test(logging): move recovery filter coverage

Co-authored-by: Codex <noreply@openai.com>

backend/Cargo.lock

@@ -2628,6 +2628,7 @@ dependencies = [
  "tokio-tungstenite 0.28.0",
  "tokio-util",
  "tracing",
+ "tracing-subscriber",
  "tungstenite 0.27.0",
  "url",
  "uuid",

backend/apps/gateway/tests/bootstrap.rs

@@ -134,7 +134,7 @@ fn config_loader_should_reject_removed_fingerprint_section() {
 }
 
 #[test]
-fn config_loader_should_reject_invalid_codex_core_version() {
+fn config_loader_should_reject_invalid_codex_cli_version() {
     assert_rejected(valid_config().replace("codex_version: '0.146.0'", "codex_version: 'latest'"));
 }
 

backend/crates/gateway-admin/src/backup/task.rs

@@ -220,7 +220,8 @@ impl BackupTask {
                     sha256: Some(artifact.sha256.clone()),
                     ..Default::default()
                 };
-                self.repository
+                let Some(uploading_record) = self
+                    .repository
                     .transition_status(
                         &record.id,
                         BackupStatus::Dumping,
@@ -229,9 +230,12 @@ impl BackupTask {
                         now,
                     )
                     .await
-                    .map_err(repo_error)?;
+                    .map_err(repo_error)?
+                else {
+                    return Ok(());
+                };
                 self.upload_and_verify(
-                    record,
+                    &uploading_record,
                     &artifact.path,
                     artifact.size_bytes,
                     &artifact.sha256,
@@ -394,7 +398,8 @@ impl BackupTask {
             sha256: Some(artifact.sha256.clone()),
             ..Default::default()
         };
-        self.repository
+        let Some(uploading_record) = self
+            .repository
             .transition_status(
                 &record.id,
                 BackupStatus::Dumping,
@@ -403,10 +408,13 @@ impl BackupTask {
                 now,
             )
             .await
-            .map_err(repo_error)?;
+            .map_err(repo_error)?
+        else {
+            return Ok(());
+        };
 
         self.upload_and_verify(
-            record,
+            &uploading_record,
             &artifact.path,
             artifact.size_bytes,
             &artifact.sha256,
@@ -448,8 +456,12 @@ impl BackupTask {
             .await
         {
             let _ = store.delete_object(&config, &record.object_key).await;
-            self.fail_task(record, error.code(), "上传到对象存储失败")
-                .await?;
+            self.fail_task(
+                record,
+                error.code(),
+                format!("上传到对象存储失败：{}", error.message()),
+            )
+            .await?;
             return Ok(());
         }
 
@@ -475,19 +487,46 @@ impl BackupTask {
                     );
                 }
             }
-            Ok(_) => {
+            Ok(Some(remote)) => {
+                let size_matches = remote.size_bytes == size_bytes;
+                let sha256_matches = remote.sha256 == sha256;
+                warn!(
+                    backup_id = %record.id,
+                    expected_size_bytes = size_bytes,
+                    remote_size_bytes = remote.size_bytes,
+                    sha256_matches,
+                    "备份远端对象校验不一致"
+                );
                 let _ = store.delete_object(&config, &record.object_key).await;
                 self.fail_task(
                     record,
                     code::REMOTE_VERIFICATION_FAILED,
-                    "远端对象大小或校验值不一致",
+                    match (size_matches, sha256_matches) {
+                        (false, true) => "远端对象大小不一致",
+                        (true, false) => "远端对象 sha256 校验值不一致",
+                        (false, false) => "远端对象大小和 sha256 校验值均不一致",
+                        (true, true) => "远端对象校验状态异常",
+                    },
+                )
+                .await?;
+            }
+            Ok(None) => {
+                warn!(backup_id = %record.id, "上传完成后未找到远端对象");
+                self.fail_task(
+                    record,
+                    code::REMOTE_VERIFICATION_FAILED,
+                    "上传完成后未找到远端对象",
                 )
                 .await?;
             }
             Err(error) => {
                 let _ = store.delete_object(&config, &record.object_key).await;
-                self.fail_task(record, error.code(), "远端对象校验失败")
-                    .await?;
+                self.fail_task(
+                    record,
+                    error.code(),
+                    format!("远端对象校验失败：{}", error.message()),
+                )
+                .await?;
             }
         }
         Ok(())
@@ -500,10 +539,17 @@ impl BackupTask {
         error_code: &'static str,
         message: impl Into<String>,
     ) -> Result<(), WorkerTaskError> {
+        let message = message.into();
+        warn!(
+            backup_id = %record.id,
+            error_code,
+            error_message = %message,
+            "备份任务失败"
+        );
         let now = Utc::now();
         let update = StatusTransitionUpdate {
             error_code: Some(error_code.to_string()),
-            error_message: Some(message.into()),
+            error_message: Some(message),
             completed_at: Some(now),
             ..Default::default()
         };

backend/crates/gateway-admin/src/model/accounts.rs

@@ -1,6 +1,6 @@
 //! 多 Provider 账号目录与连接测试的公共事实。
 
-use std::pin::Pin;
+use std::{pin::Pin, time::SystemTime};
 
 use chrono::{DateTime, Utc};
 use futures::Stream;
@@ -13,15 +13,146 @@ use super::{PageSize, Revision, observability::TimeRange};
 pub use gateway_core::engine::credential::AccountAvailability;
 
 /// 管理页使用的归一化账号状态。
+///
+/// 只回答两个问题：是否可调度、是否需要人工干预。具体错误原因不进入此枚举，
+/// 由 [`AccountErrorReason`] 承载；新增原因只扩展 reason，不改变分类。
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub enum AccountStatus {
-    Active,
-    RateLimited,
+    /// 正常，可调度。
+    Normal,
+    /// 配额耗尽（持久化终态），等待额度恢复，无需干预。
     QuotaExhausted,
-    Expired,
-    Invalid,
+    /// 限流中（quota 快照级信号），自动恢复，无需干预。
+    RateLimited,
+    /// 管理员手动停用（`enabled = false`）。
     Disabled,
+    /// 错误（过期 / 失效 / 封禁等），需要人工干预。具体原因见 [`AccountErrorReason`]。
+    Error,
+}
+
+impl AccountStatus {
+    /// 返回稳定 wire 值。
+    #[must_use]
+    pub const fn as_str(self) -> &'static str {
+        match self {
+            Self::Normal => "normal",
+            Self::QuotaExhausted => "quota_exhausted",
+            Self::RateLimited => "rate_limited",
+            Self::Disabled => "disabled",
+            Self::Error => "error",
+        }
+    }
+
+    /// 解析稳定 wire 值。
+    #[must_use]
+    pub fn parse(value: &str) -> Option<Self> {
+        match value {
+            "normal" => Some(Self::Normal),
+            "quota_exhausted" => Some(Self::QuotaExhausted),
+            "rate_limited" => Some(Self::RateLimited),
+            "disabled" => Some(Self::Disabled),
+            "error" => Some(Self::Error),
+            _ => None,
+        }
+    }
+}
+
+/// 状态派生所需的最小事实；由 store / use_case 各自构造，派生规则唯一。
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct AccountStatusSignals {
+    pub enabled: bool,
+    pub availability: AccountAvailability,
+    pub access_token_expired: bool,
+    pub quota_limit_reached: bool,
+    /// 429 临时限流（Redis 冷却）到期时间；`Some` 表示账号当前处于临时限流。
+    pub rate_limited_until: Option<SystemTime>,
+}
+
+impl AccountStatusSignals {
+    /// 派生产出运营分类。终态优先于瞬时信号：停用 > 配额耗尽 > 错误 > 限流。
+    #[must_use]
+    pub const fn derive(self) -> AccountStatus {
+        if !self.enabled {
+            AccountStatus::Disabled
+        } else if matches!(self.availability, AccountAvailability::QuotaExhausted) {
+            AccountStatus::QuotaExhausted
+        } else if self.access_token_expired
+            || matches!(
+                self.availability,
+                AccountAvailability::Expired
+                    | AccountAvailability::Invalid
+                    | AccountAvailability::Banned
+                    | AccountAvailability::Unknown
+            )
+        {
+            AccountStatus::Error
+        } else if self.rate_limited_until.is_some() {
+            AccountStatus::RateLimited
+        } else if self.quota_limit_reached {
+            AccountStatus::QuotaExhausted
+        } else {
+            AccountStatus::Normal
+        }
+    }
+}
+
+/// 错误分类下的具体原因；仅在 `AccountStatus::Error` 时存在。
+///
+/// 受控枚举用于展示与聚合，扩展不改变状态分类。
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum AccountErrorReason {
+    Expired,
+    TokenExpired,
+    Invalid,
     Banned,
+    Unknown,
+}
+
+impl AccountErrorReason {
+    /// 返回稳定 wire 值。
+    #[must_use]
+    pub const fn as_str(self) -> &'static str {
+        match self {
+            Self::Expired => "expired",
+            Self::TokenExpired => "token_expired",
+            Self::Invalid => "invalid",
+            Self::Banned => "banned",
+            Self::Unknown => "unknown",
+        }
+    }
+
+    /// 解析稳定 wire 值。
+    #[must_use]
+    pub fn parse(value: &str) -> Option<Self> {
+        match value {
+            "expired" => Some(Self::Expired),
+            "token_expired" => Some(Self::TokenExpired),
+            "invalid" => Some(Self::Invalid),
+            "banned" => Some(Self::Banned),
+            "unknown" => Some(Self::Unknown),
+            _ => None,
+        }
+    }
+}
+
+/// 由信号派生错误原因；非错误分类（正常 / 配额耗尽 / 限流 / 停用）返回 `None`。
+///
+/// 与 [`AccountStatusSignals::derive`] 保持同一判定：`Error` 当且仅当此处返回 `Some`。
+#[must_use]
+pub fn derive_error_reason(signals: &AccountStatusSignals) -> Option<AccountErrorReason> {
+    if !signals.enabled {
+        return None;
+    }
+    match signals.availability {
+        AccountAvailability::QuotaExhausted => None,
+        AccountAvailability::Ready => signals
+            .access_token_expired
+            .then_some(AccountErrorReason::TokenExpired),
+        AccountAvailability::Expired => Some(AccountErrorReason::Expired),
+        AccountAvailability::Invalid => Some(AccountErrorReason::Invalid),
+        AccountAvailability::Banned => Some(AccountErrorReason::Banned),
+        AccountAvailability::Unknown => Some(AccountErrorReason::Unknown),
+    }
 }
 
 /// 账号列表排序字段。
@@ -78,6 +209,8 @@ pub struct AccountRecord {
     pub enabled: bool,
     pub availability: AccountAvailability,
     pub availability_observed_at: DateTime<Utc>,
+    /// 最近一次非 ready 状态变更的原始原因（上游错误原文 / 原因码）；恢复 ready 后清空。
+    pub last_error_message: Option<String>,
     pub quota_observed_at: Option<DateTime<Utc>>,
     pub created_at: DateTime<Utc>,
     pub updated_at: DateTime<Utc>,
@@ -170,13 +303,15 @@ pub struct AccountPage {
 
 /// 统一账号目录的全局状态计数，不受当前筛选和分页影响。
 ///
-/// 配额耗尽是可恢复状态，始终单列；`unavailable` 只包含过期、禁用和封禁账号。
+/// 计数与 [`AccountStatus`] 一一对应，由 store 按派生状态聚合。
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub struct AccountSummary {
     pub total: u64,
-    pub active: u64,
+    pub normal: u64,
     pub quota_exhausted: u64,
-    pub unavailable: u64,
+    pub rate_limited: u64,
+    pub disabled: u64,
+    pub error: u64,
 }
 
 /// 账号启停写入命令。

backend/crates/gateway-admin/src/model/provider_credentials.rs

@@ -11,7 +11,10 @@ use gateway_core::{
 
 use super::{
     AdminError, MutationActor, MutationContext, PageSize, Revision,
-    accounts::{AccountAvailability, AccountRecord, AccountStatus, AccountSummary, AccountUsage},
+    accounts::{
+        AccountAvailability, AccountErrorReason, AccountRecord, AccountStatus, AccountSummary,
+        AccountUsage,
+    },
 };
 
 /// Provider-owned JSON；公共层只搬运且 Debug 不输出值。
@@ -592,6 +595,9 @@ pub struct ProviderQuota {
     pub windows: Vec<ProviderQuotaWindow>,
     /// 快照级限流事实（顶层或任一窗口触顶）；Admin 状态派生用，前端不再遍历窗口猜测。
     pub limit_reached: bool,
+    /// 429 临时限流（Redis 冷却）到期时间；`Some` 时账号处于临时限流，到期自动恢复。
+    /// 与 `limit_reached`（配额耗尽）是独立维度，不污染额度窗口。
+    pub rate_limited_until: Option<DateTime<Utc>>,
     pub provider_data: Option<ProviderDocument>,
 }
 
@@ -688,6 +694,8 @@ impl fmt::Debug for ProviderExport {
 pub struct AccountDirectoryItem {
     pub account: AccountRecord,
     pub status: AccountStatus,
+    /// `status == Error` 时的具体原因；其余状态为 `None`。
+    pub error_reason: Option<AccountErrorReason>,
     pub usage: Option<super::accounts::AccountUsage>,
     pub quota: ProviderQuota,
 }

backend/crates/gateway-admin/src/use_case/accounts.rs

@@ -1,6 +1,6 @@
 //! 统一账号目录与跨 Provider 动态分派。
 
-use std::{collections::BTreeMap, sync::Arc};
+use std::{collections::BTreeMap, sync::Arc, time::SystemTime};
 
 use async_trait::async_trait;
 use chrono::{Duration, Utc};
@@ -18,7 +18,8 @@ use crate::{
         AdminError, MutationContext,
         accounts::{
             AccountConnectionTestEvent, AccountConnectionTestEventStream, AccountListQuery,
-            AccountRecord, AccountStatus, AccountUsageWindowQuery,
+            AccountRecord, AccountStatus, AccountStatusSignals, AccountUsageWindowQuery,
+            derive_error_reason,
         },
         observability::TimeRange,
         provider_credentials::{
@@ -216,8 +217,11 @@ impl DefaultAccountsService {
             std::slice::from_mut(&mut quota),
         )
         .await?;
+        let signals = account_signals(&account, &quota, now);
+        let error_reason = derive_error_reason(&signals);
         Ok(AccountDirectoryItem {
-            status: account_status(&account, quota.limit_reached, now),
+            status: signals.derive(),
+            error_reason,
             usage: usage.into_iter().next(),
             account,
             quota,
@@ -301,11 +305,16 @@ impl AccountsService for DefaultAccountsService {
             .items
             .into_iter()
             .zip(quotas)
-            .map(|(account, quota)| AccountDirectoryItem {
-                status: account_status(&account, quota.limit_reached, now),
-                usage: usage.get(&account.id).cloned(),
-                account,
-                quota,
+            .map(|(account, quota)| {
+                let signals = account_signals(&account, &quota, now);
+                let error_reason = derive_error_reason(&signals);
+                AccountDirectoryItem {
+                    status: signals.derive(),
+                    error_reason,
+                    usage: usage.get(&account.id).cloned(),
+                    account,
+                    quota,
+                }
             })
             .collect();
         Ok(AccountDirectoryPage {
@@ -512,43 +521,39 @@ fn empty_quota() -> ProviderQuota {
         refresh_token_expires_at: None,
         windows: Vec::new(),
         limit_reached: false,
+        rate_limited_until: None,
         provider_data: None,
     }
 }
 
-/// 互斥页面状态：availability 终态优先，quota 快照级 `limit_reached` 派生
-/// `rate_limited`，`Unknown` 保守映射 `invalid`（不可调度，不显示为 active）。
+/// 构造状态派生所需的信号快照。
+fn account_signals(
+    account: &AccountRecord,
+    quota: &ProviderQuota,
+    now: chrono::DateTime<Utc>,
+) -> AccountStatusSignals {
+    AccountStatusSignals {
+        enabled: account.enabled,
+        availability: account.availability,
+        access_token_expired: account
+            .access_token_expires_at
+            .is_some_and(|expires_at| expires_at <= now),
+        quota_limit_reached: quota.limit_reached,
+        rate_limited_until: quota.rate_limited_until.map(SystemTime::from),
+    }
+}
+
+/// 账号运营分类（互斥展示状态）。
 fn account_status(
     account: &AccountRecord,
     quota_limit_reached: bool,
     now: chrono::DateTime<Utc>,
 ) -> AccountStatus {
-    if !account.enabled {
-        AccountStatus::Disabled
-    } else if account.availability == crate::model::accounts::AccountAvailability::Banned {
-        AccountStatus::Banned
-    } else if account.availability == crate::model::accounts::AccountAvailability::QuotaExhausted {
-        AccountStatus::QuotaExhausted
-    } else if matches!(
-        account.availability,
-        crate::model::accounts::AccountAvailability::Invalid
-    ) {
-        AccountStatus::Invalid
-    } else if matches!(
-        account.availability,
-        crate::model::accounts::AccountAvailability::Expired
-    ) || account
-        .access_token_expires_at
-        .is_some_and(|expires_at| expires_at <= now)
-    {
-        AccountStatus::Expired
-    } else if quota_limit_reached {
-        AccountStatus::RateLimited
-    } else if account.availability == crate::model::accounts::AccountAvailability::Unknown {
-        AccountStatus::Invalid
-    } else {
-        AccountStatus::Active
-    }
+    let quota = ProviderQuota {
+        limit_reached: quota_limit_reached,
+        ..empty_quota()
+    };
+    account_signals(account, &quota, now).derive()
 }
 
 fn retained_usage_range(now: chrono::DateTime<Utc>, retention_days: u32) -> TimeRange {

backend/crates/gateway-admin/tests/backup/support.rs

@@ -13,9 +13,10 @@ use gateway_admin::{
         MutationActor, MutationContext, Revision,
         auth::AdminAuditEvent,
         backup::{
-            BackupObjectMetadata, BackupRecord, BackupRecordListQuery, BackupRecordPage,
-            BackupRecordSeed, BackupSettings, BackupStatus, BackupStorageConfig, BackupTriggerKind,
-            ConnectionTestResult, UpdateBackupScheduleCommand, UpdateBackupStorageCommand,
+            BackupError, BackupObjectMetadata, BackupRecord, BackupRecordListQuery,
+            BackupRecordPage, BackupRecordSeed, BackupSettings, BackupStatus, BackupStorageConfig,
+            BackupTriggerKind, ConnectionTestResult, UpdateBackupScheduleCommand,
+            UpdateBackupStorageCommand,
         },
     },
     ports::{
@@ -497,15 +498,23 @@ impl DatabaseDumpPort for FakeDumpPort {
 /// 内存版对象存储：把上传内容按 key 保存并支持 Head/Delete/预签名。
 pub(crate) struct FakeObjectStore {
     pub(crate) objects: Mutex<std::collections::HashMap<String, Vec<u8>>>,
+    upload_error: Option<BackupError>,
 }
 
 impl FakeObjectStore {
     pub(crate) fn new() -> Self {
         Self {
             objects: Mutex::new(std::collections::HashMap::new()),
+            upload_error: None,
         }
     }
 
+    /// 让上传返回指定的脱敏 S3 错误。
+    pub(crate) fn fail_upload(mut self, code: &'static str, message: impl Into<String>) -> Self {
+        self.upload_error = Some(BackupError::new(code, message.into()));
+        self
+    }
+
     pub(crate) fn object(&self, key: &str) -> Option<Vec<u8>> {
         self.objects.lock().expect("objects").get(key).cloned()
     }
@@ -530,6 +539,9 @@ impl BackupObjectStorePort for FakeObjectStore {
         _config: &BackupStorageConfig,
         request: UploadObjectRequest,
     ) -> Result<(), gateway_admin::model::backup::BackupError> {
+        if let Some(error) = &self.upload_error {
+            return Err(error.clone());
+        }
         let content = std::fs::read(&request.source).map_err(|_| {
             gateway_admin::model::backup::BackupError::new(
                 gateway_admin::model::backup::code::S3_UPLOAD_FAILED,

backend/crates/gateway-admin/tests/backup/task.rs

@@ -6,9 +6,9 @@ use chrono::{Duration, Utc};
 use gateway_admin::{
     backup::task::BackupTask,
     model::backup::{
-        BackupRecordSeed, BackupStatus, BackupTriggerKind, UpdateBackupScheduleCommand,
+        BackupRecordSeed, BackupStatus, BackupTriggerKind, UpdateBackupScheduleCommand, code,
     },
-    ports::backup::BackupRepository,
+    ports::backup::{BackupRepository, DatabaseDumpPort, DumpRequest},
 };
 use gateway_core::engine::CancellationToken;
 
@@ -89,6 +89,78 @@ async fn backup_task_marks_failed_when_dump_port_fails() {
     assert!(object_store.object(&seed.object_key).is_none());
 }
 
+#[tokio::test]
+async fn backup_task_marks_failed_with_the_upload_error() {
+    let repository = Arc::new(FakeBackupRepository::new(configured_settings()));
+    let dump = Arc::new(FakeDumpPort::new());
+    let object_store =
+        Arc::new(FakeObjectStore::new().fail_upload(code::S3_UPLOAD_FAILED, "对象存储拒绝上传"));
+    let seed = BackupRecordSeed {
+        id: backup_id("upload-failure"),
+        trigger_kind: BackupTriggerKind::Manual,
+        scheduled_at: None,
+        object_key: "codex/2026/08/01/upload-failure.dump".to_owned(),
+        expires_at: None,
+    };
+    repository
+        .insert_backup_record(seed.clone())
+        .await
+        .expect("insert queued");
+
+    let task = BackupTask::new(repository.clone(), dump, object_store);
+    task.run_cycle(&CancellationToken::new())
+        .await
+        .expect("run cycle");
+
+    let records = repository.all_records();
+    let failed = &records[0];
+    assert_eq!(failed.status, BackupStatus::Failed);
+    assert_eq!(failed.error_code.as_deref(), Some(code::S3_UPLOAD_FAILED));
+    assert_eq!(
+        failed.error_message.as_deref(),
+        Some("上传到对象存储失败：对象存储拒绝上传")
+    );
+}
+
+#[tokio::test]
+async fn backup_task_recovery_marks_failed_with_the_upload_error() {
+    let repository = Arc::new(FakeBackupRepository::new(configured_settings()));
+    let dump = Arc::new(FakeDumpPort::new());
+    let object_store =
+        Arc::new(FakeObjectStore::new().fail_upload(code::S3_UPLOAD_FAILED, "对象存储拒绝上传"));
+    let seed = BackupRecordSeed {
+        id: backup_id("recovered-upload-failure"),
+        trigger_kind: BackupTriggerKind::Manual,
+        scheduled_at: None,
+        object_key: "codex/2026/08/01/recovered-upload-failure.dump".to_owned(),
+        expires_at: None,
+    };
+    repository
+        .insert_backup_record(seed.clone())
+        .await
+        .expect("insert queued");
+    let cancellation = CancellationToken::new();
+    dump.dump(DumpRequest {
+        backup_id: seed.id.clone(),
+        cancellation: cancellation.clone(),
+    })
+    .await
+    .expect("stage dump");
+    repository.force_status(&seed.id, BackupStatus::Dumping);
+
+    let task = BackupTask::new(repository.clone(), dump, object_store);
+    task.run_cycle(&cancellation).await.expect("run cycle");
+
+    let records = repository.all_records();
+    let failed = &records[0];
+    assert_eq!(failed.status, BackupStatus::Failed);
+    assert_eq!(failed.error_code.as_deref(), Some(code::S3_UPLOAD_FAILED));
+    assert_eq!(
+        failed.error_message.as_deref(),
+        Some("上传到对象存储失败：对象存储拒绝上传")
+    );
+}
+
 #[tokio::test]
 async fn retention_cleans_expired_scheduled_backups() {
     let repository = Arc::new(FakeBackupRepository::new(configured_settings()));

backend/crates/gateway-admin/tests/main.rs

@@ -128,6 +128,7 @@ fn representative_quota_should_prefer_short_window_and_highest_usage() {
             quota_window("shortTerm", Some(14_400), Some(45.0)),
         ],
         limit_reached: false,
+        rate_limited_until: None,
         provider_data: None,
     };
 

backend/crates/gateway-admin/tests/use_case/accounts.rs

@@ -451,9 +451,11 @@ impl AccountStore for FakeAccountStore {
             total: accounts.len() as u64,
             summary: AccountSummary {
                 total: accounts.len() as u64,
-                active: 1,
+                normal: 1,
                 quota_exhausted: 0,
-                unavailable: 0,
+                rate_limited: 0,
+                disabled: 0,
+                error: 0,
             },
         })
     }
@@ -844,7 +846,7 @@ async fn connection_test_should_probe_unavailable_account() {
 }
 
 #[tokio::test]
-async fn connection_test_rate_limited_probe_preserves_active_account_status() {
+async fn connection_test_rate_limited_probe_preserves_normal_account_status() {
     let events = events();
     let provider = FakeProviderAdmin::new("xai", events.clone());
     let store = FakeAccountStore::new("xai", events);
@@ -867,7 +869,7 @@ async fn connection_test_rate_limited_probe_preserves_active_account_status() {
     assert!(matches!(
         events.last(),
         Some(AccountConnectionTestEvent::Failed {
-            account_status: gateway_admin::model::accounts::AccountStatus::Active,
+            account_status: gateway_admin::model::accounts::AccountStatus::Normal,
             ..
         })
     ));
@@ -948,7 +950,7 @@ async fn accounts_refresh_should_keep_guard_through_store_commit() {
     assert_eq!(result.account.account.provider_kind.as_str(), "openai");
     assert_eq!(
         result.account.status,
-        gateway_admin::model::accounts::AccountStatus::Active
+        gateway_admin::model::accounts::AccountStatus::Normal
     );
 
     assert_eq!(
@@ -984,12 +986,12 @@ async fn accounts_list_should_return_complete_directory_semantics() {
         .expect("complete account directory");
 
     assert_eq!(page.summary.total, 1);
-    assert_eq!(page.summary.active, 1);
+    assert_eq!(page.summary.normal, 1);
     let account = page.items.first().expect("account item");
     assert_eq!(account.account.provider_kind.as_str(), "openai");
     assert_eq!(
         account.status,
-        gateway_admin::model::accounts::AccountStatus::Active
+        gateway_admin::model::accounts::AccountStatus::Normal
     );
 }
 
@@ -1013,6 +1015,7 @@ async fn accounts_list_should_degrade_quota_failure_to_empty_window_without_drop
             provider_data: None,
         }],
         limit_reached: false,
+        rate_limited_until: None,
         provider_data: None,
     });
     let failing = FakeProviderAdmin::new("xai", events.clone());
@@ -1088,8 +1091,8 @@ async fn accounts_list_should_prefer_quota_exhaustion_over_stale_token_expiry()
 }
 
 #[tokio::test]
-async fn accounts_list_should_map_unknown_availability_to_invalid_not_active() {
-    // Unknown 不可调度，Admin 不得显示为 active。
+async fn accounts_list_should_map_unknown_availability_to_error_not_normal() {
+    // Unknown 不可调度，Admin 不得显示为 normal。
     let provider = FakeProviderAdmin::new("openai", events());
     let mut account = account_record("openai");
     account.availability = AccountAvailability::Unknown;
@@ -1111,13 +1114,14 @@ async fn accounts_list_should_map_unknown_availability_to_invalid_not_active() {
 
     assert_eq!(
         page.items.first().expect("account item").status,
-        gateway_admin::model::accounts::AccountStatus::Invalid,
+        gateway_admin::model::accounts::AccountStatus::Error,
     );
 }
 
 #[tokio::test]
 async fn accounts_list_should_map_rate_limited_from_quota_limit_reached() {
-    // ready + 快照级 quota.limitReached → rate_limited。
+    // ready + 快照级 quota.limitReached → quota_exhausted（配额耗尽）；
+    // rate_limited 只来自 ProviderQuota.rate_limited_until（429 Redis 冷却）。
     let provider = FakeProviderAdmin::new("openai", events());
     provider.set_quota(ProviderQuota {
         observed_at: Some(Utc::now()),
@@ -1135,6 +1139,7 @@ async fn accounts_list_should_map_rate_limited_from_quota_limit_reached() {
             provider_data: None,
         }],
         limit_reached: true,
+        rate_limited_until: Some(Utc::now() + TimeDelta::minutes(5)),
         provider_data: None,
     });
     let store = FakeAccountStore::new("openai", events());
@@ -1159,6 +1164,51 @@ async fn accounts_list_should_map_rate_limited_from_quota_limit_reached() {
     );
 }
 
+#[tokio::test]
+async fn accounts_list_should_map_quota_exhausted_when_only_limit_reached() {
+    // ready + 快照级 quota.limitReached、无 429 冷却 → quota_exhausted。
+    let provider = FakeProviderAdmin::new("openai", events());
+    provider.set_quota(ProviderQuota {
+        observed_at: Some(Utc::now()),
+        refresh_token_expires_at: None,
+        windows: vec![ProviderQuotaWindow {
+            key: "primary".to_owned(),
+            group: "shortTerm".to_owned(),
+            label: "5小时限额".to_owned(),
+            source: None,
+            window_seconds: Some(5 * 60 * 60),
+            used_percent: Some(100.0),
+            reset_at: Some(Utc::now() + TimeDelta::hours(1)),
+            limit_reached: true,
+            local_usage: None,
+            provider_data: None,
+        }],
+        limit_reached: true,
+        rate_limited_until: None,
+        provider_data: None,
+    });
+    let store = FakeAccountStore::new("openai", events());
+
+    let page = accounts_service(provider, store)
+        .await
+        .accounts()
+        .list(AccountListQuery {
+            page: 1,
+            page_size: gateway_admin::model::PageSize::new(20).expect("page size"),
+            provider_kind: None,
+            search: None,
+            status: None,
+            sort: None,
+        })
+        .await
+        .expect("quota exhausted account directory");
+
+    assert_eq!(
+        page.items.first().expect("account item").status,
+        gateway_admin::model::accounts::AccountStatus::QuotaExhausted,
+    );
+}
+
 #[tokio::test]
 async fn accounts_list_should_attach_local_usage_to_quota_windows() {
     let provider = FakeProviderAdmin::new("openai", events());
@@ -1179,6 +1229,7 @@ async fn accounts_list_should_attach_local_usage_to_quota_windows() {
             provider_data: None,
         }],
         limit_reached: false,
+        rate_limited_until: None,
         provider_data: None,
     });
     let store = FakeAccountStore::new("openai", events());
@@ -1287,6 +1338,7 @@ fn empty_quota() -> ProviderQuota {
         refresh_token_expires_at: None,
         windows: Vec::new(),
         limit_reached: false,
+        rate_limited_until: None,
         provider_data: None,
     }
 }
@@ -1332,6 +1384,7 @@ pub(super) fn account_record(kind: &str) -> AccountRecord {
         enabled: true,
         availability: AccountAvailability::Ready,
         availability_observed_at: now,
+        last_error_message: None,
         quota_observed_at: None,
         created_at: now,
         updated_at: now,

backend/crates/gateway-api/src/admin/accounts.rs

@@ -91,7 +91,8 @@ impl ListQuery {
         let status = match self.status.as_deref().map(str::trim) {
             None | Some("") => None,
             Some(value) => Some(
-                parse_account_status(value).ok_or_else(|| WireValidationError::new("status"))?,
+                DomainAccountStatus::parse(&value.to_ascii_lowercase())
+                    .ok_or_else(|| WireValidationError::new("status"))?,
             ),
         };
         let sort = match (self.sort_by.as_deref(), self.sort_direction.as_deref()) {
@@ -127,17 +128,6 @@ fn parse_provider(value: &str) -> Result<Option<ProviderKind>, WireValidationErr
         .map_err(|_| WireValidationError::new("provider"))
 }
 
-fn parse_account_status(value: &str) -> Option<DomainAccountStatus> {
-    match value.trim().to_ascii_lowercase().as_str() {
-        "active" => Some(DomainAccountStatus::Active),
-        "expired" => Some(DomainAccountStatus::Expired),
-        "quota_exhausted" => Some(DomainAccountStatus::QuotaExhausted),
-        "disabled" => Some(DomainAccountStatus::Disabled),
-        "banned" => Some(DomainAccountStatus::Banned),
-        _ => None,
-    }
-}
-
 fn parse_sort_field(value: &str) -> Option<AccountSortField> {
     match value.trim() {
         "email" => Some(AccountSortField::Email),
@@ -172,9 +162,11 @@ pub struct AccountPageData {
 #[serde(rename_all = "camelCase")]
 pub struct AccountSummaryView {
     pub total: u64,
-    pub active: u64,
+    pub normal: u64,
     pub quota_exhausted: u64,
-    pub unavailable: u64,
+    pub rate_limited: u64,
+    pub disabled: u64,
+    pub error: u64,
 }
 
 /// 一条安全账号视图。
@@ -194,6 +186,10 @@ pub struct AccountView {
     pub has_refresh_token: bool,
     pub status: String,
     pub display_status: String,
+    /// `status == "error"` 时的具体原因；其余状态为 `null`。
+    pub error_reason: Option<String>,
+    /// 最近一次失败的上游错误描述；仅错误状态存在。
+    pub error_message: Option<String>,
     pub availability: String,
     pub enabled: bool,
     pub access_token_expires_at: Option<String>,
@@ -214,6 +210,8 @@ pub struct AccountView {
 pub struct AccountQuotaView {
     pub refreshed_at_display: String,
     pub limit_reached: bool,
+    /// 429 临时限流（Redis 冷却）到期时间展示；非限流中为 `null`。
+    pub rate_limited_until: Option<String>,
     pub windows: Vec<AccountQuotaWindowView>,
 }
 
@@ -527,7 +525,7 @@ impl From<DomainConnectionTestEvent> for AccountConnectionTestEvent {
             DomainConnectionTestEvent::Completed { account_status } => serde_json::json!({
                 "type": "test_complete",
                 "success": true,
-                "accountStatus": account_status_name(account_status)
+                "accountStatus": account_status.as_str()
             }),
             DomainConnectionTestEvent::Failed {
                 message,
@@ -545,7 +543,7 @@ impl From<DomainConnectionTestEvent> for AccountConnectionTestEvent {
                 "upstreamStatus": upstream_status,
                 "upstreamContentType": upstream_content_type,
                 "upstreamBody": upstream_body,
-                "accountStatus": account_status_name(account_status)
+                "accountStatus": account_status.as_str()
             }),
         };
         Self { data }
@@ -1333,9 +1331,11 @@ fn account_page_data(
         page: PageMeta::new(page, u32::from(page_size), result.total, total_pages),
         summary: AccountSummaryView {
             total: result.summary.total,
-            active: result.summary.active,
+            normal: result.summary.normal,
             quota_exhausted: result.summary.quota_exhausted,
-            unavailable: result.summary.unavailable,
+            rate_limited: result.summary.rate_limited,
+            disabled: result.summary.disabled,
+            error: result.summary.error,
         },
     }
 }
@@ -1366,10 +1366,11 @@ fn account_view(item: AccountDirectoryItem, now: DateTime<Utc>) -> AccountView {
     let AccountDirectoryItem {
         account,
         status,
+        error_reason,
         usage,
         quota,
     } = item;
-    let status = account_status_name(status).to_owned();
+    let status = status.as_str().to_owned();
     let expires_at = account.access_token_expires_at.as_ref().map(china_rfc3339);
     let added_at = china_rfc3339(&account.created_at);
     let updated_at = china_rfc3339(&account.updated_at);
@@ -1388,6 +1389,8 @@ fn account_view(item: AccountDirectoryItem, now: DateTime<Utc>) -> AccountView {
         has_refresh_token: account.has_refresh_token,
         status: status.clone(),
         display_status: status,
+        error_reason: error_reason.map(|reason| reason.as_str().to_owned()),
+        error_message: account.last_error_message,
         availability: account.availability.as_str().to_owned(),
         enabled: account.enabled,
         access_token_expires_at: expires_at,
@@ -1417,10 +1420,12 @@ fn account_quota_view(
         .observed_at
         .map_or_else(|| "—".to_owned(), |value| relative_time(value, now));
     let windows = quota.windows.into_iter().map(quota_window_view).collect();
+    let rate_limited_until = quota.rate_limited_until.map(|until| china_datetime(&until));
     (
         AccountQuotaView {
             refreshed_at_display,
             limit_reached: quota.limit_reached,
+            rate_limited_until,
             windows,
         },
         refresh_token_expires_at,
@@ -1713,18 +1718,6 @@ fn provider_document_value(document: ProviderDocument) -> Value {
     Value::Object(document.into_provider_data().into_inner())
 }
 
-fn account_status_name(status: DomainAccountStatus) -> &'static str {
-    match status {
-        DomainAccountStatus::Active => "active",
-        DomainAccountStatus::RateLimited => "rate_limited",
-        DomainAccountStatus::QuotaExhausted => "quota_exhausted",
-        DomainAccountStatus::Expired => "expired",
-        DomainAccountStatus::Invalid => "invalid",
-        DomainAccountStatus::Disabled => "disabled",
-        DomainAccountStatus::Banned => "banned",
-    }
-}
-
 fn relative_time(value: DateTime<Utc>, now: DateTime<Utc>) -> String {
     let elapsed = now.signed_duration_since(value);
     if elapsed.num_seconds() < 0 {

backend/crates/gateway-api/src/admin/observability/query.rs

@@ -310,16 +310,6 @@ pub(crate) fn usage_range(
     domain::TimeRange::new(start, end).map_err(|_| WireValidationError::new("timeRange"))
 }
 
-pub(crate) fn dashboard_range(
-    start: Option<&str>,
-    end: Option<&str>,
-) -> Result<domain::TimeRange, WireValidationError> {
-    let end = parse_datetime(end)?.unwrap_or_else(Utc::now);
-    let start =
-        parse_datetime(start)?.unwrap_or_else(|| domain::china_day_start(end) - Duration::days(1));
-    domain::TimeRange::new(start, end).map_err(|_| WireValidationError::new("timeRange"))
-}
-
 pub(crate) fn dashboard_today_range(
     start: Option<&str>,
     end: Option<&str>,

backend/crates/gateway-api/src/admin/observability/routes.rs

@@ -38,7 +38,8 @@ where
     S: AdminSessionState + Send + Sync,
 {
     let kind = query.trend_kind().map_err(map_wire_error)?;
-    let range = dashboard_range(query.start_time.as_deref(), query.end_time.as_deref())
+    // 概览默认按中国时区当日统计，与单独趋势接口保持同一口径。
+    let range = dashboard_today_range(query.start_time.as_deref(), query.end_time.as_deref())
         .map_err(map_wire_error)?;
     let result = state
         .admin_services()

backend/crates/gateway-api/tests/admin/accounts.rs

@@ -10,7 +10,7 @@ mod query {
             "pageSize": 20,
             "provider": "xai",
             "search": "  operator  ",
-            "status": "active",
+            "status": "normal",
             "sortBy": "lastUsedAt",
             "sortDirection": "desc"
         }))
@@ -23,7 +23,7 @@ mod query {
             Some(ref provider) if provider.as_str() == "xai"
         ));
         assert_eq!(query.search.as_deref(), Some("operator"));
-        assert_eq!(query.status, Some(AccountStatus::Active));
+        assert_eq!(query.status, Some(AccountStatus::Normal));
         assert_eq!(
             query.sort.expect("sort").field,
             AccountSortField::LastUsedAt
@@ -34,6 +34,19 @@ mod query {
         );
     }
 
+    #[test]
+    fn account_query_should_parse_rate_limited_status() {
+        let query: ListQuery = serde_json::from_value(json!({
+            "status": "rate_limited"
+        }))
+        .expect("deserialize account query");
+
+        assert_eq!(
+            query.validate().expect("validate account query").status,
+            Some(AccountStatus::RateLimited)
+        );
+    }
+
     #[test]
     fn account_query_should_reject_unbounded_page_size() {
         let query: ListQuery =
@@ -331,7 +344,7 @@ mod actions {
                 text: "OK".to_owned(),
             },
             DomainConnectionTestEvent::Completed {
-                account_status: AccountStatus::Active,
+                account_status: AccountStatus::Normal,
             },
             DomainConnectionTestEvent::Failed {
                 message: "upstream unavailable".to_owned(),
@@ -365,7 +378,7 @@ mod actions {
                     }
                 }),
                 json!({ "type": "content", "text": "OK" }),
-                json!({ "type": "test_complete", "success": true, "accountStatus": "active" }),
+                json!({ "type": "test_complete", "success": true, "accountStatus": "normal" }),
                 json!({
                     "type": "error",
                     "error": "upstream unavailable",

backend/crates/gateway-api/tests/admin/mod.rs

@@ -83,6 +83,7 @@ pub(super) struct AdminTestFixture {
     pub diagnostics: Arc<Mutex<Vec<DiagnosticObservation>>>,
     pub ops_errors: Arc<Mutex<Vec<OpsError>>>,
     pub dashboard_observation: Arc<Mutex<Option<DashboardObservation>>>,
+    pub dashboard_summary_range: Arc<Mutex<Option<TimeRange>>>,
 }
 
 impl AdminTestFixture {
@@ -100,12 +101,14 @@ impl AdminTestFixture {
         let diagnostics = Arc::new(Mutex::new(Vec::new()));
         let ops_errors = Arc::new(Mutex::new(Vec::new()));
         let dashboard_observation = Arc::new(Mutex::new(None));
+        let dashboard_summary_range = Arc::new(Mutex::new(None));
         let unused = Arc::new(UnusedStore {
             usage_records: Arc::clone(&usage_records),
             usage_detail: Arc::clone(&usage_detail),
             diagnostics: Arc::clone(&diagnostics),
             ops_errors: Arc::clone(&ops_errors),
             dashboard_observation: Arc::clone(&dashboard_observation),
+            dashboard_summary_range: Arc::clone(&dashboard_summary_range),
         });
         let stores = AdminStorePorts::new(
             unused.clone(),
@@ -142,6 +145,7 @@ impl AdminTestFixture {
             diagnostics,
             ops_errors,
             dashboard_observation,
+            dashboard_summary_range,
         }
     }
 
@@ -413,6 +417,7 @@ struct UnusedStore {
     diagnostics: Arc<Mutex<Vec<DiagnosticObservation>>>,
     ops_errors: Arc<Mutex<Vec<OpsError>>>,
     dashboard_observation: Arc<Mutex<Option<DashboardObservation>>>,
+    dashboard_summary_range: Arc<Mutex<Option<TimeRange>>>,
 }
 
 #[async_trait]
@@ -523,7 +528,11 @@ impl AccountStore for UnusedStore {
 
 #[async_trait]
 impl ObservabilityStore for UnusedStore {
-    async fn dashboard_summary(&self, _: TimeRange) -> AdminStoreResult<DashboardObservation> {
+    async fn dashboard_summary(&self, range: TimeRange) -> AdminStoreResult<DashboardObservation> {
+        *self
+            .dashboard_summary_range
+            .lock()
+            .expect("dashboard summary range") = Some(range);
         self.dashboard_observation
             .lock()
             .expect("dashboard observation")

backend/crates/gateway-api/tests/admin/observability/response.rs

@@ -270,6 +270,61 @@ async fn dashboard_summary_should_include_quota_exhaustion_in_unavailable_headli
     );
 }
 
+#[tokio::test]
+async fn dashboard_summary_should_default_to_current_china_day() {
+    use axum::{
+        body::Body,
+        http::{Request, StatusCode, header},
+    };
+    use chrono::{Duration, Utc};
+    use gateway_admin::model::observability::{
+        AccountPoolMetrics, AttemptMetrics, DashboardObservation, RequestMetrics, TimeRange,
+        china_day_start,
+    };
+    use gateway_api::admin::observability;
+    use tower::ServiceExt as _;
+
+    use crate::admin::{AdminTestFixture, AdminTestState};
+
+    let fixture = AdminTestFixture::new().await;
+    fixture.auth.insert_session("valid-session");
+    let now = Utc::now();
+    *fixture
+        .dashboard_observation
+        .lock()
+        .expect("dashboard observation") = Some(DashboardObservation {
+        range: TimeRange::new(now - Duration::hours(1), now).expect("dashboard range"),
+        requests: RequestMetrics::default(),
+        attempts: AttemptMetrics::default(),
+        provider_accounts: AccountPoolMetrics::default(),
+        trend: Vec::new(),
+        account_usage: Vec::new(),
+        recent_requests: Vec::new(),
+    });
+
+    let response = observability::router::<AdminTestState>()
+        .with_state(fixture.state())
+        .oneshot(
+            Request::builder()
+                .uri("/api/admin/dashboard/summary")
+                .header(header::COOKIE, "cpr_admin_session=valid-session")
+                .header("x-request-id", "req_dashboard_today_range")
+                .body(Body::empty())
+                .expect("dashboard request"),
+        )
+        .await
+        .expect("dashboard response");
+    let range = fixture
+        .dashboard_summary_range
+        .lock()
+        .expect("dashboard summary range")
+        .take()
+        .expect("recorded dashboard summary range");
+
+    assert_eq!(response.status(), StatusCode::OK);
+    assert_eq!(range.start, china_day_start(range.end));
+}
+
 #[tokio::test]
 async fn usage_detail_should_keep_attempt_snapshot_contract() {
     use axum::{

backend/crates/gateway-core/src/engine/coordinator.rs

@@ -660,6 +660,9 @@ where
             ProviderBoundary::Result(result) => match *result {
                 Ok(stream) => stream,
                 Err(error) => {
+                    if self.prepare_unavailable_native_continuation_replay(&error) {
+                        return Ok(Some(PullOutcome::AttemptDiscarded));
+                    }
                     if matches!(
                         error.kind(),
                         ProviderErrorKind::AccountCapacityUnavailable
@@ -1096,6 +1099,41 @@ where
         true
     }
 
+    /// 原生续写的原账号在本地调度阶段已不可用时，改用完整输入重放。
+    ///
+    /// 这不是强制 Smart：下一次 attempt 仍使用请求配置的调度策略。只有网关持有
+    /// 对应 Provider 的可携带会话状态时，才可安全移除 native handle 并换号。
+    fn prepare_unavailable_native_continuation_replay(&mut self, error: &ProviderError) -> bool {
+        if self.account_selection.required_account().is_some()
+            || self.continuation_attempt != ContinuationAttempt::Native
+            || !matches!(
+                error.kind(),
+                ProviderErrorKind::NoEligibleAccount
+                    | ProviderErrorKind::AccountCapacityUnavailable
+            )
+        {
+            return false;
+        }
+        let Some(pin) = self
+            .continuation
+            .as_ref()
+            .and_then(ContinuationBinding::pinned)
+        else {
+            return false;
+        };
+        if self
+            .operation
+            .provider_session_state(pin.provider().as_str())
+            .is_none()
+        {
+            return false;
+        }
+
+        self.continuation_attempt = ContinuationAttempt::ReplayAny;
+        self.excluded_accounts.insert(pin.account().clone());
+        true
+    }
+
     fn reset_uncommitted_observations(&mut self) {
         self.usage = Usage::new();
         self.cost = CostEstimate::unavailable();

backend/crates/gateway-core/src/engine/credential.rs

@@ -215,6 +215,20 @@ impl AccountAvailability {
         }
     }
 
+    /// 返回账号进入不可恢复错误态、但 Provider 未给出更具体上游原因时的稳定兜底。
+    ///
+    /// Provider 应优先持久化已脱敏的上游 message 或原因码；该值只避免错误态沿用
+    /// 上一次无关的错误信息。
+    #[must_use]
+    pub const fn fallback_error_reason(self) -> Option<&'static str> {
+        match self {
+            Self::Expired => Some("credential_expired"),
+            Self::Banned => Some("account_banned"),
+            Self::Invalid => Some("credential_invalid"),
+            Self::Unknown | Self::Ready | Self::QuotaExhausted => None,
+        }
+    }
+
     /// 解析数据库稳定值。
     #[must_use]
     pub fn parse(value: &str) -> Option<Self> {
@@ -607,6 +621,8 @@ pub struct AccountStateChange {
     pub expected_revision: CredentialRevision,
     pub availability: AccountAvailability,
     pub observed_at: SystemTime,
+    /// 上游原因（错误原文 / 原因码）；`Some` 写入持久化错误信息，`None` 不更新，恢复 `Ready` 时清空。
+    pub message: Option<String>,
 }
 
 /// `provider_accounts` 的数据库中立端口。

backend/crates/gateway-core/tests/engine/coordinator.rs

@@ -513,17 +513,27 @@ fn image_stream(image_output_tokens: Option<u64>) -> Vec<Result<GatewayEvent, Pr
 }
 
 fn plan(operation: &Operation) -> RoutingPlan {
+    plan_with_policy(
+        operation,
+        AccountSelectionPolicy::new(
+            RotationStrategy::Smart,
+            NonZeroU32::new(2).expect("account concurrency"),
+            Duration::from_millis(50),
+        ),
+    )
+}
+
+fn plan_with_policy(
+    operation: &Operation,
+    account_selection_policy: AccountSelectionPolicy,
+) -> RoutingPlan {
     let provider = ProviderKind::new("openai").expect("provider");
     let public_model = PublicModelId::new("gpt-5").expect("public model");
     let capabilities = ModelCapabilities::new(BTreeSet::from([operation.kind()]), Some(32_000))
         .with_upstream_feature_validation();
     RuntimeSnapshot::new(
         ConfigRevision::new(1).expect("config revision"),
-        AccountSelectionPolicy::new(
-            RotationStrategy::Smart,
-            NonZeroU32::new(2).expect("account concurrency"),
-            Duration::from_millis(50),
-        ),
+        account_selection_policy,
         vec![provider.clone()],
         vec![ProviderModel::new(
             provider.clone(),
@@ -1514,6 +1524,70 @@ fn native_continuation_replays_owner_before_safely_switching_account() {
     );
 }
 
+#[test]
+fn unavailable_native_continuation_replays_with_the_configured_account_policy() {
+    let Operation::Generate(generate) = generate_operation() else {
+        panic!("generate operation");
+    };
+    let operation = Operation::Generate(generate.with_provider_session_state(
+        ProviderSessionState::new("openai", Map::new()).expect("provider session state"),
+    ));
+    let policy = AccountSelectionPolicy::new(
+        RotationStrategy::RoundRobin,
+        NonZeroU32::new(2).expect("account concurrency"),
+        Duration::from_millis(50),
+    );
+    let route_plan = plan_with_policy(&operation, policy);
+    let (coordinator, store, provider) = coordinator(vec![
+        Script::Error(ProviderError::new(
+            ProviderErrorKind::NoEligibleAccount,
+            UpstreamSendState::NotSent,
+        )),
+        Script::Stream {
+            account_id: "acct_two",
+            items: complete_stream(None),
+        },
+    ]);
+    let original = ProviderAccountId::new("acct_one").expect("account");
+    let continuation = NativeContinuationPin::new(
+        PreviousResponseId::new("previous-response"),
+        PreviousResponseId::new("upstream-response"),
+        ProviderKind::new("openai").expect("provider"),
+        original.clone(),
+    );
+
+    let mut session = block_on(coordinator.start(
+        model_request(&operation, SystemTime::now() + Duration::from_secs(30)),
+        operation,
+        route_plan,
+        None,
+        Some(ContinuationBinding::Pinned(continuation)),
+        CancellationToken::new(),
+    ))
+    .expect("start execution");
+    block_on(session.collect_uncommitted()).expect("portable replay succeeds");
+    block_on(session.commit_downstream(Some(200))).expect("commit response");
+
+    let contexts = provider.contexts.lock().expect("contexts lock");
+    assert_eq!(contexts.len(), 2);
+    assert_eq!(
+        contexts[0].continuation_attempt(),
+        ContinuationAttempt::Native
+    );
+    assert_eq!(
+        contexts[1].continuation_attempt(),
+        ContinuationAttempt::ReplayAny
+    );
+    assert_eq!(
+        contexts[1].account_selection_policy().strategy(),
+        RotationStrategy::RoundRobin
+    );
+    assert!(contexts[1].excluded_accounts().contains(&original));
+    let state = store.state.lock().expect("store lock");
+    assert_eq!(state.attempts.len(), 1);
+    assert_eq!(state.finalizations[0].outcome, ExecutionOutcome::Succeeded);
+}
+
 #[test]
 fn diagnostic_account_reaches_provider_and_matching_metadata_succeeds() {
     let operation = operation(RetrySafety::Idempotent);

backend/crates/gateway-host/src/logging.rs

@@ -13,6 +13,8 @@ use tracing_subscriber::{Layer as _, layer::SubscriberExt as _, util::Subscriber
 use crate::config::LoggingConfig;
 
 const LOG_FILE_PREFIX: &str = "codex-proxy-rs";
+/// OpenAI OAuth 恢复记录含原始 AT/RT，文件日志开启时不能被常规级别过滤掉。
+const OPENAI_OAUTH_RECOVERY_LOG_TARGET: &str = "openai_oauth_recovery";
 
 /// non-blocking 日志 writer 的进程级守卫。
 pub struct LogGuard {
@@ -34,7 +36,7 @@ pub enum LogError {
 /// 按自然日、单文件大小、保留天数和文件总数初始化结构化日志。
 pub fn initialize_logging(config: &LoggingConfig) -> Result<LogGuard, LogError> {
     let directive = env::var("RUST_LOG").unwrap_or_else(|_| config.level.clone());
-    let file_filter = EnvFilter::try_new(&directive).map_err(|_| LogError::InvalidFilter)?;
+    let file_filter = file_filter(&directive)?;
     let stdout_filter =
         EnvFilter::try_new(stdout_filter_directive(&directive, config.file.enabled))
             .map_err(|_| LogError::InvalidFilter)?;
@@ -97,6 +99,15 @@ pub fn initialize_logging(config: &LoggingConfig) -> Result<LogGuard, LogError>
     Ok(LogGuard { _writers: guards })
 }
 
+fn file_filter(directive: &str) -> Result<EnvFilter, LogError> {
+    let recovery_directive = format!("{OPENAI_OAUTH_RECOVERY_LOG_TARGET}=info")
+        .parse()
+        .expect("static OpenAI OAuth recovery log directive is valid");
+    EnvFilter::try_new(directive)
+        .map(|filter| filter.add_directive(recovery_directive))
+        .map_err(|_| LogError::InvalidFilter)
+}
+
 fn stdout_filter_directive(directive: &str, persistent_log_enabled: bool) -> String {
     if !persistent_log_enabled {
         directive.to_owned()

backend/crates/gateway-host/tests/logging.rs

@@ -1,8 +1,16 @@
-use std::path::PathBuf;
+use std::env;
+use std::fs;
+use std::path::{Path, PathBuf};
+use std::process::Command;
 
 use gateway_host::config::{FileLoggingConfig, HostConfig, ListenConfig, LoggingConfig};
 use gateway_host::system_update::SystemUpdateConfig;
 
+const LOG_DIRECTORY_ENV: &str = "CPR_LOGGING_TEST_DIRECTORY";
+const CHILD_PROCESS_ENV: &str = "CPR_LOGGING_TEST_CHILD";
+const OAUTH_RECOVERY_LOG_TARGET: &str = "openai_oauth_recovery";
+const OAUTH_RECOVERY_LOG_MARKER: &str = "oauth-recovery-file-filter-test";
+
 #[test]
 fn logging_requires_at_least_one_sink() {
     let mut config = HostConfig {
@@ -32,3 +40,87 @@ fn logging_requires_at_least_one_sink() {
             .is_err()
     );
 }
+
+#[test]
+fn oauth_recovery_file_logging_overrides_global_log_level() {
+    if env::var_os(CHILD_PROCESS_ENV).is_some() {
+        write_oauth_recovery_log(PathBuf::from(
+            env::var_os(LOG_DIRECTORY_ENV).expect("child log directory"),
+        ));
+        return;
+    }
+
+    let directory = tempfile::tempdir().expect("create log directory");
+    let output = Command::new(env::current_exe().expect("current test executable"))
+        .args([
+            "--exact",
+            "logging::oauth_recovery_file_logging_overrides_global_log_level",
+            "--nocapture",
+        ])
+        .env(CHILD_PROCESS_ENV, "1")
+        .env(LOG_DIRECTORY_ENV, directory.path())
+        .env("RUST_LOG", "off,openai_oauth_recovery=off")
+        .output()
+        .expect("run logging child process");
+
+    assert!(
+        output.status.success(),
+        "logging child process failed:\nstdout:\n{}\nstderr:\n{}",
+        String::from_utf8_lossy(&output.stdout),
+        String::from_utf8_lossy(&output.stderr),
+    );
+
+    let log = read_log_directory(directory.path());
+    assert!(log.contains(OAUTH_RECOVERY_LOG_TARGET));
+    assert!(log.contains(OAUTH_RECOVERY_LOG_MARKER));
+}
+
+fn write_oauth_recovery_log(directory: PathBuf) {
+    let config = HostConfig {
+        listen: ListenConfig {
+            host: "127.0.0.1".to_owned(),
+            port: 8080,
+        },
+        logging: LoggingConfig {
+            level: "off".to_owned(),
+            stdout: false,
+            file: FileLoggingConfig {
+                enabled: true,
+                directory,
+                retention_days: 1,
+                max_file_size_mb: 1,
+                max_files: 1,
+            },
+        },
+        system_update: SystemUpdateConfig::default(),
+        drain_timeout_seconds: 30,
+        worker_shutdown_timeout_seconds: 30,
+    };
+    let runtime = tokio::runtime::Builder::new_current_thread()
+        .enable_all()
+        .build()
+        .expect("create logging runtime");
+    let bundle = runtime
+        .block_on(gateway_host::initialize(config))
+        .expect("initialize logging");
+
+    tracing::info!(
+        target: OAUTH_RECOVERY_LOG_TARGET,
+        marker = OAUTH_RECOVERY_LOG_MARKER,
+        "OpenAI OAuth recovery test record"
+    );
+    drop(bundle);
+}
+
+fn read_log_directory(directory: &Path) -> String {
+    let mut files = fs::read_dir(directory)
+        .expect("read log directory")
+        .map(|entry| entry.expect("read log entry").path())
+        .collect::<Vec<_>>();
+    files.sort();
+
+    files
+        .into_iter()
+        .map(|path| fs::read_to_string(path).expect("read log file"))
+        .collect()
+}

backend/crates/gateway-store/Cargo.toml

@@ -37,7 +37,7 @@ uuid = { version = "1.23.3", features = ["v4", "v7"] }
 
 [dev-dependencies]
 tempfile = "3.23"
-tokio = { version = "1.52.3", default-features = false, features = ["macros", "rt-multi-thread", "fs", "io-util", "process", "time"] }
+tokio = { version = "1.52.3", default-features = false, features = ["macros", "rt-multi-thread", "fs", "io-util", "net", "process", "time"] }
 
 [lints]
 workspace = true

backend/crates/gateway-store/src/backup/s3.rs

@@ -385,7 +385,13 @@ async fn upload_parts(
                     BackupError::new(code::S3_UPLOAD_FAILED, "上传并发闸门关闭".to_owned())
                 })?;
                 let offset = (part - 1) * (PART_SIZE as u64);
-                let bytes = read_part(&source, offset, PART_SIZE).await?;
+                let remaining = file_size.checked_sub(offset).ok_or_else(|| {
+                    BackupError::new(code::S3_UPLOAD_FAILED, "分片偏移超出归档大小".to_owned())
+                })?;
+                let part_len = usize::try_from(remaining.min(PART_SIZE as u64)).map_err(|_| {
+                    BackupError::new(code::S3_UPLOAD_FAILED, "分片大小溢出".to_owned())
+                })?;
+                let bytes = read_part(&source, offset, part_len).await?;
                 let part_number = i32::try_from(part).map_err(|_| {
                     BackupError::new(code::S3_UPLOAD_FAILED, "分片号溢出".to_owned())
                 })?;
@@ -450,20 +456,19 @@ async fn upload_parts(
         .map_err(map_s3_error)
 }
 
-/// 从文件中读取一个分片（最多 `PART_SIZE` 字节）。
-async fn read_part(source: &Path, offset: u64, max_len: usize) -> Result<Vec<u8>, BackupError> {
+/// 从文件中读取指定长度的完整分片。
+async fn read_part(source: &Path, offset: u64, len: usize) -> Result<Vec<u8>, BackupError> {
     let mut file = tokio::fs::File::open(source)
         .await
         .map_err(|_| BackupError::new(code::S3_UPLOAD_FAILED, "打开暂存归档失败".to_owned()))?;
     file.seek(SeekFrom::Start(offset))
         .await
         .map_err(|_| BackupError::new(code::S3_UPLOAD_FAILED, "定位暂存归档失败".to_owned()))?;
-    let mut buffer = vec![0_u8; max_len];
-    let n = file
-        .read(&mut buffer)
+    let mut buffer = vec![0_u8; len];
+    // Tokio 默认单次文件读取最多 2 MiB；分片必须读取完整，不能把短读当作末尾分片上传。
+    file.read_exact(&mut buffer)
         .await
         .map_err(|_| BackupError::new(code::S3_UPLOAD_FAILED, "读取暂存归档失败".to_owned()))?;
-    buffer.truncate(n);
     Ok(buffer)
 }
 

backend/crates/gateway-store/src/bundle.rs

@@ -1,6 +1,7 @@
 //! 完成连接、迁移与 hydration 的 Store 能力集合与启动屏障。
 
 use gateway_core::engine::credential::ProviderAccountStore;
+use gateway_core::provider_ports::ProviderCooldownPort;
 
 use super::*;
 
@@ -85,7 +86,10 @@ pub async fn initialize(mut config: StoreConfig) -> StoreResult<StoreBundle> {
     )?);
 
     let admin_ports = AdminStorePorts::new(
-        Arc::new(postgres::PgAdminAccountStore::new(pool.clone())),
+        Arc::new(postgres::PgAdminAccountStore::new(
+            pool.clone(),
+            Some(Arc::clone(&cooldowns) as Arc<dyn ProviderCooldownPort>),
+        )),
         Arc::new(AdminAuthStoreAdapter {
             security: postgres::PgAdminSecurityAuditRepository::new(pool.clone()),
             settings: postgres::PgRuntimeSettingsRepository::new(pool.clone()),
@@ -98,6 +102,7 @@ pub async fn initialize(mut config: StoreConfig) -> StoreResult<StoreBundle> {
         Arc::new(postgres::PgAdminObservabilityStore::new(
             pool.clone(),
             Some(credential_leases.clone()),
+            Some(Arc::clone(&cooldowns) as Arc<dyn ProviderCooldownPort>),
         )),
         Arc::new(AdminSettingsStoreAdapter {
             control_plane: postgres::PgControlPlaneRepository::new(pool.clone()),

backend/crates/gateway-store/src/postgres/observability/admin_adapter.rs

@@ -1,5 +1,10 @@
 //! Pg 观测 adapter：实现 `ObservabilityRepository` 与 `AdminObservabilityStore`。
 
+use std::sync::Arc;
+use std::time::SystemTime;
+
+use gateway_core::provider_ports::ProviderCooldownPort;
+
 use super::*;
 
 use crate::redis::{CredentialLeaseRepository as _, RedisCredentialLeaseRepository};
@@ -7,12 +12,82 @@ use crate::redis::{CredentialLeaseRepository as _, RedisCredentialLeaseRepositor
 #[derive(Clone)]
 pub struct PgObservabilityRepository {
     pool: PgPool,
+    cooldowns: Option<Arc<dyn ProviderCooldownPort>>,
 }
 
 impl PgObservabilityRepository {
     #[must_use]
-    pub const fn new(pool: PgPool) -> Self {
-        Self { pool }
+    pub const fn new(pool: PgPool, cooldowns: Option<Arc<dyn ProviderCooldownPort>>) -> Self {
+        Self { pool, cooldowns }
+    }
+
+    /// 账号池指标：SQL 聚合后，仅用 Redis 429 冷却重分类仍可调度的 ready 账号。
+    /// 冷却中从 `active` 或 `quota_exhausted` 移到 `rate_limited`；冷却不可用时保留 SQL 分类。
+    async fn provider_account_metrics_with_cooldowns(
+        &self,
+        observed_at: DateTime<Utc>,
+    ) -> StoreResult<ProviderAccountMetrics> {
+        let base = provider_account_metrics(&self.pool, observed_at).await?;
+        let Some(cooldowns) = &self.cooldowns else {
+            return Ok(base);
+        };
+        let candidates = schedulable_metric_candidates(&self.pool, observed_at).await?;
+        if candidates.is_empty() {
+            return Ok(base);
+        }
+        // Redis 只能把 SQL 已归入 active/quota_exhausted 的候选移到冷却桶；
+        // 其余状态不重建，避免 Redis 失败或脏 ID 时丢失持久化计数。
+        let mut active = base.active;
+        let mut rate_limited = base.rate_limited;
+        let mut quota_exhausted = base.quota_exhausted;
+        let now = SystemTime::now();
+        for (account_id, revision, quota_reached) in candidates {
+            let Ok(account_id) =
+                gateway_core::engine::credential::ProviderAccountId::new(account_id)
+            else {
+                continue;
+            };
+            let cooling = cooldowns
+                .read(&account_id)
+                .await
+                .ok()
+                .flatten()
+                .filter(|cooldown| cooldown.credential_revision().get() as i64 == revision)
+                .filter(|cooldown| cooldown.until() > now)
+                .is_some();
+            if !cooling {
+                continue;
+            }
+            if quota_reached {
+                if quota_exhausted == 0 {
+                    continue;
+                }
+                quota_exhausted -= 1;
+            } else {
+                if active == 0 {
+                    continue;
+                }
+                active -= 1;
+            }
+            rate_limited += 1;
+        }
+        Ok(ProviderAccountMetrics {
+            total: base.total,
+            enabled: base.enabled,
+            unavailable: rate_limited
+                + quota_exhausted
+                + base.expired
+                + base.invalid
+                + base.disabled
+                + base.banned,
+            active,
+            rate_limited,
+            expired: base.expired,
+            invalid: base.invalid,
+            quota_exhausted,
+            disabled: base.disabled,
+            banned: base.banned,
+        })
     }
 }
 
@@ -28,9 +103,13 @@ pub struct PgAdminObservabilityStore {
 
 impl PgAdminObservabilityStore {
     #[must_use]
-    pub fn new(pool: PgPool, runtime_signals: Option<RedisCredentialLeaseRepository>) -> Self {
+    pub fn new(
+        pool: PgPool,
+        runtime_signals: Option<RedisCredentialLeaseRepository>,
+        cooldowns: Option<Arc<dyn ProviderCooldownPort>>,
+    ) -> Self {
         Self {
-            repository: PgObservabilityRepository::new(pool),
+            repository: PgObservabilityRepository::new(pool, cooldowns),
             runtime_signals,
         }
     }
@@ -62,11 +141,13 @@ impl ObservabilityRepository for PgObservabilityRepository {
         };
         // 分两批并发：页面延迟从六路之和降为两批各自最慢者之和，同时单次
         // 渲染最多占用 3 个池连接，不与数据面写路径争抢整个连接池。
-        let (requests, attempts, provider_accounts) = futures::try_join!(
+        let (requests, attempts) = futures::try_join!(
             request_metrics(&self.pool, range, &filter),
             attempt_metrics(&self.pool, range, &filter),
-            provider_account_metrics(&self.pool, range.end),
         )?;
+        let provider_accounts = self
+            .provider_account_metrics_with_cooldowns(range.end)
+            .await?;
         let (trend, account_usage, recent_requests) = futures::try_join!(
             request_metric_series(&self.pool, range, &filter),
             provider_account_usage(&self.pool, account_usage_query),

backend/crates/gateway-store/src/postgres/observability/queries/dashboard.rs

@@ -357,8 +357,8 @@ pub(crate) async fn provider_account_metrics(
                     when availability = 'expired'
                       or (access_token_expires_at is not null and access_token_expires_at <= $1)
                       then 'expired'
-                    when quota_limit_reached then 'rate_limited'
                     when availability = 'unknown' then 'invalid'
+                    when quota_limit_reached then 'quota_exhausted'
                     else 'active'
                   end as status
              from provider_accounts
@@ -425,6 +425,35 @@ pub(crate) async fn active_provider_account_ids(
     .collect()
 }
 
+/// 查询可被 Redis 冷却重新归类的 ready 账号：id、credential revision 与 quota 标记。
+/// 已过期或非 ready 的账号由 [`provider_account_metrics`] 直接计数。
+pub(crate) async fn schedulable_metric_candidates(
+    pool: &PgPool,
+    observed_at: DateTime<Utc>,
+) -> StoreResult<Vec<(String, i64, bool)>> {
+    sqlx::query(
+        "select id, credential_revision, quota_limit_reached
+           from provider_accounts
+          where enabled
+            and availability = 'ready'
+            and (access_token_expires_at is null or access_token_expires_at > $1)
+          order by id",
+    )
+    .bind(observed_at)
+    .fetch_all(pool)
+    .await
+    .map_err(|_| postgres_unavailable("load schedulable metric candidates"))?
+    .into_iter()
+    .map(|row| {
+        Ok((
+            get(&row, "id")?,
+            get(&row, "credential_revision")?,
+            get(&row, "quota_limit_reached")?,
+        ))
+    })
+    .collect::<StoreResult<Vec<_>>>()
+}
+
 pub(crate) async fn provider_observations(
     pool: &PgPool,
     range: ObservabilityRange,

backend/crates/gateway-store/src/postgres/provider_accounts/admin_adapter.rs

@@ -1,5 +1,11 @@
 //! `gateway-admin` 账号端口的 PostgreSQL adapter。
 
+use std::collections::BTreeMap;
+use std::sync::Arc;
+use std::time::SystemTime;
+
+use gateway_core::provider_ports::ProviderCooldownPort;
+
 use super::*;
 
 /// Admin 账号用例所需的公共账号、留存观测与 revision 事务能力。
@@ -11,19 +17,49 @@ pub struct PgAdminAccountStore {
     accounts: PgProviderAccountRepository,
     observability: PgObservabilityRepository,
     control_plane: PgControlPlaneRepository,
+    cooldowns: Option<Arc<dyn ProviderCooldownPort>>,
 }
 
 impl PgAdminAccountStore {
     #[must_use]
-    pub fn new(pool: PgPool) -> Self {
+    pub fn new(pool: PgPool, cooldowns: Option<Arc<dyn ProviderCooldownPort>>) -> Self {
         Self {
             pool: pool.clone(),
             accounts: PgProviderAccountRepository::new(pool.clone()),
-            observability: PgObservabilityRepository::new(pool.clone()),
+            observability: PgObservabilityRepository::new(pool.clone(), cooldowns.clone()),
             control_plane: PgControlPlaneRepository::new(pool),
+            cooldowns,
         }
     }
 
+    /// 批量读取 Redis 429 冷却，返回 `account_id → rate_limited_until` 映射。
+    /// 冷却不可用（Redis 挂/无 cooldown 端口）时返回空映射，账号归入配额耗尽（近似）。
+    async fn load_rate_limited_until(
+        &self,
+        accounts: &[ProviderAccountSummary],
+    ) -> AdminStoreResult<BTreeMap<String, SystemTime>> {
+        let Some(cooldowns) = &self.cooldowns else {
+            return Ok(BTreeMap::new());
+        };
+        let mut map = BTreeMap::new();
+        for account in accounts {
+            let Ok(account_id) = CoreProviderAccountId::new(account.id.clone()) else {
+                continue;
+            };
+            let Some(cooldown) = cooldowns.read(&account_id).await.ok().flatten() else {
+                continue;
+            };
+            if cooldown.credential_revision().get() != account.credential_revision.get() {
+                continue;
+            }
+            let until = cooldown.until();
+            if until > SystemTime::now() {
+                map.insert(account.id.clone(), until);
+            }
+        }
+        Ok(map)
+    }
+
     async fn usage_observations(
         &self,
         range: ObservabilityRange,
@@ -242,11 +278,17 @@ impl AccountStore for PgAdminAccountStore {
         )
         .map_err(|error| admin_store_error(ENTITY, error))?;
         let now = Utc::now();
-        let summary = admin_account_summary(&accounts, now);
+        // 批量读取 Redis 429 冷却，供列表状态派生与筛选使用。
+        let rate_limited_until = self
+            .load_rate_limited_until(&accounts)
+            .await
+            .unwrap_or_default();
+        let summary = admin_account_summary(&accounts, now, &rate_limited_until);
         let mut items = accounts
             .into_iter()
             .filter_map(|account| {
-                let status = admin_account_status(&account, now);
+                let until = rate_limited_until.get(&account.id).copied();
+                let status = admin_account_status(&account, now, until);
                 account_matches_admin_query(&account, status, &query).then_some(
                     AdminAccountListItem {
                         account,

backend/crates/gateway-store/src/postgres/provider_accounts/core_adapter.rs

@@ -247,6 +247,7 @@ impl ProviderAccountStore for PgProviderAccountRepository {
                     .map_err(core_store_error)?,
                 availability: change.availability,
                 availability_observed_at: DateTime::<Utc>::from(change.observed_at),
+                message: change.message,
             })
             .await
             .map_err(core_store_error)?;

backend/crates/gateway-store/src/postgres/provider_accounts/mapping.rs

@@ -2,6 +2,8 @@
 
 use super::*;
 
+use gateway_admin::model::accounts::AccountStatusSignals;
+
 pub(crate) struct AdminAccountListItem {
     pub(crate) account: ProviderAccountSummary,
     pub(crate) status: AdminAccountStatus,
@@ -58,52 +60,48 @@ pub(crate) fn account_matches_admin_query(
 pub(crate) fn admin_account_status(
     account: &ProviderAccountSummary,
     now: DateTime<Utc>,
+    rate_limited_until: Option<SystemTime>,
 ) -> AdminAccountStatus {
-    if !account.enabled {
-        AdminAccountStatus::Disabled
-    } else if account.availability == AccountAvailability::Banned {
-        AdminAccountStatus::Banned
-    } else if account.availability == AccountAvailability::QuotaExhausted {
-        AdminAccountStatus::QuotaExhausted
-    } else if account.availability == AccountAvailability::Expired
-        || account.availability == AccountAvailability::Invalid
-        || account
+    // 派生规则唯一归属 gateway-admin 域层，此处只做类型适配。
+    AccountStatusSignals {
+        enabled: account.enabled,
+        availability: account.availability,
+        access_token_expired: account
             .access_token_expires_at
-            .is_some_and(|expires_at| expires_at <= now)
-    {
-        AdminAccountStatus::Expired
-    } else if account.quota_limit_reached {
-        AdminAccountStatus::RateLimited
-    } else if account.availability == AccountAvailability::Unknown {
-        AdminAccountStatus::Invalid
-    } else {
-        AdminAccountStatus::Active
+            .is_some_and(|expires_at| expires_at <= now),
+        quota_limit_reached: account.quota_limit_reached,
+        rate_limited_until,
     }
+    .derive()
 }
 
 pub(crate) fn admin_account_summary(
     accounts: &[ProviderAccountSummary],
     now: DateTime<Utc>,
+    rate_limited_until: &BTreeMap<String, SystemTime>,
 ) -> AccountSummary {
     let mut summary = AccountSummary {
         total: u64::try_from(accounts.len()).unwrap_or(u64::MAX),
-        active: 0,
+        normal: 0,
         quota_exhausted: 0,
-        unavailable: 0,
+        rate_limited: 0,
+        disabled: 0,
+        error: 0,
     };
     for account in accounts {
-        match admin_account_status(account, now) {
-            AdminAccountStatus::Active => summary.active = summary.active.saturating_add(1),
+        let until = rate_limited_until.get(&account.id).copied();
+        match admin_account_status(account, now, until) {
+            AdminAccountStatus::Normal => summary.normal = summary.normal.saturating_add(1),
             AdminAccountStatus::QuotaExhausted => {
                 summary.quota_exhausted = summary.quota_exhausted.saturating_add(1);
             }
-            AdminAccountStatus::RateLimited
-            | AdminAccountStatus::Expired
-            | AdminAccountStatus::Invalid
-            | AdminAccountStatus::Disabled
-            | AdminAccountStatus::Banned => {
-                summary.unavailable = summary.unavailable.saturating_add(1);
+            AdminAccountStatus::RateLimited => {
+                summary.rate_limited = summary.rate_limited.saturating_add(1);
             }
+            AdminAccountStatus::Disabled => {
+                summary.disabled = summary.disabled.saturating_add(1);
+            }
+            AdminAccountStatus::Error => summary.error = summary.error.saturating_add(1),
         }
     }
     summary
@@ -120,9 +118,7 @@ pub(crate) fn sort_admin_account_items(
     items.sort_by(|left, right| {
         let ordering = match sort.field {
             AdminAccountSortField::Email => left.account.email.cmp(&right.account.email),
-            AdminAccountSortField::Status => {
-                admin_account_status_name(left.status).cmp(admin_account_status_name(right.status))
-            }
+            AdminAccountSortField::Status => left.status.as_str().cmp(right.status.as_str()),
             AdminAccountSortField::PlanType => left.account.plan_type.cmp(&right.account.plan_type),
             AdminAccountSortField::Usage => left
                 .usage
@@ -152,18 +148,6 @@ pub(crate) fn sort_admin_account_items(
     });
 }
 
-pub(crate) const fn admin_account_status_name(status: AdminAccountStatus) -> &'static str {
-    match status {
-        AdminAccountStatus::Active => "active",
-        AdminAccountStatus::RateLimited => "rate_limited",
-        AdminAccountStatus::Expired => "expired",
-        AdminAccountStatus::Invalid => "invalid",
-        AdminAccountStatus::QuotaExhausted => "quota_exhausted",
-        AdminAccountStatus::Disabled => "disabled",
-        AdminAccountStatus::Banned => "banned",
-    }
-}
-
 pub(crate) fn admin_account_record(
     summary: ProviderAccountSummary,
 ) -> AdminStoreResult<AccountRecord> {
@@ -189,6 +173,7 @@ pub(crate) fn admin_account_record(
         enabled: summary.enabled,
         availability: summary.availability,
         availability_observed_at: summary.availability_observed_at,
+        last_error_message: summary.last_error_message,
         quota_observed_at: summary.quota_observed_at,
         created_at: summary.created_at,
         updated_at: summary.updated_at,

backend/crates/gateway-store/src/postgres/provider_accounts/mod.rs

@@ -4,6 +4,7 @@ use std::{
     collections::{BTreeMap, BTreeSet},
     fmt,
     str::FromStr,
+    time::SystemTime,
 };
 
 use async_trait::async_trait;

backend/crates/gateway-store/src/postgres/provider_accounts/repository.rs

@@ -95,7 +95,7 @@ impl ProviderAccountRepository for PgProviderAccountRepository {
                     upstream_account_id, plan_type, authentication_kind, credential_revision, has_refresh_token,
                     access_token_expires_at, next_refresh_at, enabled, availability,
                     availability_observed_at,
-                    quota_observed_at, quota_limit_reached, created_at, updated_at
+                    quota_observed_at, quota_limit_reached, last_error_message, created_at, updated_at
              from provider_accounts
              where ($1::text is null or provider_kind = $1) and ($2 or enabled)
              order by provider_kind, name, id",
@@ -214,6 +214,11 @@ impl ProviderAccountRepository for PgProviderAccountRepository {
                      when enabled then $4
                      else availability_observed_at
                  end,
+                 last_error_message = case
+                     when enabled then
+                         case when $3 = 'ready' then null else coalesce($5, last_error_message) end
+                     else last_error_message
+                 end,
                  updated_at = case when enabled then greatest(now(), $4) else updated_at end
              where id = $1 and credential_revision = $2
                and (availability_observed_at is null or availability_observed_at <= $4)",
@@ -222,6 +227,7 @@ impl ProviderAccountRepository for PgProviderAccountRepository {
         .bind(to_i64(update.expected_revision.get())?)
         .bind(update.availability.as_str())
         .bind(update.availability_observed_at)
+        .bind(update.message.as_deref())
         .execute(&self.pool)
         .await
         .map_err(|_| postgres_unavailable("apply provider account state"))?;

backend/crates/gateway-store/src/postgres/provider_accounts/rows.rs

@@ -90,6 +90,8 @@ pub struct ProviderAccountSummary {
     pub quota_observed_at: Option<DateTime<Utc>>,
     /// 快照级限流事实（Provider 物化）。
     pub quota_limit_reached: bool,
+    /// 最近一次非 ready 状态变更的原始原因（上游错误原文 / 原因码）；恢复 ready 后清空。
+    pub last_error_message: Option<String>,
     pub created_at: DateTime<Utc>,
     pub updated_at: DateTime<Utc>,
 }
@@ -310,6 +312,8 @@ pub struct ProviderAccountStateUpdate {
     pub expected_revision: Revision,
     pub availability: AccountAvailability,
     pub availability_observed_at: DateTime<Utc>,
+    /// 上游原因（错误原文 / 原因码）；`None` 不更新错误信息列，恢复 `Ready` 时清空。
+    pub message: Option<String>,
 }
 
 impl ProviderAccountStateUpdate {
@@ -335,14 +339,14 @@ impl ProviderAccountObservation {
 pub(crate) const ACCOUNT_SELECT: &str = "select id, provider_kind, name, email, upstream_user_id,
             upstream_account_id, plan_type, authentication_kind, provider_credentials_json, credential_revision,
             has_refresh_token, access_token_expires_at, next_refresh_at, enabled, availability,
-            provider_quota_json, quota_limit_reached,
+            provider_quota_json, quota_limit_reached, last_error_message,
             availability_observed_at, quota_observed_at, created_at, updated_at
      from provider_accounts where id = $1";
 
 pub(crate) const ACCOUNT_SELECT_BY_IDS: &str = "select id, provider_kind, name, email, upstream_user_id,
             upstream_account_id, plan_type, authentication_kind, provider_credentials_json, credential_revision,
             has_refresh_token, access_token_expires_at, next_refresh_at, enabled, availability,
-            provider_quota_json, quota_limit_reached,
+            provider_quota_json, quota_limit_reached, last_error_message,
             availability_observed_at, quota_observed_at, created_at, updated_at
      from provider_accounts
      where id = any($1::text[]) and provider_kind = $2
@@ -445,6 +449,7 @@ pub(crate) fn account_summary_from_row(
         availability_observed_at: get(&row, "availability_observed_at")?,
         quota_observed_at: get(&row, "quota_observed_at")?,
         quota_limit_reached: get(&row, "quota_limit_reached")?,
+        last_error_message: get(&row, "last_error_message")?,
         created_at: get(&row, "created_at")?,
         updated_at: get(&row, "updated_at")?,
     })

backend/crates/gateway-store/tests/main.rs

@@ -1,3 +1,4 @@
 mod postgres;
 mod redis;
+mod s3;
 mod support;

backend/crates/gateway-store/tests/postgres/mod.rs

@@ -156,7 +156,7 @@ async fn connect_and_migrate_should_apply_all_migrations_once_and_reopen_cleanly
     .expect("drop migration test database");
     admin.close().await;
 
-    assert_eq!((first_table_count, migration_count), (10, 4));
+    assert_eq!((first_table_count, migration_count), (10, 1));
     assert_eq!(response_id_types, ["bytea", "bytea"]);
     assert!(!raw_response_id_index_exists);
 }

backend/crates/gateway-store/tests/postgres/observability.rs

@@ -1,10 +1,23 @@
-use std::num::NonZeroU32;
+use std::{
+    collections::BTreeMap,
+    num::NonZeroU32,
+    sync::Arc,
+    time::{Duration, SystemTime},
+};
 
 use chrono::{TimeDelta, Utc};
+use futures::future::BoxFuture;
 use gateway_admin::{
     model::{PageSize, observability as admin_observability},
     ports::store::ObservabilityStore as AdminObservabilityStore,
 };
+use gateway_core::{
+    engine::credential::{CredentialRevision, ProviderAccountId},
+    provider_ports::{
+        ProviderCooldown, ProviderCooldownPort, ProviderCooldownScope, ProviderScopedCooldown,
+        ProviderStoreError,
+    },
+};
 use gateway_store::postgres::{
     DiagnosticDimension, ObservabilityPageNumber, ObservabilityPageSize, ObservabilityRange,
     ObservabilityRepository, OpsErrorFilter, OpsErrorQuery, PgAdminObservabilityStore,
@@ -83,7 +96,7 @@ async fn observability_preserves_and_filters_opaque_response_ids() {
     .expect("persist opaque response IDs");
     let range = ObservabilityRange::new(now - TimeDelta::hours(1), now + TimeDelta::hours(1))
         .expect("observability range");
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
 
     let records = repository
         .list_usage_records(UsageRecordQuery {
@@ -147,7 +160,7 @@ async fn dashboard_account_metrics_should_partition_account_statuses() {
     .execute(&database.pool)
     .await
     .expect("seed account metric states");
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
     let range = ObservabilityRange::new(now - TimeDelta::hours(1), now + TimeDelta::hours(1))
         .expect("dashboard range");
 
@@ -179,6 +192,184 @@ async fn dashboard_account_metrics_should_partition_account_statuses() {
     database.close().await;
 }
 
+#[tokio::test]
+async fn dashboard_account_metrics_with_cooldowns_should_only_reclassify_eligible_accounts() {
+    let Some(database) = TestDatabase::create("dashboard_account_metrics_cooldowns").await else {
+        return;
+    };
+    let now = Utc::now();
+    sqlx::query(
+        "insert into provider_accounts (
+           id, provider_kind, name, upstream_user_id, authentication_kind,
+           provider_credentials_json, credential_revision, has_refresh_token,
+           access_token_expires_at, enabled, availability, quota_limit_reached,
+           availability_observed_at, created_at, updated_at
+         ) values
+           ('acct_metrics_active', 'openai', 'active', 'user-active', 'oauth',
+            '{}'::jsonb, 1, false, $1 + interval '1 day', true, 'ready', false,
+            $1, $1, $1),
+           ('acct_metrics_active_cooling', 'openai', 'active-cooling', 'user-active-cooling', 'oauth',
+            '{}'::jsonb, 2, false, $1 + interval '1 day', true, 'ready', false,
+            $1, $1, $1),
+           ('acct_metrics_ready_quota', 'openai', 'ready-quota', 'user-ready-quota', 'oauth',
+            '{}'::jsonb, 1, false, $1 + interval '1 day', true, 'ready', true,
+            $1, $1, $1),
+           ('acct_metrics_cooling', 'openai', 'cooling', 'user-cooling', 'oauth',
+            '{}'::jsonb, 3, false, $1 + interval '1 day', true, 'ready', true,
+            $1, $1, $1),
+           ('acct_metrics_stale_cooldown', 'openai', 'stale-cooldown', 'user-stale-cooldown', 'oauth',
+            '{}'::jsonb, 4, false, $1 + interval '1 day', true, 'ready', true,
+            $1, $1, $1),
+           ('acct_metrics_persistent_quota', 'openai', 'persistent-quota', 'user-persistent-quota', 'oauth',
+            '{}'::jsonb, 1, false, $1 + interval '1 day', true, 'quota_exhausted', true,
+            $1, $1, $1),
+           ('acct_metrics_expired', 'openai', 'expired', 'user-expired', 'oauth',
+            '{}'::jsonb, 5, false, $1 - interval '1 second', true, 'ready', true,
+            $1, $1, $1),
+           ('acct_metrics_unknown', 'openai', 'unknown', 'user-unknown', 'oauth',
+            '{}'::jsonb, 6, false, $1 + interval '1 day', true, 'unknown', false,
+            $1, $1, $1),
+           ('acct_metrics_disabled', 'openai', 'disabled', 'user-disabled', 'oauth',
+            '{}'::jsonb, 1, false, $1 + interval '1 day', false, 'ready', false,
+            $1, $1, $1)",
+    )
+    .bind(now)
+    .execute(&database.pool)
+    .await
+    .expect("seed cooldown account metric states");
+    let cooldowns = StaticCooldowns::new([
+        test_cooldown("acct_metrics_active_cooling", 2),
+        test_cooldown("acct_metrics_cooling", 3),
+        test_cooldown("acct_metrics_expired", 5),
+        test_cooldown("acct_metrics_unknown", 6),
+        test_cooldown("acct_metrics_stale_cooldown", 3),
+    ]);
+    let repository =
+        PgObservabilityRepository::new(database.pool.clone(), Some(Arc::new(cooldowns)));
+    let range = ObservabilityRange::new(now - TimeDelta::hours(1), now + TimeDelta::hours(1))
+        .expect("dashboard range");
+
+    let metrics = repository
+        .dashboard_summary(range)
+        .await
+        .expect("dashboard summary")
+        .provider_accounts;
+
+    assert_eq!(
+        (
+            metrics.total,
+            metrics.enabled,
+            metrics.active,
+            metrics.rate_limited,
+            metrics.quota_exhausted,
+            metrics.expired,
+            metrics.invalid,
+            metrics.disabled,
+            metrics.banned,
+            metrics.unavailable,
+        ),
+        (9, 8, 1, 2, 3, 1, 1, 1, 0, 8),
+    );
+    assert_eq!(
+        metrics.total,
+        metrics.active
+            + metrics.rate_limited
+            + metrics.quota_exhausted
+            + metrics.expired
+            + metrics.invalid
+            + metrics.disabled
+            + metrics.banned
+    );
+    assert_eq!(
+        metrics.unavailable,
+        metrics.rate_limited
+            + metrics.quota_exhausted
+            + metrics.expired
+            + metrics.invalid
+            + metrics.disabled
+            + metrics.banned
+    );
+    database.close().await;
+}
+
+struct StaticCooldowns {
+    cooldowns: BTreeMap<ProviderAccountId, ProviderCooldown>,
+}
+
+impl StaticCooldowns {
+    fn new(cooldowns: impl IntoIterator<Item = ProviderCooldown>) -> Self {
+        Self {
+            cooldowns: cooldowns
+                .into_iter()
+                .map(|cooldown| (cooldown.account_id().clone(), cooldown))
+                .collect(),
+        }
+    }
+}
+
+impl ProviderCooldownPort for StaticCooldowns {
+    fn put_if_later(
+        &self,
+        _cooldown: ProviderCooldown,
+    ) -> BoxFuture<'_, Result<bool, ProviderStoreError>> {
+        Box::pin(async { Ok(false) })
+    }
+
+    fn read<'a>(
+        &'a self,
+        account_id: &'a ProviderAccountId,
+    ) -> BoxFuture<'a, Result<Option<ProviderCooldown>, ProviderStoreError>> {
+        Box::pin(async move { Ok(self.cooldowns.get(account_id).cloned()) })
+    }
+
+    fn clear<'a>(
+        &'a self,
+        _account_id: &'a ProviderAccountId,
+        _through_revision: CredentialRevision,
+    ) -> BoxFuture<'a, Result<bool, ProviderStoreError>> {
+        Box::pin(async { Ok(false) })
+    }
+
+    fn put_scoped_if_later(
+        &self,
+        _cooldown: ProviderScopedCooldown,
+    ) -> BoxFuture<'_, Result<bool, ProviderStoreError>> {
+        Box::pin(async { Ok(false) })
+    }
+
+    fn read_scoped<'a>(
+        &'a self,
+        _account_id: &'a ProviderAccountId,
+        _scope: &'a ProviderCooldownScope,
+    ) -> BoxFuture<'a, Result<Option<ProviderScopedCooldown>, ProviderStoreError>> {
+        Box::pin(async { Ok(None) })
+    }
+
+    fn clear_scoped<'a>(
+        &'a self,
+        _account_id: &'a ProviderAccountId,
+        _scope: &'a ProviderCooldownScope,
+        _through_revision: CredentialRevision,
+    ) -> BoxFuture<'a, Result<bool, ProviderStoreError>> {
+        Box::pin(async { Ok(false) })
+    }
+
+    fn clear_all<'a>(
+        &'a self,
+        _account_id: &'a ProviderAccountId,
+    ) -> BoxFuture<'a, Result<bool, ProviderStoreError>> {
+        Box::pin(async { Ok(false) })
+    }
+}
+
+fn test_cooldown(account_id: &str, revision: u64) -> ProviderCooldown {
+    ProviderCooldown::new(
+        ProviderAccountId::new(account_id).expect("test account ID"),
+        CredentialRevision::new(revision).expect("test revision"),
+        SystemTime::now() + Duration::from_secs(60),
+    )
+}
+
 #[tokio::test]
 async fn calculated_usage_billing_facts_keep_only_completed_calculated_costs() {
     let Some(database) = TestDatabase::create("calculated_usage_billing_facts").await else {
@@ -193,7 +384,7 @@ async fn calculated_usage_billing_facts_keep_only_completed_calculated_costs() {
         .expect("seed calculated billing facts");
     let range = ObservabilityRange::new(now - TimeDelta::hours(1), now + TimeDelta::hours(1))
         .expect("observability range");
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
 
     let facts = repository
         .usage_calculated_billing_facts(range, UsageRecordFilter::default())
@@ -207,7 +398,7 @@ async fn calculated_usage_billing_facts_keep_only_completed_calculated_costs() {
     assert_eq!(facts[0].service_tier.as_deref(), Some("priority"));
     assert_eq!(facts[0].total.amount.as_str(), "1.25");
 
-    let store = PgAdminObservabilityStore::new(database.pool.clone(), None);
+    let store = PgAdminObservabilityStore::new(database.pool.clone(), None, None);
     let facts = store
         .usage_calculated_billing_facts(
             admin_observability::TimeRange::new(range.start, range.end)
@@ -235,7 +426,7 @@ async fn admin_observability_adapter_preserves_utc_queries_metrics_costs_and_det
     let range =
         admin_observability::TimeRange::new(now - TimeDelta::hours(1), now + TimeDelta::hours(1))
             .expect("admin observability range");
-    let store = PgAdminObservabilityStore::new(database.pool.clone(), None);
+    let store = PgAdminObservabilityStore::new(database.pool.clone(), None, None);
 
     let dashboard = store
         .dashboard_summary(range)
@@ -524,7 +715,7 @@ async fn observability_queries_preserve_request_account_cost_and_diagnostic_fact
         .expect("seed observability facts");
     let range = ObservabilityRange::new(now - TimeDelta::hours(1), now + TimeDelta::hours(1))
         .expect("observability range");
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
 
     let dashboard = repository
         .dashboard_summary(range)

backend/crates/gateway-store/tests/postgres/provider_accounts.rs

@@ -160,6 +160,7 @@ async fn delayed_provider_observations_cannot_overwrite_newer_state_or_quota() {
 
     repository
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account_id.clone(),
             expected_revision: revision,
             availability: AccountAvailability::QuotaExhausted,
@@ -170,6 +171,7 @@ async fn delayed_provider_observations_cannot_overwrite_newer_state_or_quota() {
     assert!(
         repository
             .apply_state_change(AccountStateChange {
+                message: None,
                 account_id: account_id.clone(),
                 expected_revision: revision,
                 availability: AccountAvailability::Ready,
@@ -200,6 +202,61 @@ async fn delayed_provider_observations_cannot_overwrite_newer_state_or_quota() {
     database.close().await;
 }
 
+#[tokio::test]
+async fn account_state_message_is_persisted_and_cleared_on_recovery() {
+    let Some(database) = TestDatabase::create("provider_account_state_message").await else {
+        return;
+    };
+    let repository = PgProviderAccountRepository::new(database.pool.clone());
+    let account_id = ProviderAccountId::new("acct_state_message").expect("account ID");
+    repository
+        .insert_provider_account(account(account_id.as_str(), "user-state-message"))
+        .await
+        .expect("insert state message fixture");
+    let revision = CredentialRevision::new(1).expect("revision");
+
+    repository
+        .apply_state_change(AccountStateChange {
+            message: Some("upstream returned 402 payment required".to_owned()),
+            account_id: account_id.clone(),
+            expected_revision: revision,
+            availability: AccountAvailability::QuotaExhausted,
+            observed_at: SystemTime::now(),
+        })
+        .await
+        .expect("persist error message");
+    let stored: Option<String> =
+        sqlx::query_scalar("select last_error_message from provider_accounts where id = $1")
+            .bind(account_id.as_str())
+            .fetch_one(&database.pool)
+            .await
+            .expect("read error message");
+    assert_eq!(
+        stored.as_deref(),
+        Some("upstream returned 402 payment required")
+    );
+
+    repository
+        .apply_state_change(AccountStateChange {
+            message: None,
+            account_id: account_id.clone(),
+            expected_revision: revision,
+            availability: AccountAvailability::Ready,
+            observed_at: SystemTime::now(),
+        })
+        .await
+        .expect("recover account");
+    let cleared: Option<String> =
+        sqlx::query_scalar("select last_error_message from provider_accounts where id = $1")
+            .bind(account_id.as_str())
+            .fetch_one(&database.pool)
+            .await
+            .expect("read cleared error message");
+    assert_eq!(cleared, None);
+
+    database.close().await;
+}
+
 #[tokio::test]
 async fn terminal_admin_list_filters_and_sorts_before_pagination_with_retained_usage() {
     let Some(database) = TestDatabase::create("provider_account_terminal_list").await else {
@@ -231,6 +288,21 @@ async fn terminal_admin_list_filters_and_sorts_before_pagination_with_retained_u
             .await
             .expect("insert account list fixture");
     }
+    repository
+        .compare_and_swap_quota(QuotaObservation {
+            account_id: ProviderAccountId::new("acct_charlie").expect("account id"),
+            expected_revision: CredentialRevision::new(1).expect("credential revision"),
+            quota: Some(OpaqueProviderData::new(
+                json!({"rate_limit": {"limit_reached": true}})
+                    .as_object()
+                    .expect("quota object")
+                    .clone(),
+            )),
+            observed_at: Some(SystemTime::now()),
+            limit_reached: Some(true),
+        })
+        .await
+        .expect("seed rate-limited account");
 
     seed_model_request(
         &database.pool,
@@ -289,7 +361,7 @@ async fn terminal_admin_list_filters_and_sorts_before_pagination_with_retained_u
     .await
     .expect("seed charlie usage");
 
-    let store = PgAdminAccountStore::new(database.pool.clone());
+    let store = PgAdminAccountStore::new(database.pool.clone(), None);
     let usage_page = store
         .list_accounts(AccountListQuery {
             page: 1,
@@ -307,14 +379,18 @@ async fn terminal_admin_list_filters_and_sorts_before_pagination_with_retained_u
     assert_eq!(usage_page.config_revision.get(), 1);
     assert_eq!(usage_page.total, 6);
     assert_eq!(usage_page.summary.total, 6);
-    assert_eq!(usage_page.summary.active, 2);
-    assert_eq!(usage_page.summary.quota_exhausted, 1);
-    assert_eq!(usage_page.summary.unavailable, 3);
+    assert_eq!(usage_page.summary.normal, 1);
+    assert_eq!(usage_page.summary.quota_exhausted, 2);
+    assert_eq!(usage_page.summary.rate_limited, 0);
+    assert_eq!(usage_page.summary.disabled, 1);
+    assert_eq!(usage_page.summary.error, 2);
     assert_eq!(
         usage_page.summary.total,
-        usage_page.summary.active
+        usage_page.summary.normal
             + usage_page.summary.quota_exhausted
-            + usage_page.summary.unavailable
+            + usage_page.summary.rate_limited
+            + usage_page.summary.disabled
+            + usage_page.summary.error
     );
     assert_eq!(
         usage_page
@@ -354,7 +430,7 @@ async fn terminal_admin_list_filters_and_sorts_before_pagination_with_retained_u
             page_size: PageSize::new(10).expect("page size"),
             provider_kind: Some(ProviderKind::new("openai").expect("Provider kind")),
             search: Some("ALPHA@EXAMPLE".to_owned()),
-            status: Some(AccountStatus::Active),
+            status: Some(AccountStatus::Normal),
             sort: None,
         })
         .await
@@ -364,18 +440,20 @@ async fn terminal_admin_list_filters_and_sorts_before_pagination_with_retained_u
     assert_eq!(filtered.items[0].id, "acct_alpha");
     assert_eq!(filtered.items[0].provider_kind.as_str(), "openai");
 
-    let banned = store
+    let error_accounts = store
         .list_accounts(AccountListQuery {
             page: 1,
             page_size: PageSize::new(10).expect("page size"),
             provider_kind: None,
             search: None,
-            status: Some(AccountStatus::Banned),
+            status: Some(AccountStatus::Error),
             sort: None,
         })
         .await
-        .expect("filter banned accounts");
-    assert_eq!(banned.items[0].id, "acct_beta");
+        .expect("filter error accounts");
+    assert_eq!(error_accounts.items.len(), 2);
+    assert_eq!(error_accounts.items[0].id, "acct_beta");
+    assert_eq!(error_accounts.items[1].id, "acct_invalid");
     database.close().await;
 }
 
@@ -403,7 +481,7 @@ async fn terminal_credential_list_preserves_grouped_filters_and_unpaged_collecti
             .expect("insert xAI credential fixture");
     }
 
-    let store = PgAdminAccountStore::new(database.pool.clone());
+    let store = PgAdminAccountStore::new(database.pool.clone(), None);
     let provider = ProviderKind::new("xai").expect("xAI Provider kind");
     let complete = store
         .list_credentials(
@@ -484,7 +562,7 @@ async fn terminal_admin_usage_chunks_large_selections_and_preserves_exact_costs(
     let mut account_ids = vec!["acct_usage_exact".to_owned()];
     account_ids.extend((0..200).map(|index| format!("missing_account_{index}")));
 
-    let usage = PgAdminAccountStore::new(database.pool.clone())
+    let usage = PgAdminAccountStore::new(database.pool.clone(), None)
         .load_account_usage(
             TimeRange {
                 start: now - TimeDelta::hours(1),
@@ -576,7 +654,7 @@ async fn terminal_admin_quota_window_usage_prefers_provider_total_and_falls_back
     .await
     .expect("separate provider total from fallback components");
 
-    let mut usage = PgAdminAccountStore::new(database.pool.clone())
+    let mut usage = PgAdminAccountStore::new(database.pool.clone(), None)
         .load_account_usage_by_windows(&[
             AccountUsageWindowQuery {
                 account_id: "acct_quota_window".to_owned(),
@@ -616,7 +694,7 @@ async fn terminal_admin_mutations_keep_revision_account_and_audit_atomic() {
         .insert_provider_account(account("acct_terminal_mutation", "user-terminal-mutation"))
         .await
         .expect("insert mutation account");
-    let store = PgAdminAccountStore::new(database.pool.clone());
+    let store = PgAdminAccountStore::new(database.pool.clone(), None);
     let context = MutationContext {
         actor: MutationActor::System,
         request_id: "request_terminal_mutation".to_owned(),
@@ -689,7 +767,7 @@ async fn terminal_admin_delete_removes_enabled_accounts_in_one_transaction() {
             .expect("insert enabled account");
     }
 
-    let revision = PgAdminAccountStore::new(database.pool.clone())
+    let revision = PgAdminAccountStore::new(database.pool.clone(), None)
         .delete_accounts(
             DeleteAccounts {
                 account_ids: vec![
@@ -1193,6 +1271,7 @@ async fn xai_free_quota_exhaustion_state_is_persisted() {
 
     repository
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account_id.clone(),
             expected_revision: CredentialRevision::new(1).expect("credential revision"),
             availability: AccountAvailability::QuotaExhausted,
@@ -1231,6 +1310,7 @@ async fn xai_resettable_usage_limit_state_is_persisted() {
     let observed_at = SystemTime::now();
     repository
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account_id.clone(),
             expected_revision: CredentialRevision::new(1).expect("credential revision"),
             availability: AccountAvailability::Ready,
@@ -1280,6 +1360,7 @@ async fn disabled_account_preserves_user_state_during_refresh_writes() {
 
     repository
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account_id.clone(),
             expected_revision: CredentialRevision::new(1).expect("credential revision"),
             availability: AccountAvailability::Ready,

backend/crates/gateway-store/tests/postgres/snapshots.rs

@@ -39,7 +39,7 @@ async fn request_snapshots_should_survive_account_deletion() {
         .await
         .expect("delete provider account");
 
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
     let page = repository
         .list_usage_records(usage_query(started_at, UsageRecordFilter::default()))
         .await
@@ -154,7 +154,7 @@ async fn attempts_should_keep_their_own_account_snapshots() {
         .await
         .expect("delete provider accounts");
 
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
     let detail = repository
         .usage_record_detail("req_snap_ab")
         .await
@@ -276,7 +276,7 @@ async fn ops_errors_should_keep_request_and_event_snapshots_after_account_deleti
         .await
         .expect("delete provider accounts");
 
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
     let errors = repository
         .list_ops_errors(OpsErrorQuery {
             range: range_around(started_at),
@@ -376,7 +376,7 @@ async fn diagnostics_should_group_same_email_accounts_by_stable_ref() {
         .await
         .expect("delete provider accounts");
 
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
     let diagnostics = repository
         .usage_diagnostics(
             range_around(started_at),
@@ -445,7 +445,7 @@ async fn diagnostics_should_fallback_to_name_then_ref_for_missing_snapshots() {
         .expect("insert legacy request without live account");
     finalize_request(&database.pool, "req_snap_legacy", started_at).await;
 
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
     let diagnostics = repository
         .usage_diagnostics(
             range_around(started_at),
@@ -511,7 +511,7 @@ async fn failure_diagnostics_should_only_include_errored_requests() {
     .await
     .expect("mark request failed");
 
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
     let diagnostics = repository
         .usage_diagnostics(
             range_around(started_at),
@@ -566,7 +566,7 @@ async fn api_key_diagnostics_should_display_key_name_and_fallback_to_ref() {
         .await
         .expect("set request b latency");
 
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
     let diagnostics = repository
         .usage_diagnostics(
             range_around(started_at),
@@ -633,7 +633,7 @@ async fn renaming_account_should_not_rewrite_historical_snapshots() {
     .await
     .expect("rename provider account");
 
-    let repository = PgObservabilityRepository::new(database.pool.clone());
+    let repository = PgObservabilityRepository::new(database.pool.clone(), None);
     let page = repository
         .list_usage_records(usage_query(started_at, UsageRecordFilter::default()))
         .await

backend/crates/gateway-store/tests/s3.rs

@@ -0,0 +1,141 @@
+use std::io;
+
+use chrono::Utc;
+use gateway_admin::model::backup::{BackupObjectMetadata, BackupStorageConfig};
+use gateway_admin::ports::backup::{BackupObjectStorePort, UploadObjectRequest};
+use gateway_core::engine::CancellationToken;
+use gateway_store::backup::s3::S3ObjectStoreAdapter;
+use secrecy::SecretString;
+use tokio::io::{AsyncReadExt as _, AsyncWriteExt as _};
+use tokio::net::{TcpListener, TcpStream};
+
+const PART_SIZE_ABOVE_TOKIO_DEFAULT: usize = 2 * 1024 * 1024 + 1;
+
+#[tokio::test]
+async fn upload_file_sends_full_part_beyond_tokio_default_buffer_limit() {
+    let directory = tempfile::tempdir().expect("create temporary directory");
+    let source = directory.path().join("archive.dump");
+    let expected = vec![b'x'; PART_SIZE_ABOVE_TOKIO_DEFAULT];
+    tokio::fs::write(&source, &expected)
+        .await
+        .expect("write temporary archive");
+
+    let listener = TcpListener::bind("127.0.0.1:0")
+        .await
+        .expect("bind multipart probe server");
+    let endpoint = format!(
+        "http://{}",
+        listener.local_addr().expect("read listener address")
+    );
+    let server = tokio::spawn(serve_multipart_probe(listener));
+    let storage = BackupStorageConfig {
+        storage_revision: 1,
+        endpoint,
+        region: "auto".to_owned(),
+        bucket: "backup-bucket".to_owned(),
+        access_key_id: "test-access-key".to_owned(),
+        secret_access_key: SecretString::from("test-secret-key"),
+        prefix: "backups".to_owned(),
+        force_path_style: true,
+    };
+    let request = UploadObjectRequest {
+        object_key: "backups/archive.dump".to_owned(),
+        source,
+        metadata: BackupObjectMetadata::new(
+            "backup_test".to_owned(),
+            "a".repeat(64),
+            Utc::now(),
+            PART_SIZE_ABOVE_TOKIO_DEFAULT as u64,
+        ),
+        cancellation: CancellationToken::new(),
+    };
+
+    S3ObjectStoreAdapter::new()
+        .upload_file(&storage, request)
+        .await
+        .expect("upload multipart archive");
+    let uploaded_len = server
+        .await
+        .expect("multipart probe task did not panic")
+        .expect("serve multipart probe");
+
+    assert_eq!(uploaded_len, PART_SIZE_ABOVE_TOKIO_DEFAULT);
+}
+
+async fn serve_multipart_probe(listener: TcpListener) -> io::Result<usize> {
+    let mut uploaded_len = None;
+    for _ in 0..3 {
+        let (mut stream, _) = listener.accept().await?;
+        let (request, body) = read_http_request(&mut stream).await?;
+        let response = if request.starts_with("POST ") && request.contains("uploads") {
+            xml_response(
+                "<InitiateMultipartUploadResult xmlns=\"http://s3.amazonaws.com/doc/2006-03-01/\"><UploadId>probe-upload</UploadId></InitiateMultipartUploadResult>",
+            )
+        } else if request.starts_with("PUT ") && request.contains("partNumber=") {
+            uploaded_len = Some(body.len());
+            "HTTP/1.1 200 OK\r\ncontent-length: 0\r\nconnection: close\r\netag: \"probe-part\"\r\n\r\n".to_owned()
+        } else if request.starts_with("POST ") && request.contains("uploadId=") {
+            xml_response(
+                "<CompleteMultipartUploadResult xmlns=\"http://s3.amazonaws.com/doc/2006-03-01/\"><ETag>\"probe-part\"</ETag></CompleteMultipartUploadResult>",
+            )
+        } else {
+            return Err(io::Error::other("unexpected multipart probe request"));
+        };
+        stream.write_all(response.as_bytes()).await?;
+        stream.shutdown().await?;
+    }
+    uploaded_len.ok_or_else(|| io::Error::other("multipart probe did not receive an upload part"))
+}
+
+async fn read_http_request(stream: &mut TcpStream) -> io::Result<(String, Vec<u8>)> {
+    let mut bytes = Vec::new();
+    let header_end = loop {
+        if let Some(position) = bytes.windows(4).position(|window| window == b"\r\n\r\n") {
+            break position + 4;
+        }
+        let mut chunk = [0_u8; 8 * 1024];
+        let read = stream.read(&mut chunk).await?;
+        if read == 0 {
+            return Err(io::Error::new(
+                io::ErrorKind::UnexpectedEof,
+                "multipart probe request ended before headers",
+            ));
+        }
+        bytes.extend_from_slice(&chunk[..read]);
+    };
+    let headers = String::from_utf8_lossy(&bytes[..header_end]).into_owned();
+    let content_len = headers
+        .lines()
+        .filter_map(|line| line.split_once(':'))
+        .find(|(name, _)| name.eq_ignore_ascii_case("content-length"))
+        .and_then(|(_, value)| value.trim().parse::<usize>().ok())
+        .unwrap_or_default();
+    if headers
+        .to_ascii_lowercase()
+        .contains("expect: 100-continue")
+    {
+        stream.write_all(b"HTTP/1.1 100 Continue\r\n\r\n").await?;
+    }
+    while bytes.len() < header_end + content_len {
+        let mut chunk = [0_u8; 8 * 1024];
+        let read = stream.read(&mut chunk).await?;
+        if read == 0 {
+            return Err(io::Error::new(
+                io::ErrorKind::UnexpectedEof,
+                "multipart probe request ended before body",
+            ));
+        }
+        bytes.extend_from_slice(&chunk[..read]);
+    }
+    Ok((
+        headers,
+        bytes[header_end..header_end + content_len].to_vec(),
+    ))
+}
+
+fn xml_response(body: &str) -> String {
+    format!(
+        "HTTP/1.1 200 OK\r\ncontent-type: application/xml\r\ncontent-length: {}\r\nconnection: close\r\n\r\n{body}",
+        body.len()
+    )
+}

backend/crates/providers/openai/Cargo.toml

@@ -58,6 +58,7 @@ zstd = "0.13"
 [dev-dependencies]
 tempfile = "3.27.0"
 tokio = { version = "1.52.3", default-features = false, features = ["rt-multi-thread", "test-util"] }
+tracing-subscriber = { version = "0.3.23", features = ["json"] }
 wiremock = "0.6.5"
 
 [[test]]

backend/crates/providers/openai/src/admin.rs

@@ -50,7 +50,7 @@ use crate::credential::{
     CodexCredentialAdminService, CodexCredentialCatalogError, CodexCredentialCatalogService,
     CodexCredentialQuotaError, CodexCredentialQuotaService, CodexOAuthAdmin, CodexOAuthAdminError,
     CodexOAuthPendingClaimOutcome, CodexOAuthPendingStore, CodexOAuthPendingStoreError,
-    CodexPendingAuthorization, CodexQuotaWindowKind, CodexQuotaWindowRole,
+    CodexPendingAuthorization, CodexQuotaWindow, CodexQuotaWindowKind, CodexQuotaWindowRole,
     CompleteCodexOAuthAuthorization, CompletedCodexOAuthCredential, ExportManagedCodexCredential,
     RotateManagedCodexCredential, StartCodexOAuthAuthorization, StoredCodexPendingAuthorization,
     refresh_time,
@@ -194,12 +194,13 @@ impl ProviderAdmin for OpenAiAdminProvider {
         let profile = self.profile.snapshot();
         let release = self.desktop_release.snapshot();
         let user_agent = profile.user_agent();
+        let client_identity = format!("{}; {}", profile.originator, profile.desktop_version);
         let release = dashboard_desktop_release(&profile, release);
         Some(DashboardWireProfile {
             provider: self.provider_kind.as_str().to_owned(),
             product: profile.originator,
-            version: profile.desktop_version,
-            build: Some(profile.desktop_build),
+            version: profile.codex_version,
+            build: None,
             target: DashboardWireTarget {
                 os_type: profile.os_type,
                 os_version: profile.os_version,
@@ -208,8 +209,8 @@ impl ProviderAdmin for OpenAiAdminProvider {
             },
             user_agent,
             attributes: vec![DashboardWireAttribute {
-                label: "Codex Core".to_owned(),
-                value: profile.codex_version,
+                label: "客户端标识".to_owned(),
+                value: client_identity,
             }],
             verified_at: Some(profile.verified_at),
             release: Some(release),
@@ -475,7 +476,14 @@ impl ProviderAdmin for OpenAiAdminProvider {
                 .await
                 .map_err(map_quota_error)?
         };
-        Ok(project_quota(snapshot, &account))
+        let mut quota = project_quota(snapshot, &account);
+        quota.rate_limited_until = self
+            .quota
+            .rate_limited_until(account.id(), account.revision())
+            .await
+            .map_err(map_quota_error)?
+            .map(DateTime::<Utc>::from);
+        Ok(quota)
     }
 
     async fn models(
@@ -740,6 +748,7 @@ fn empty_quota() -> ProviderQuota {
         refresh_token_expires_at: None,
         windows: Vec::new(),
         limit_reached: false,
+        rate_limited_until: None,
         provider_data: None,
     }
 }
@@ -774,6 +783,7 @@ fn project_quota_snapshot(snapshot: CodexAccountQuotaSnapshot) -> ProviderQuota
     let windows: Vec<ProviderQuotaWindow> = snapshot
         .windows()
         .iter()
+        .filter(|window| should_project_quota_window(window))
         .map(|window| {
             let mut data = Map::new();
             data.insert(
@@ -790,6 +800,7 @@ fn project_quota_snapshot(snapshot: CodexAccountQuotaSnapshot) -> ProviderQuota
                 label: codex_quota_window_label(
                     window.limit_name(),
                     window.kind(),
+                    window.role(),
                     window.source(),
                     window.window_seconds(),
                 ),
@@ -811,12 +822,24 @@ fn project_quota_snapshot(snapshot: CodexAccountQuotaSnapshot) -> ProviderQuota
         refresh_token_expires_at: None,
         windows,
         limit_reached,
+        rate_limited_until: None,
         provider_data: Some(ProviderDocument::new(OpaqueProviderData::new(
             provider_data,
         ))),
     }
 }
 
+/// `secondary_window` 有时只是上游的空占位：没有时长、重置时间，也没有用量。
+/// 它不能被可靠翻译成 5 小时或周额度，因此不应占用账号面板；带有实际事实的
+/// 次级窗口（时长、重置、非零用量或触顶）仍完整保留。
+fn should_project_quota_window(window: &CodexQuotaWindow) -> bool {
+    window.role() != CodexQuotaWindowRole::Secondary
+        || window.window_seconds().is_some()
+        || window.reset_at().is_some()
+        || window.used_percent().is_some_and(|used| used > 0.0)
+        || window.limit_reached()
+}
+
 fn quota_windows_limit_reached(windows: &[ProviderQuotaWindow]) -> bool {
     windows.iter().any(|window| window.limit_reached)
 }
@@ -882,45 +905,66 @@ const fn quota_role(role: CodexQuotaWindowRole) -> &'static str {
     }
 }
 
-/// 窗口展示名，优先使用上游 `limit_name`（原文，`_` 转空格）；
-/// 无 limit_name 时按窗口时长猜（±5% 容差，汉化输出），猜不中兜底「额度」。
+/// 按窗口时长显示汉化额度名；非核心桶补上上游桶名称作为前缀。
 fn codex_quota_window_label(
     limit_name: Option<&str>,
     kind: CodexQuotaWindowKind,
+    role: CodexQuotaWindowRole,
     source: &str,
     window_seconds: Option<u64>,
 ) -> String {
-    if let Some(limit_name) = limit_name {
-        return limit_name.replace('_', " ");
-    }
     let base = match kind {
-        CodexQuotaWindowKind::Monthly => "月限额".to_owned(),
-        CodexQuotaWindowKind::Weekly => "周限额".to_owned(),
+        CodexQuotaWindowKind::Monthly => "月额度".to_owned(),
+        CodexQuotaWindowKind::Weekly => "周额度".to_owned(),
         CodexQuotaWindowKind::ShortTerm => {
             if window_seconds.is_some_and(|seconds| seconds > 86_400) {
-                "周限额".to_owned()
+                "周额度".to_owned()
             } else {
-                "5小时限额".to_owned()
+                "5小时额度".to_owned()
             }
         }
-        CodexQuotaWindowKind::Other => custom_quota_window_label(window_seconds),
+        CodexQuotaWindowKind::Other => custom_quota_window_label(window_seconds, role),
     };
-    if matches!(source, "core" | "codex" | "code_review" | "spend_control") {
-        return base;
+    if let Some(prefix) = codex_quota_window_prefix(limit_name, source) {
+        return format!("{prefix} · {base}");
     }
-    format!("{source} · {base}")
+    base
 }
 
-fn custom_quota_window_label(window_seconds: Option<u64>) -> String {
+fn codex_quota_window_prefix(limit_name: Option<&str>, source: &str) -> Option<String> {
+    let source_identifier = quota_identifier(source);
+    if matches!(source_identifier.as_str(), "core" | "codex") {
+        return None;
+    }
+    let label = limit_name.unwrap_or(source);
+    match quota_identifier(label).as_str() {
+        "premium" => Some("Premium".to_owned()),
+        "code_review" => Some("代码审查".to_owned()),
+        "spend_control" => Some("消费控制".to_owned()),
+        _ => Some(label.replace('_', " ")),
+    }
+}
+
+fn quota_identifier(value: &str) -> String {
+    value.trim().to_ascii_lowercase().replace(['-', ' '], "_")
+}
+
+fn custom_quota_window_label(window_seconds: Option<u64>, role: CodexQuotaWindowRole) -> String {
     let Some(seconds) = window_seconds.filter(|seconds| *seconds > 0) else {
-        return "额度".to_owned();
+        // 与官方客户端一致：没有时长时不臆测为 5 小时或周额度；保留
+        // primary/secondary 语义，让用户知道这是上游未标明时长的独立窗口。
+        return match role {
+            CodexQuotaWindowRole::Primary => "主额度".to_owned(),
+            CodexQuotaWindowRole::Secondary => "次级额度".to_owned(),
+            CodexQuotaWindowRole::Monthly => "月额度".to_owned(),
+        };
     };
     if seconds % 86_400 == 0 {
-        format!("{}日限额", seconds / 86_400)
+        format!("{}日额度", seconds / 86_400)
     } else if seconds % 3_600 == 0 {
-        format!("{}小时限额", seconds / 3_600)
+        format!("{}小时额度", seconds / 3_600)
     } else {
-        format!("{}分钟限额", seconds.div_ceil(60))
+        format!("{}分钟额度", seconds.div_ceil(60))
     }
 }
 
@@ -1337,14 +1381,15 @@ fn map_credential_admin_error(error: CodexCredentialAdminError) -> ProviderAdmin
         | Error::MissingRefreshToken
         | Error::RefreshRejected
         | Error::AccountBanned
+        | Error::AccountProfileRejected
         | Error::IdentityRejected => ProviderAdminErrorKind::Invalid,
         Error::NotFound => ProviderAdminErrorKind::NotFound,
         Error::RefreshLeaseUnavailable | Error::RefreshAmbiguous => {
             ProviderAdminErrorKind::Conflict
         }
-        Error::RefreshUnavailable | Error::IdentityUnavailable => {
-            ProviderAdminErrorKind::Unavailable
-        }
+        Error::RefreshUnavailable
+        | Error::AccountProfileUnavailable
+        | Error::IdentityUnavailable => ProviderAdminErrorKind::Unavailable,
     })
 }
 
@@ -1358,6 +1403,8 @@ const fn credential_admin_error_code(error: CodexCredentialAdminError) -> &'stat
         CodexCredentialAdminError::RefreshLeaseUnavailable => "refresh_lease_unavailable",
         CodexCredentialAdminError::RefreshRejected => "refresh_rejected",
         CodexCredentialAdminError::AccountBanned => "account_banned",
+        CodexCredentialAdminError::AccountProfileRejected => "account_profile_rejected",
+        CodexCredentialAdminError::AccountProfileUnavailable => "account_profile_unavailable",
         CodexCredentialAdminError::RefreshUnavailable => "refresh_unavailable",
         CodexCredentialAdminError::RefreshAmbiguous => "refresh_ambiguous",
         CodexCredentialAdminError::IdentityRejected => "identity_rejected",

backend/crates/providers/openai/src/config.rs

@@ -43,10 +43,8 @@ impl OpenAiConfig {
         self.quota.validate()?;
         self.auth.validate()?;
         self.wire_profile.validate()?;
-        self.identity_secret_path = source_dir
-            .parent()
-            .unwrap_or(source_dir)
-            .join(".runtime/data/identity_hmac_secret");
+        let project_root = source_dir.parent().unwrap_or(source_dir);
+        self.identity_secret_path = project_root.join(".runtime/data/identity_hmac_secret");
         Ok(())
     }
 
@@ -278,6 +276,7 @@ impl CodexAuthSettings {
 #[serde(deny_unknown_fields)]
 pub struct CodexWireProfileConfig {
     pub originator: String,
+    /// npm `@openai/codex` 启动版本；运行时发布任务会更新此画像字段。
     pub codex_version: String,
     pub desktop_version: String,
     pub desktop_build: String,
@@ -292,10 +291,10 @@ impl Default for CodexWireProfileConfig {
     fn default() -> Self {
         Self {
             originator: "Codex Desktop".to_owned(),
-            codex_version: "0.145.0".to_owned(),
-            desktop_version: "26.721.31836".to_owned(),
-            desktop_build: "5828".to_owned(),
-            os_type: "macOS".to_owned(),
+            codex_version: "0.146.0".to_owned(),
+            desktop_version: "26.727.51351".to_owned(),
+            desktop_build: "6119".to_owned(),
+            os_type: "Mac OS".to_owned(),
             os_version: "15.7.1".to_owned(),
             arch: "arm64".to_owned(),
             terminal: "unknown".to_owned(),

backend/crates/providers/openai/src/credential/admin.rs

@@ -26,8 +26,9 @@ use serde_json::Value;
 use thiserror::Error;
 
 use super::identity::{
-    CodexAccountIdentityVerifier, CodexIdentityExpectation, CodexIdentityVerificationError,
+    CodexAuthenticatedAccountSource, CodexIdentityExpectation, CodexIdentityVerificationError,
 };
+use super::recovery_log::{CodexOAuthRecoveryOperation, record_oauth_recovery};
 use super::security::CodexCredentialCodec;
 use super::token_client::{RefreshFailure, TokenRefresher};
 use super::types::{
@@ -40,6 +41,8 @@ const PROVIDER_NAME: &str = "openai";
 const CODEX_CREDENTIAL_SCHEMA_VERSION: u32 = 1;
 const MAX_BATCH: usize = 200;
 const MAX_IMPORT_DOCUMENT_BYTES: usize = 64 * 1024 * 1024;
+const MAX_IMPORT_EMAIL_BYTES: usize = 512;
+const MAX_IMPORT_PLAN_BYTES: usize = 128;
 
 pub struct ImportCodexOAuthCredential {
     pub account_id: String,
@@ -365,6 +368,10 @@ pub enum CodexCredentialAdminError {
     RefreshUnavailable,
     #[error("Codex refresh send state is ambiguous")]
     RefreshAmbiguous,
+    #[error("Codex account profile request was rejected")]
+    AccountProfileRejected,
+    #[error("Codex account profile request is unavailable")]
+    AccountProfileUnavailable,
     #[error("Codex refreshed identity was rejected")]
     IdentityRejected,
     #[error("Codex identity verification is unavailable")]
@@ -387,8 +394,7 @@ impl CodexCredentialAdmin {
         if input.name.trim().is_empty() {
             return Err(CodexCredentialAdminError::InvalidInput);
         }
-        let access_token_expires_at =
-            required_time(input.verified_account.access_token_expires_at)?;
+        let access_token_expires_at = optional_time(input.verified_account.access_token_expires_at);
         let revision =
             CredentialRevision::new(1).map_err(|_| CodexCredentialAdminError::InvalidCredential)?;
         let upstream_user_id = input.verified_account.chatgpt_user_id.clone();
@@ -402,7 +408,7 @@ impl CodexCredentialAdmin {
             upstream_user_id,
             CODEX_AUTHENTICATION_KIND_OAUTH.to_owned(),
             revision,
-            Some(access_token_expires_at),
+            access_token_expires_at,
         )
         .with_profile(
             input.verified_account.email,
@@ -499,12 +505,60 @@ impl CodexCredentialAdmin {
         self.prepare_oauth_rotation(input, false)
     }
 
-    /// 完整 OAuth 授权按目标数据 ID 替换凭据与上游账号身份。
-    pub(crate) fn prepare_reauthorization(
+    /// 构建一次成功 RT exchange 的 CAS 写入，保留已存账号身份资料。
+    ///
+    /// Refresh endpoint 已经完成本次 token 交换的授权；这里不对新 access
+    /// token 重新执行身份验证。
+    pub(crate) fn prepare_refreshed_oauth_rotation(
         &self,
-        input: RotateManagedCodexCredential,
+        current: LoadedCredential,
+        secret: CodexOAuthSecret,
+        access_token_expires_at: DateTime<Utc>,
+        next_refresh_at: Option<DateTime<Utc>>,
     ) -> Result<PreparedCodexCredentialRotation, CodexCredentialAdminError> {
-        self.prepare_oauth_rotation(input, true)
+        if current.account.provider().as_str() != PROVIDER_NAME
+            || current.account.authentication_kind() != CODEX_AUTHENTICATION_KIND_OAUTH
+        {
+            return Err(CodexCredentialAdminError::IdentityMismatch);
+        }
+        let mut data = CodexCredentialCodec::decode_complete(&current.credential)
+            .map_err(|_| CodexCredentialAdminError::InvalidCredential)?;
+        let oauth = data
+            .oauth_mut()
+            .ok_or(CodexCredentialAdminError::IdentityMismatch)?;
+        oauth.access_token = secret.access_token.expose_secret().to_owned();
+        oauth.refresh_token = secret
+            .refresh_token
+            .as_ref()
+            .map(|value| value.expose_secret().to_owned());
+        oauth.id_token = secret
+            .id_token
+            .as_ref()
+            .map(|value| value.expose_secret().to_owned());
+        let credential = CodexCredentialCodec::encode_complete(data)
+            .map_err(|_| CodexCredentialAdminError::InvalidCredential)?;
+        let profile = ProviderAccountUpdate {
+            account_id: current.account.id().clone(),
+            name: current.account.name().to_owned(),
+            email: current.account.email().map(str::to_owned),
+            plan_type: current.account.plan_type().map(str::to_owned),
+        };
+        let credential = CredentialCasUpdate::new(
+            current.account.id().clone(),
+            current.account.revision(),
+            profile.clone(),
+            credential,
+            secret.refresh_token.is_some(),
+            Some(SystemTime::from(access_token_expires_at)),
+            optional_time(next_refresh_at),
+        )
+        .map_err(|_| CodexCredentialAdminError::InvalidCredential)?;
+        Ok(PreparedCodexCredentialRotation {
+            profile,
+            credential,
+            replacement_identity: None,
+            refresh_guards: None,
+        })
     }
 
     fn prepare_oauth_rotation(
@@ -588,7 +642,7 @@ impl CodexCredentialAdmin {
 /// 有状态的 Codex 手工刷新边界；消费调用方刚读取的当前 credential 并准备 CAS。
 pub struct CodexCredentialAdminService {
     refresher: Arc<dyn TokenRefresher>,
-    verifier: Arc<dyn CodexAccountIdentityVerifier>,
+    account_source: Arc<dyn CodexAuthenticatedAccountSource>,
     leases: Arc<dyn ProviderLeasePort>,
     runtime_policy: Arc<dyn ProviderRuntimePolicyPort>,
 }
@@ -598,7 +652,7 @@ impl fmt::Debug for CodexCredentialAdminService {
         formatter
             .debug_struct("CodexCredentialAdminService")
             .field("refresher", &"TokenRefresher")
-            .field("verifier", &"CodexAccountIdentityVerifier")
+            .field("account_source", &"CodexAuthenticatedAccountSource")
             .field("leases", &"ProviderLeasePort")
             .field("runtime_policy", &"ProviderRuntimePolicyPort")
             .finish()
@@ -608,19 +662,19 @@ impl fmt::Debug for CodexCredentialAdminService {
 impl CodexCredentialAdminService {
     pub fn new(
         refresher: Arc<dyn TokenRefresher>,
-        verifier: Arc<dyn CodexAccountIdentityVerifier>,
+        account_source: Arc<dyn CodexAuthenticatedAccountSource>,
         leases: Arc<dyn ProviderLeasePort>,
         runtime_policy: Arc<dyn ProviderRuntimePolicyPort>,
     ) -> Self {
         Self {
             refresher,
-            verifier,
+            account_source,
             leases,
             runtime_policy,
         }
     }
 
-    /// 官方 RT exchange + AT 身份验证；结果由 App 在同一 revision/audit 事务中提交。
+    /// 官方 RT exchange；结果由 App 在同一 revision/audit 事务中提交。
     pub async fn manual_refresh(
         &self,
         current: LoadedCredential,
@@ -632,7 +686,6 @@ impl CodexCredentialAdminService {
         }
         let runtime = CodexCredentialCodec::decode(&current.credential)
             .map_err(|_| CodexCredentialAdminError::InvalidCredential)?;
-        let expectation = current_identity_expectation(&current.account, &runtime)?;
         let oauth = runtime
             .authentication
             .oauth()
@@ -677,9 +730,6 @@ impl CodexCredentialAdminService {
             .refresh(refresh_token.expose_secret())
             .await
             .map_err(map_refresh_failure)?;
-        if tokens.access_token.trim().is_empty() || tokens.expires_in.is_zero() {
-            return Err(CodexCredentialAdminError::InvalidCredential);
-        }
         let secret = CodexOAuthSecret {
             access_token: SecretString::from(tokens.access_token),
             refresh_token: tokens
@@ -688,24 +738,30 @@ impl CodexCredentialAdminService {
                 .or_else(|| oauth.refresh_token.clone()),
             id_token: oauth.id_token.clone(),
         };
-        let verified_account = self
-            .verifier
-            .verify(&secret, &expectation)
-            .await
-            .and_then(super::identity::CodexIdentityVerification::into_complete)
-            .map_err(map_identity_error)?;
+        Self::record_recovery_log(
+            CodexOAuthRecoveryOperation::ManualRefresh,
+            Some(account_id.as_str()),
+            &secret,
+        );
+        if secret.access_token.expose_secret().trim().is_empty() || tokens.expires_in.is_zero() {
+            return Err(CodexCredentialAdminError::InvalidCredential);
+        }
+        let access_token_expires_at = SystemTime::now()
+            .checked_add(tokens.expires_in)
+            .map(DateTime::<Utc>::from)
+            .ok_or(CodexCredentialAdminError::InvalidCredential)?;
         let next_refresh_at = refresh_time(
             policy,
             &account_id,
-            verified_account.access_token_expires_at,
+            Some(access_token_expires_at),
             secret.refresh_token.is_some(),
         )?;
-        let mut prepared = CodexCredentialAdmin.prepare_rotation(RotateManagedCodexCredential {
+        let mut prepared = CodexCredentialAdmin.prepare_refreshed_oauth_rotation(
             current,
             secret,
-            verified_account,
+            access_token_expires_at,
             next_refresh_at,
-        })?;
+        )?;
         prepared.refresh_guards = Some(ProviderRefreshGuards {
             _capacity: capacity_guard,
             _account: account_guard,
@@ -713,7 +769,7 @@ impl CodexCredentialAdminService {
         Ok(prepared)
     }
 
-    /// 从 OAuth token 对补全账号事实，并归一到唯一 `NewProviderAccount` 写入路径。
+    /// 用 access token 补全账号事实，并归一到唯一 `NewProviderAccount` 写入路径。
     pub async fn prepare_import_document(
         &self,
         payload: Value,
@@ -743,9 +799,23 @@ impl CodexCredentialAdminService {
                     if !account_ids.insert(account_id.clone()) {
                         return Err(CodexCredentialAdminError::InvalidInput);
                     }
-                    let (secret, verified_account) = self
-                        .verify_import_tokens(access_token.clone(), refresh_token.clone())
+                    let typed_account_id = ProviderAccountId::new(account_id.clone())
+                        .map_err(|_| CodexCredentialAdminError::InvalidInput)?;
+                    let (secret, access_token_expires_at) = self
+                        .resolve_import_tokens(
+                            &typed_account_id,
+                            access_token.clone(),
+                            refresh_token.clone(),
+                        )
                         .await?;
+                    let mut verified_account = complete_oauth_account_profile(
+                        self.account_source.as_ref(),
+                        &secret,
+                        access_token_expires_at,
+                    )
+                    .await
+                    .map_err(map_account_profile_error)?;
+                    apply_import_metadata(&mut verified_account, &candidate);
                     if !upstream_identities.insert((
                         verified_account.chatgpt_user_id.clone(),
                         verified_account.chatgpt_account_id.clone(),
@@ -765,16 +835,17 @@ impl CodexCredentialAdminService {
                     };
                     let next_refresh_at = refresh_time(
                         refresh_policy,
-                        &ProviderAccountId::new(account_id.clone())
-                            .map_err(|_| CodexCredentialAdminError::InvalidInput)?,
+                        &typed_account_id,
                         verified_account.access_token_expires_at,
                         secret.refresh_token.is_some(),
                     )?;
                     CodexCredentialAdmin.prepare_import(ImportCodexOAuthCredential {
                         account_id,
-                        name: verified_account
-                            .email
+                        name: candidate
+                            .name
                             .clone()
+                            .or_else(|| candidate.email.clone())
+                            .or_else(|| verified_account.email.clone())
                             .filter(|name| !name.trim().is_empty())
                             .unwrap_or_else(|| "Codex OAuth".to_owned()),
                         secret,
@@ -815,11 +886,12 @@ impl CodexCredentialAdminService {
         Ok(PreparedCodexAccountImport { accounts })
     }
 
-    async fn verify_import_tokens(
+    async fn resolve_import_tokens(
         &self,
+        account_id: &ProviderAccountId,
         access_token: Option<String>,
         refresh_token: Option<String>,
-    ) -> Result<(CodexOAuthSecret, CodexAccountProfile), CodexCredentialAdminError> {
+    ) -> Result<(CodexOAuthSecret, Option<DateTime<Utc>>), CodexCredentialAdminError> {
         let access_token = access_token
             .as_deref()
             .map(normalize_bearer)
@@ -831,19 +903,15 @@ impl CodexCredentialAdminService {
         if let Some(access_token) = access_token {
             let secret = CodexOAuthSecret {
                 access_token: SecretString::from(access_token),
-                refresh_token: refresh_token.clone().map(SecretString::from),
+                refresh_token: refresh_token.map(SecretString::from),
                 id_token: None,
             };
-            match self.verifier.verify_import(&secret).await {
-                Ok(profile) => return Ok((secret, profile)),
-                Err(CodexIdentityVerificationError::Unavailable) => {
-                    return Err(CodexCredentialAdminError::IdentityUnavailable);
-                }
-                Err(CodexIdentityVerificationError::Rejected) if refresh_token.is_none() => {
-                    return Err(CodexCredentialAdminError::IdentityRejected);
-                }
-                Err(CodexIdentityVerificationError::Rejected) => {}
-            }
+            Self::record_recovery_log(
+                CodexOAuthRecoveryOperation::ImportDirect,
+                Some(account_id.as_str()),
+                &secret,
+            );
+            return Ok((secret, None));
         }
         let refresh_token = refresh_token.ok_or(CodexCredentialAdminError::InvalidCredential)?;
         let tokens = self
@@ -851,9 +919,6 @@ impl CodexCredentialAdminService {
             .refresh(&refresh_token)
             .await
             .map_err(map_refresh_failure)?;
-        if tokens.access_token.trim().is_empty() || tokens.expires_in.is_zero() {
-            return Err(CodexCredentialAdminError::InvalidCredential);
-        }
         let secret = CodexOAuthSecret {
             access_token: SecretString::from(tokens.access_token),
             // RT 未轮换时仍保留导入提供的 RT，保证本次补全不会丢失后续刷新能力。
@@ -863,12 +928,35 @@ impl CodexCredentialAdminService {
                 .or_else(|| Some(SecretString::from(refresh_token))),
             id_token: None,
         };
-        let profile = self
-            .verifier
-            .verify_import(&secret)
-            .await
-            .map_err(map_identity_error)?;
-        Ok((secret, profile))
+        Self::record_recovery_log(
+            CodexOAuthRecoveryOperation::ImportRefreshToken,
+            Some(account_id.as_str()),
+            &secret,
+        );
+        if secret.access_token.expose_secret().trim().is_empty() || tokens.expires_in.is_zero() {
+            return Err(CodexCredentialAdminError::InvalidCredential);
+        }
+        let access_token_expires_at = SystemTime::now()
+            .checked_add(tokens.expires_in)
+            .map(DateTime::<Utc>::from)
+            .ok_or(CodexCredentialAdminError::InvalidCredential)?;
+        Ok((secret, Some(access_token_expires_at)))
+    }
+
+    fn record_recovery_log(
+        operation: CodexOAuthRecoveryOperation,
+        account_id: Option<&str>,
+        secret: &CodexOAuthSecret,
+    ) {
+        record_oauth_recovery(
+            operation,
+            account_id,
+            secret.access_token.expose_secret(),
+            secret
+                .refresh_token
+                .as_ref()
+                .map(ExposeSecret::expose_secret),
+        );
     }
 
     fn prepare_agent_identity_import(
@@ -939,27 +1027,6 @@ impl CodexCredentialAdminService {
     }
 }
 
-fn current_identity_expectation(
-    account: &ProviderAccount,
-    credential: &super::security::CodexRuntimeCredential,
-) -> Result<CodexIdentityExpectation, CodexCredentialAdminError> {
-    let account_id = account
-        .upstream_account_id()
-        .ok_or(CodexCredentialAdminError::InvalidCredential)?;
-    let principal = credential
-        .principal
-        .as_ref()
-        .ok_or(CodexCredentialAdminError::InvalidCredential)?;
-    CodexIdentityExpectation::current(
-        principal.oauth_subject.clone(),
-        principal.poid.clone(),
-        account_id.to_owned(),
-        account.upstream_user_id().to_owned(),
-        credential.installation_id.clone(),
-    )
-    .map_err(|_| CodexCredentialAdminError::InvalidCredential)
-}
-
 fn required_time(
     value: Option<chrono::DateTime<chrono::Utc>>,
 ) -> Result<SystemTime, CodexCredentialAdminError> {
@@ -981,9 +1048,12 @@ pub(crate) fn refresh_time(
     if !has_refresh_token {
         return Ok(None);
     }
-    let expires_at = required_time(access_token_expires_at)?;
+    let observed_at = SystemTime::now();
+    let expires_at = access_token_expires_at
+        .map(SystemTime::from)
+        .unwrap_or(observed_at);
     policy
-        .next_attempt_at(account_id, expires_at, SystemTime::now())
+        .next_attempt_at(account_id, expires_at, observed_at)
         .map(DateTime::<Utc>::from)
         .map(Some)
         .map_err(|_| CodexCredentialAdminError::InvalidCredential)
@@ -1016,15 +1086,6 @@ const fn map_refresh_failure(error: RefreshFailure) -> CodexCredentialAdminError
     }
 }
 
-const fn map_identity_error(error: CodexIdentityVerificationError) -> CodexCredentialAdminError {
-    match error {
-        CodexIdentityVerificationError::Rejected => CodexCredentialAdminError::IdentityRejected,
-        CodexIdentityVerificationError::Unavailable => {
-            CodexCredentialAdminError::IdentityUnavailable
-        }
-    }
-}
-
 fn parse_import_document(
     payload: &Value,
 ) -> Result<Vec<ParsedCodexImportAccount>, CodexCredentialAdminError> {
@@ -1037,20 +1098,28 @@ fn parse_import_document(
         if !is_openai_oauth_candidate(value) {
             continue;
         }
-        accounts.push(ParsedCodexImportAccount {
-            id: None,
-            name: None,
-            email: None,
-            plan_type: None,
-            chatgpt_account_id: None,
-            chatgpt_user_id: None,
-            authentication: parse_oauth_import_tokens(value)?,
-            status: None,
-        });
+        accounts.push(parse_oauth_import_account(value)?);
     }
     Ok(accounts)
 }
 
+fn parse_oauth_import_account(
+    value: &Value,
+) -> Result<ParsedCodexImportAccount, CodexCredentialAdminError> {
+    let credentials = value.get("credentials").unwrap_or(value);
+    Ok(ParsedCodexImportAccount {
+        id: first_string(value, &["id"]),
+        name: first_string(value, &["name", "label"]),
+        email: first_string(value, &["email"]).or_else(|| first_string(credentials, &["email"])),
+        plan_type: first_string(value, &["plan_type", "planType"])
+            .or_else(|| first_string(credentials, &["plan_type", "planType"])),
+        chatgpt_account_id: None,
+        chatgpt_user_id: None,
+        authentication: parse_oauth_import_tokens(value)?,
+        status: first_string(value, &["status"]),
+    })
+}
+
 fn parse_agent_identity_account(
     value: &Value,
 ) -> Result<ParsedCodexImportAccount, CodexCredentialAdminError> {
@@ -1212,6 +1281,59 @@ fn normalize_bearer(value: &str) -> String {
         .to_owned()
 }
 
+pub(crate) async fn complete_oauth_account_profile(
+    account_source: &dyn CodexAuthenticatedAccountSource,
+    secret: &CodexOAuthSecret,
+    access_token_expires_at: Option<DateTime<Utc>>,
+) -> Result<CodexAccountProfile, CodexIdentityVerificationError> {
+    let account = account_source
+        .fetch(secret, &CodexIdentityExpectation::default())
+        .await?;
+    Ok(CodexAccountProfile {
+        email: account.email,
+        oauth_subject: account.chatgpt_user_id.clone(),
+        poid: None,
+        chatgpt_account_id: account.chatgpt_account_id,
+        chatgpt_user_id: account.chatgpt_user_id,
+        plan_type: account.plan_type,
+        access_token_expires_at,
+    })
+}
+
+fn apply_import_metadata(profile: &mut CodexAccountProfile, candidate: &ParsedCodexImportAccount) {
+    if profile.email.is_none() {
+        profile.email = candidate
+            .email
+            .as_deref()
+            .and_then(|value| normalize_import_text(value, MAX_IMPORT_EMAIL_BYTES));
+    }
+    if profile.plan_type.is_none() {
+        profile.plan_type = candidate
+            .plan_type
+            .as_deref()
+            .and_then(|value| normalize_import_text(value, MAX_IMPORT_PLAN_BYTES));
+    }
+}
+
+const fn map_account_profile_error(
+    error: CodexIdentityVerificationError,
+) -> CodexCredentialAdminError {
+    match error {
+        CodexIdentityVerificationError::Rejected => {
+            CodexCredentialAdminError::AccountProfileRejected
+        }
+        CodexIdentityVerificationError::Unavailable => {
+            CodexCredentialAdminError::AccountProfileUnavailable
+        }
+    }
+}
+
+fn normalize_import_text(value: &str, max_bytes: usize) -> Option<String> {
+    let value = value.trim();
+    (!value.is_empty() && value.len() <= max_bytes && !value.chars().any(char::is_control))
+        .then(|| value.to_owned())
+}
+
 fn import_runtime_state(
     status: Option<&str>,
 ) -> Result<(bool, AccountAvailability), CodexCredentialAdminError> {

backend/crates/providers/openai/src/credential/affinity.rs

@@ -10,19 +10,33 @@ use crate::transport::request::derive_conversation_anchor;
 pub(crate) fn derive_codex_session_affinity_key(
     request: &CodexResponsesRequest,
 ) -> Option<ProviderSessionAffinityKey> {
-    if let Some(conversation_id) = request
+    let session_key = if let Some(conversation_id) = request
         .local_conversation_id
         .as_deref()
         .map(str::trim)
         .filter(|value| !value.is_empty())
     {
         if let Ok(key) = ProviderSessionAffinityKey::try_new(conversation_id.to_owned()) {
-            return Some(key);
+            Some(key)
+        } else {
+            opaque_affinity_key("local-conversation", conversation_id)
         }
-        return opaque_affinity_key("local-conversation", conversation_id);
-    }
-    let (domain, value) = derive_conversation_anchor(request)?;
-    opaque_affinity_key(domain, &value)
+    } else {
+        let (domain, value) = derive_conversation_anchor(request)?;
+        opaque_affinity_key(domain, &value)
+    };
+
+    let session_discriminator = request.subagent_kind();
+    session_key.and_then(|session_key| match session_discriminator.as_deref() {
+        None => Some(session_key),
+        Some(discriminator) => {
+            // 仅扩展原会话锚点的键空间；后续查找、绑定和调度仍使用同一套亲和逻辑。
+            opaque_affinity_key(
+                "subagent-session",
+                &format!("{discriminator}\0{}", session_key.expose_to_store()),
+            )
+        }
+    })
 }
 
 fn opaque_affinity_key(domain: &str, value: &str) -> Option<ProviderSessionAffinityKey> {

backend/crates/providers/openai/src/credential/mod.rs

@@ -8,6 +8,7 @@ mod cookie;
 mod identity;
 mod oauth;
 mod quota;
+mod recovery_log;
 mod refresh;
 mod repository;
 mod security;

backend/crates/providers/openai/src/credential/oauth.rs

@@ -1,6 +1,10 @@
 //! Codex Authorization Code + PKCE/OIDC 管理流。
 
-use std::{fmt, sync::Arc, time::Duration};
+use std::{
+    fmt,
+    sync::Arc,
+    time::{Duration, SystemTime},
+};
 
 use async_trait::async_trait;
 use base64::{Engine as _, engine::general_purpose::URL_SAFE_NO_PAD};
@@ -13,7 +17,7 @@ use gateway_admin::model::{
     },
 };
 use gateway_core::engine::credential::{
-    NewProviderAccount, ProviderAccountId, ProviderAccountStore,
+    LoadedCredential, NewProviderAccount, ProviderAccountId, ProviderAccountStore,
 };
 use gateway_core::provider_ports::ProviderRuntimePolicyPort;
 use secrecy::{ExposeSecret, SecretString};
@@ -24,11 +28,11 @@ use uuid::Uuid;
 
 use super::admin::{
     CodexCredentialAdmin, CodexCredentialAdminError, ImportCodexOAuthCredential,
-    PreparedCodexCredentialRotation, RotateManagedCodexCredential, refresh_time,
-};
-use super::identity::{
-    CodexAccountIdentityVerifier, CodexIdentityExpectation, CodexIdentityVerificationError,
+    PreparedCodexCredentialRotation, complete_oauth_account_profile, refresh_time,
 };
+use super::identity::{CodexAuthenticatedAccountSource, CodexIdentityVerificationError};
+use super::recovery_log::{CodexOAuthRecoveryOperation, record_oauth_recovery};
+use super::security::CodexCredentialCodec;
 use super::token_client::{
     AuthorizationCodeExchangeError, AuthorizationCodeExchanger, AuthorizationCodeGrant,
     OFFICIAL_CODEX_OAUTH_CLIENT_ID, OFFICIAL_CODEX_REDIRECT_URI,
@@ -344,7 +348,7 @@ pub trait CodexOAuthAdmin: Send + Sync {
 pub struct CodexOAuthAdminService {
     pending: Arc<dyn CodexOAuthPendingStore>,
     exchanger: Arc<dyn AuthorizationCodeExchanger>,
-    verifier: Arc<dyn CodexAccountIdentityVerifier>,
+    account_source: Arc<dyn CodexAuthenticatedAccountSource>,
     store: Arc<dyn ProviderAccountStore>,
     runtime_policy: Arc<dyn ProviderRuntimePolicyPort>,
     credentials: CodexCredentialAdmin,
@@ -396,7 +400,7 @@ impl CodexOAuthAdminService {
     pub fn new(
         pending: Arc<dyn CodexOAuthPendingStore>,
         exchanger: Arc<dyn AuthorizationCodeExchanger>,
-        verifier: Arc<dyn CodexAccountIdentityVerifier>,
+        account_source: Arc<dyn CodexAuthenticatedAccountSource>,
         store: Arc<dyn ProviderAccountStore>,
         runtime_policy: Arc<dyn ProviderRuntimePolicyPort>,
         credentials: CodexCredentialAdmin,
@@ -404,7 +408,7 @@ impl CodexOAuthAdminService {
         Self {
             pending,
             exchanger,
-            verifier,
+            account_source,
             store,
             runtime_policy,
             credentials,
@@ -487,8 +491,6 @@ impl CodexOAuthAdminService {
         callback_url: &str,
     ) -> Result<(PendingAuthorizationMutation, CompletedCodexOAuthCredential), CodexOAuthAdminError>
     {
-        let (mut secret, id_token) = self.exchange_pending(&pending, callback_url).await?;
-        let mutation = pending.mutation.clone();
         let current = if let Some(target) = pending.reauthorization() {
             let current = self
                 .store
@@ -499,13 +501,19 @@ impl CodexOAuthAdminService {
         } else {
             None
         };
-        let expectation = CodexIdentityExpectation::default();
-        let profile = self
-            .verifier
-            .verify_authorization(&secret, &id_token, &pending.nonce, &expectation)
-            .await
-            .and_then(super::identity::CodexIdentityVerification::into_complete)
-            .map_err(map_identity_error)?;
+        let fallback_refresh_token = current
+            .as_ref()
+            .map(current_oauth_refresh_token)
+            .transpose()?
+            .flatten();
+        let (mut secret, id_token, expires_in) = self
+            .exchange_pending(&pending, callback_url, fallback_refresh_token)
+            .await?;
+        let mutation = pending.mutation.clone();
+        let access_token_expires_at = SystemTime::now()
+            .checked_add(expires_in)
+            .map(DateTime::<Utc>::from)
+            .ok_or(CodexOAuthAdminError::Credential)?;
         secret.id_token = Some(id_token);
         let policy = self
             .runtime_policy
@@ -516,28 +524,35 @@ impl CodexOAuthAdminService {
             let next_refresh_at = refresh_time(
                 policy,
                 current.account.id(),
-                profile.access_token_expires_at,
+                Some(access_token_expires_at),
                 secret.refresh_token.is_some(),
             )
             .map_err(map_admin_error)?;
             CompletedCodexOAuthCredential::Reauthorize(
                 self.credentials
-                    .prepare_reauthorization(RotateManagedCodexCredential {
+                    .prepare_refreshed_oauth_rotation(
                         current,
                         secret,
-                        verified_account: profile,
+                        access_token_expires_at,
                         next_refresh_at,
-                    })
+                    )
                     .map_err(map_admin_error)?,
             )
         } else {
             let account_id = format!("acct_{}", Uuid::now_v7().simple());
             let typed_account_id = ProviderAccountId::new(account_id.clone())
                 .map_err(|_| CodexOAuthAdminError::Credential)?;
+            let profile = complete_oauth_account_profile(
+                self.account_source.as_ref(),
+                &secret,
+                Some(access_token_expires_at),
+            )
+            .await
+            .map_err(map_account_profile_error)?;
             let next_refresh_at = refresh_time(
                 policy,
                 &typed_account_id,
-                profile.access_token_expires_at,
+                Some(access_token_expires_at),
                 secret.refresh_token.is_some(),
             )
             .map_err(map_admin_error)?;
@@ -654,7 +669,9 @@ impl CodexOAuthAdminService {
         &self,
         pending: &CodexPendingAuthorization,
         callback_url: &str,
-    ) -> Result<(super::types::CodexOAuthSecret, SecretString), CodexOAuthAdminError> {
+        fallback_refresh_token: Option<SecretString>,
+    ) -> Result<(super::types::CodexOAuthSecret, SecretString, Duration), CodexOAuthAdminError>
+    {
         let (code, callback_state) = callback_parts(callback_url)?;
         if !constant_time_equal(
             pending.state.expose_secret().as_bytes(),
@@ -670,10 +687,36 @@ impl CodexOAuthAdminService {
             })
             .await
             .map_err(map_exchange_error)?;
-        Ok((tokens.secret, tokens.id_token))
+        let mut secret = tokens.secret;
+        if secret.refresh_token.is_none() {
+            secret.refresh_token = fallback_refresh_token;
+        }
+        record_oauth_recovery(
+            CodexOAuthRecoveryOperation::AuthorizationCode,
+            pending
+                .reauthorization()
+                .map(|account_id| account_id.as_str()),
+            secret.access_token.expose_secret(),
+            secret
+                .refresh_token
+                .as_ref()
+                .map(ExposeSecret::expose_secret),
+        );
+        Ok((secret, tokens.id_token, tokens.expires_in))
     }
 }
 
+fn current_oauth_refresh_token(
+    current: &LoadedCredential,
+) -> Result<Option<SecretString>, CodexOAuthAdminError> {
+    let runtime = CodexCredentialCodec::decode(&current.credential)
+        .map_err(|_| CodexOAuthAdminError::Credential)?;
+    Ok(runtime
+        .authentication
+        .oauth()
+        .and_then(|oauth| oauth.refresh_token.clone()))
+}
+
 fn authorization_url(
     pending: &CodexPendingAuthorization,
     oauth_client_id: &str,
@@ -699,27 +742,19 @@ fn authorization_url(
 
 fn callback_parts(value: &str) -> Result<(SecretString, SecretString), CodexOAuthAdminError> {
     let url = Url::parse(value).map_err(|_| CodexOAuthAdminError::CallbackRejected)?;
-    if url.scheme() != "http"
-        || url.host_str() != Some("localhost")
-        || url.port() != Some(1455)
-        || url.path() != "/auth/callback"
-        || url.fragment().is_some()
-    {
-        return Err(CodexOAuthAdminError::CallbackRejected);
-    }
     let mut code = None;
     let mut state = None;
     for (key, value) in url.query_pairs() {
         match key.as_ref() {
             "code" => set_unique_callback_parameter(&mut code, value.into_owned())?,
             "state" => set_unique_callback_parameter(&mut state, value.into_owned())?,
-            // OAuth 服务可能附加 `scope`、`iss` 等标准参数。
-            // 固定回调地址与唯一的 `code`/`state` 才是安全边界，无关参数不参与校验。
+            // 回调地址只承载 query 参数；安全绑定由唯一的 code/state、
+            // 服务端保存的 state 和 PKCE token exchange 共同完成。
             _ => {}
         }
     }
-    let code = code.filter(|value| valid_secret(value));
-    let state = state.filter(|value| valid_secret(value));
+    let code = code.filter(|value| !value.is_empty());
+    let state = state.filter(|value| !value.is_empty());
     match (code, state) {
         (Some(code), Some(state)) => Ok((SecretString::from(code), SecretString::from(state))),
         _ => Err(CodexOAuthAdminError::CallbackRejected),
@@ -813,9 +848,9 @@ fn map_exchange_error(error: AuthorizationCodeExchangeError) -> CodexOAuthAdminE
     }
 }
 
-fn map_identity_error(error: CodexIdentityVerificationError) -> CodexOAuthAdminError {
+const fn map_account_profile_error(error: CodexIdentityVerificationError) -> CodexOAuthAdminError {
     match error {
-        CodexIdentityVerificationError::Rejected => CodexOAuthAdminError::IdentityRejected,
+        CodexIdentityVerificationError::Rejected => CodexOAuthAdminError::TokenRejected,
         CodexIdentityVerificationError::Unavailable => CodexOAuthAdminError::UpstreamUnavailable,
     }
 }

backend/crates/providers/openai/src/credential/quota/mod.rs

@@ -20,12 +20,13 @@ use gateway_core::engine::credential::{
     AccountAvailability, AccountQuotaSignals, CredentialRevision, OpaqueProviderData,
     ProviderAccount, ProviderAccountId, ProviderAccountStore, QuotaObservation, QuotaWriteOutcome,
 };
+use gateway_core::provider_ports::{ProviderCooldown, ProviderCooldownPort};
 use gateway_protocol::openai::events::{
     ParsedRateLimits, RateLimitDetails, RateLimitWindow, parse_rate_limit_headers,
 };
 use reqwest::Client;
 use secrecy::ExposeSecret;
-use serde_json::{Map, Number, Value};
+use serde_json::{Map, Value};
 use thiserror::Error;
 use tokio::sync::Mutex;
 use uuid::Uuid;
@@ -35,7 +36,7 @@ use crate::transport::{CodexBackendClient, CodexClientError, CodexRequestContext
 
 use super::agent_identity::{CodexAgentIdentityTaskService, PreparedCodexRuntimeCredential};
 use super::repository::{CodexCredentialRepository, CredentialRepositoryError};
-use recovery::{QuotaRecoveryEvidence, quota_state_transition, quota_success_state};
+use recovery::{quota_state_reason, quota_state_transition, quota_success_state};
 use snapshot::{
     parse_account_quota_snapshot, quota_projection_ttl, quota_snapshot_from_observation,
     scheduling_signals_from_snapshot,
@@ -120,6 +121,7 @@ pub struct CodexCredentialQuotaService {
     http: Client,
     base_url: String,
     agent_identity: Arc<CodexAgentIdentityTaskService>,
+    cooldowns: Arc<dyn ProviderCooldownPort>,
     scheduling: CodexQuotaSchedulingProjection,
 }
 
@@ -379,6 +381,7 @@ impl CodexCredentialQuotaService {
         http: Client,
         base_url: String,
         agent_identity: Arc<CodexAgentIdentityTaskService>,
+        cooldowns: Arc<dyn ProviderCooldownPort>,
     ) -> Self {
         Self {
             store: Arc::clone(repository.store()),
@@ -387,11 +390,12 @@ impl CodexCredentialQuotaService {
             http,
             base_url,
             agent_identity,
+            cooldowns,
             scheduling: CodexQuotaSchedulingProjection::default(),
         }
     }
 
-    /// 写入新快照前读取旧确认 reset（402 恢复用「旧 reset 已到期」判定）。
+    /// 写入新快照前读取已确认耗尽窗口的 reset，用于判断窗口是否真的推进。
     async fn previous_confirmed_reset(&self, account: &ProviderAccount) -> Option<SystemTime> {
         let observations = self
             .store
@@ -406,6 +410,55 @@ impl CodexCredentialQuotaService {
         snapshot.fact().resets_at().map(SystemTime::from)
     }
 
+    /// 将已确认的额度拒绝投影为 100%，使展示、调度和恢复共用同一窗口事实。
+    ///
+    /// `reset_at` 来自官方 `usage_limit_reached` 错误体；缺失时保留当前快照的 reset。
+    pub(crate) async fn record_confirmed_exhaustion(
+        &self,
+        account: &ProviderAccount,
+        reset_at: Option<i64>,
+    ) -> Result<(), CodexCredentialQuotaError> {
+        let existing = self
+            .store
+            .get_quotas(std::slice::from_ref(account.id()))
+            .await?
+            .into_iter()
+            .find(|observation| {
+                observation.account_id == *account.id()
+                    && observation.expected_revision == account.revision()
+            })
+            .and_then(|observation| observation.quota)
+            .map(OpaqueProviderData::into_inner)
+            .unwrap_or_default();
+        let quota = confirmed_exhaustion_projection(
+            normalize_quota_window_placeholders(existing),
+            reset_at,
+        );
+        let observed_at = SystemTime::now();
+        let outcome = self
+            .store
+            .compare_and_swap_quota(QuotaObservation {
+                account_id: account.id().clone(),
+                expected_revision: account.revision(),
+                quota: Some(OpaqueProviderData::new(quota.clone())),
+                observed_at: Some(observed_at),
+                limit_reached: Some(true),
+            })
+            .await?;
+        if outcome == QuotaWriteOutcome::Conflict {
+            return Err(CodexCredentialQuotaError::RevisionConflict);
+        }
+        if let Ok(snapshot) = parse_account_quota_snapshot(
+            account.id().clone(),
+            account.revision(),
+            observed_at,
+            &Value::Object(quota),
+        ) {
+            self.scheduling.observe(&snapshot);
+        }
+        Ok(())
+    }
+
     /// 批量预热请求级额度投影；持久层或 Provider JSON 异常只退化为未知额度。
     pub async fn prepare_scheduling(&self, accounts: &[ProviderAccount]) {
         if self.scheduling.hydration_targets(accounts).is_empty() {
@@ -495,21 +548,23 @@ impl CodexCredentialQuotaService {
                         // 402：真额度耗尽，写入 QuotaExhausted（worker 按 reset_at 自动恢复）。
                         Some(AccountAvailability::QuotaExhausted) => {
                             summary.exhausted += 1;
-                            let _ = self
-                                .repository
-                                .apply_state(
-                                    &account,
-                                    AccountAvailability::QuotaExhausted,
-                                    observed_at,
-                                )
-                                .await;
+                            self.persist_upstream_state(
+                                &account,
+                                AccountAvailability::QuotaExhausted,
+                                &error,
+                                observed_at,
+                            )
+                            .await;
                         }
                         Some(AccountAvailability::Banned) => {
                             summary.banned += 1;
-                            let _ = self
-                                .repository
-                                .apply_state(&account, AccountAvailability::Banned, observed_at)
-                                .await;
+                            self.persist_upstream_state(
+                                &account,
+                                AccountAvailability::Banned,
+                                &error,
+                                observed_at,
+                            )
+                            .await;
                         }
                         // 429/503 等临时拒绝：不写 availability（限流不改变账号可用性），仅记失败待重试。
                         None | Some(_) => {
@@ -535,18 +590,44 @@ impl CodexCredentialQuotaService {
         observed_at: SystemTime,
         summary: &mut CodexQuotaSyncSummary,
     ) -> Result<(), CodexCredentialQuotaError> {
-        let snapshot = parse_account_quota_snapshot(
+        let refreshed_snapshot = parse_account_quota_snapshot(
             account.id().clone(),
             account.revision(),
             observed_at,
             value,
         )?;
-        let fact = snapshot.fact();
+        let refreshed_fact = refreshed_snapshot.fact();
         let previous_reset_at = self.previous_confirmed_reset(account).await;
-        let object = value
-            .as_object()
-            .cloned()
-            .ok_or(CodexCredentialQuotaError::InvalidCredentialData)?;
+        let now = SystemTime::now();
+        let object = normalize_quota_window_placeholders(
+            value
+                .as_object()
+                .cloned()
+                .ok_or(CodexCredentialQuotaError::InvalidCredentialData)?,
+        );
+        let object = if account.availability() == AccountAvailability::QuotaExhausted
+            && quota_success_state(
+                account.availability(),
+                refreshed_fact,
+                previous_reset_at,
+                now,
+            )
+            .is_none()
+        {
+            confirmed_exhaustion_projection(
+                object,
+                confirmed_exhaustion_reset_at(previous_reset_at, refreshed_fact),
+            )
+        } else {
+            object
+        };
+        let snapshot = parse_account_quota_snapshot(
+            account.id().clone(),
+            account.revision(),
+            observed_at,
+            &Value::Object(object.clone()),
+        )?;
+        let fact = snapshot.fact();
         let outcome = self
             .store
             .compare_and_swap_quota(QuotaObservation {
@@ -577,10 +658,9 @@ impl CodexCredentialQuotaService {
         }
         if let Some(availability) = quota_success_state(
             current.availability(),
-            fact,
+            refreshed_fact,
             previous_reset_at,
-            SystemTime::now(),
-            QuotaRecoveryEvidence::RefreshedSnapshot,
+            now,
         ) {
             let _ = self
                 .repository
@@ -599,6 +679,15 @@ impl CodexCredentialQuotaService {
         let Some(rate_limits) = parse_rate_limit_headers(headers) else {
             return Ok(false);
         };
+        let has_quota_facts = rate_limits
+            .limits
+            .values()
+            .any(|details| passive_rate_limit_snapshot(details).is_some());
+        let previous_reset_at = if account.availability() == AccountAvailability::QuotaExhausted {
+            self.previous_confirmed_reset(account).await
+        } else {
+            None
+        };
         let existing = self
             .store
             .get_quotas(std::slice::from_ref(account.id()))
@@ -607,12 +696,48 @@ impl CodexCredentialQuotaService {
             .find(|observation| {
                 observation.account_id == *account.id()
                     && observation.expected_revision == account.revision()
-            })
+            });
+        let existing_observed_at = existing
+            .as_ref()
+            .and_then(|observation| observation.observed_at);
+        let existing = existing
             .and_then(|observation| observation.quota)
             .map(OpaqueProviderData::into_inner)
             .unwrap_or_default();
+        // 套餐、credits 等元数据可以更新，但没有额度窗口事实时必须保留旧观察时刻，
+        // 也不能借旧快照重新推导 quota state。
+        if !has_quota_facts {
+            let Some(observed_at) = existing_observed_at else {
+                return Ok(false);
+            };
+            let outcome = self
+                .store
+                .compare_and_swap_quota(QuotaObservation {
+                    account_id: account.id().clone(),
+                    expected_revision: account.revision(),
+                    quota: Some(OpaqueProviderData::new(merge_passive_quota(
+                        existing,
+                        &rate_limits,
+                    ))),
+                    observed_at: Some(observed_at),
+                    limit_reached: None,
+                })
+                .await?;
+            return Ok(outcome != QuotaWriteOutcome::Conflict);
+        }
         let quota = merge_passive_quota(existing, &rate_limits);
         let observed_at = SystemTime::now();
+        let refreshed_fact = parse_codex_quota_usage(&Value::Object(quota.clone()))?;
+        let quota = if account.availability() == AccountAvailability::QuotaExhausted {
+            // 被动成功响应不能解除确认额度耗尽；继续展示 100%，直到权威 usage 快照
+            // 证明旧 reset 已跨过且新窗口已经推进。
+            confirmed_exhaustion_projection(
+                quota,
+                confirmed_exhaustion_reset_at(previous_reset_at, refreshed_fact),
+            )
+        } else {
+            quota
+        };
         let fact = parse_codex_quota_usage(&Value::Object(quota.clone()))?;
         let snapshot = parse_account_quota_snapshot(
             account.id().clone(),
@@ -627,123 +752,68 @@ impl CodexCredentialQuotaService {
                 expected_revision: account.revision(),
                 quota: Some(OpaqueProviderData::new(quota)),
                 observed_at: Some(observed_at),
-                limit_reached: Some(true),
+                limit_reached: Some(fact.exhausted()),
             })
             .await?;
         if outcome == QuotaWriteOutcome::Conflict {
             return Ok(false);
         }
-        // 可用性写回必须基于最新账号状态：revision fence 只覆盖凭据轮换，
-        // 不覆盖其他路径的仅状态变化（Cooldown/QuotaExhausted 转换）。
-        let Some(current) = self.store.get_account(account.id()).await? else {
-            return Ok(false);
-        };
-        if current.revision() != account.revision() {
-            return Ok(false);
-        }
         if let Ok(snapshot) = snapshot {
             self.scheduling.observe(&snapshot);
         }
-        if let Some(availability) = quota_success_state(
-            current.availability(),
-            fact,
-            None,
-            observed_at,
-            QuotaRecoveryEvidence::SuccessfulResponse,
-        ) {
-            self.repository
-                .apply_state(&current, availability, observed_at)
-                .await?;
-        }
         Ok(true)
     }
 
-    /// 真实 429 的单一事实入口：合成/更新 quota 主窗口（`limit_reached=true`、
-    /// `used_percent=100`、`reset_at=retry-after`）。availability 不改变，
-    /// 窗口 reset 到期后由投影自动滚动恢复（429 只记录限流窗口）。
-    /// 已有更晚的 `reset_at` 不会被缩短（never-shrink）。
+    /// 真实 429 的单一事实入口：写入 Redis 临时限流冷却（`until = now + retry_after`）。
+    /// availability 与 quota 主窗口（额度重置时间）都不改变——临时限流是独立维度，
+    /// 到期由 Redis key 过期自动解除，不污染配额耗尽状态。
+    /// 已有更晚的冷却不会被缩短（put_if_later）。
     pub async fn apply_rate_limit_429(
         &self,
         account: &ProviderAccount,
         retry_after: Option<Duration>,
         observed_at: SystemTime,
     ) -> Result<(), CodexCredentialQuotaError> {
-        let mut object = self
-            .store
-            .get_quotas(std::slice::from_ref(account.id()))
-            .await?
-            .into_iter()
-            .find(|observation| {
-                observation.account_id == *account.id()
-                    && observation.expected_revision == account.revision()
-            })
-            .and_then(|observation| observation.quota)
-            .map(OpaqueProviderData::into_inner)
-            .unwrap_or_default();
-        let new_reset_at = observed_at
+        let until = observed_at
             .checked_add(retry_after.unwrap_or(DEFAULT_RATE_LIMIT_COOLDOWN))
-            .and_then(|reset| reset.duration_since(UNIX_EPOCH).ok())
-            .map(|duration| duration.as_secs() as i64);
-        let existing_reset_at = object
-            .get("rate_limit")
-            .and_then(Value::as_object)
-            .and_then(|rate_limit| rate_limit.get("primary_window"))
-            .and_then(Value::as_object)
-            .and_then(|window| window.get("reset_at"))
-            .and_then(Value::as_i64);
-        let reset_at = match (existing_reset_at, new_reset_at) {
-            (Some(existing), Some(next)) if existing > next => Some(existing),
-            (_, next) => next,
-        };
-        let mut rate_limit = object
-            .get("rate_limit")
-            .cloned()
-            .and_then(|value| value.as_object().cloned())
-            .unwrap_or_default();
-        rate_limit.insert("allowed".to_owned(), Value::Bool(false));
-        rate_limit.insert("limit_reached".to_owned(), Value::Bool(true));
-        let mut primary = rate_limit
-            .get("primary_window")
-            .cloned()
-            .and_then(|value| value.as_object().cloned())
-            .unwrap_or_default();
-        let used_percent = primary
-            .get("used_percent")
-            .and_then(Value::as_f64)
-            .unwrap_or(0.0)
-            .max(100.0);
-        primary.insert(
-            "used_percent".to_owned(),
-            Number::from_f64(used_percent)
-                .map(Value::Number)
-                .unwrap_or_else(|| Value::Number(Number::from(100))),
-        );
-        if let Some(reset_at) = reset_at {
-            primary.insert("reset_at".to_owned(), Value::from(reset_at));
-        }
-        rate_limit.insert("primary_window".to_owned(), Value::Object(primary));
-        object.insert("rate_limit".to_owned(), Value::Object(rate_limit));
-        if self
-            .store
-            .compare_and_swap_quota(QuotaObservation {
-                account_id: account.id().clone(),
-                expected_revision: account.revision(),
-                quota: Some(OpaqueProviderData::new(object)),
-                observed_at: Some(observed_at),
-                limit_reached: Some(true),
-            })
-            .await?
-            == QuotaWriteOutcome::Conflict
-        {
-            return Ok(());
-        }
-        // 429 合成的限流窗口要立即反映到调度信号（reset 到期由投影滚动解除）。
-        if let Ok(Some(snapshot)) = self.read_snapshot_for(account).await {
-            self.scheduling.observe(&snapshot);
-        }
+            .unwrap_or(observed_at);
+        self.cooldowns
+            .put_if_later(ProviderCooldown::new(
+                account.id().clone(),
+                account.revision(),
+                until,
+            ))
+            .await
+            .map_err(|error| CodexCredentialQuotaError::Store {
+                detail: error.to_string(),
+            })?;
         Ok(())
     }
 
+    /// 读取账号当前是否处于临时限流（429）冷却，及到期时间。
+    pub async fn rate_limited_until(
+        &self,
+        account_id: &ProviderAccountId,
+        credential_revision: CredentialRevision,
+    ) -> Result<Option<SystemTime>, CodexCredentialQuotaError> {
+        let Some(cooldown) = self.cooldowns.read(account_id).await.map_err(|error| {
+            CodexCredentialQuotaError::Store {
+                detail: error.to_string(),
+            }
+        })?
+        else {
+            return Ok(None);
+        };
+        if cooldown.credential_revision() != credential_revision {
+            return Ok(None);
+        }
+        let until = cooldown.until();
+        if until <= SystemTime::now() {
+            return Ok(None);
+        }
+        Ok(Some(until))
+    }
+
     /// 读取单账号最后一次落库的 Provider quota，并由 Codex 域解析展示窗口。
     pub async fn read_account(
         &self,
@@ -845,9 +915,7 @@ impl CodexCredentialQuotaService {
                 Err(CodexQuotaFetchError::Upstream { account, error }) => {
                     // 402/deactivated → 写终态；429/503 等临时拒绝不写 availability（限流不改变账号可用性）。
                     if let Some(availability) = quota_state_transition(&error) {
-                        let _ = self
-                            .repository
-                            .apply_state(&account, availability, observed_at)
+                        self.persist_upstream_state(&account, availability, &error, observed_at)
                             .await;
                     }
                     return Err(CodexCredentialQuotaError::Upstream {
@@ -855,17 +923,43 @@ impl CodexCredentialQuotaService {
                     });
                 }
             };
-        let snapshot = parse_account_quota_snapshot(
+        let refreshed_snapshot = parse_account_quota_snapshot(
             account.id().clone(),
             account.revision(),
             observed_at,
             &value,
         )?;
+        let refreshed_fact = refreshed_snapshot.fact();
         let previous_reset_at = self.previous_confirmed_reset(&account).await;
-        let object = value
-            .as_object()
-            .cloned()
-            .ok_or(CodexCredentialQuotaError::InvalidCredentialData)?;
+        let now = SystemTime::now();
+        let object = normalize_quota_window_placeholders(
+            value
+                .as_object()
+                .cloned()
+                .ok_or(CodexCredentialQuotaError::InvalidCredentialData)?,
+        );
+        let object = if account.availability() == AccountAvailability::QuotaExhausted
+            && quota_success_state(
+                account.availability(),
+                refreshed_fact,
+                previous_reset_at,
+                now,
+            )
+            .is_none()
+        {
+            confirmed_exhaustion_projection(
+                object,
+                confirmed_exhaustion_reset_at(previous_reset_at, refreshed_fact),
+            )
+        } else {
+            object
+        };
+        let snapshot = parse_account_quota_snapshot(
+            account.id().clone(),
+            account.revision(),
+            observed_at,
+            &Value::Object(object.clone()),
+        )?;
         if self
             .store
             .compare_and_swap_quota(QuotaObservation {
@@ -891,10 +985,9 @@ impl CodexCredentialQuotaService {
         self.scheduling.observe(&snapshot);
         if let Some(availability) = quota_success_state(
             current.availability(),
-            snapshot.fact(),
+            refreshed_fact,
             previous_reset_at,
-            SystemTime::now(),
-            QuotaRecoveryEvidence::RefreshedSnapshot,
+            now,
         ) {
             self.repository
                 .apply_state(&current, availability, observed_at)
@@ -903,6 +996,29 @@ impl CodexCredentialQuotaService {
         Ok(snapshot)
     }
 
+    async fn persist_upstream_state(
+        &self,
+        account: &ProviderAccount,
+        availability: AccountAvailability,
+        upstream_error: &CodexClientError,
+        observed_at: SystemTime,
+    ) {
+        let reason = quota_state_reason(upstream_error).map(str::to_owned);
+        if let Err(error) = self
+            .repository
+            .apply_state_with_message(account, availability, observed_at, reason.clone())
+            .await
+        {
+            tracing::warn!(
+                account_id = %account.id(),
+                ?availability,
+                reason = reason.as_deref().unwrap_or_default(),
+                error = %error,
+                "OpenAI quota upstream state write failed"
+            );
+        }
+    }
+
     async fn fetch_usage_with_recovery(
         &self,
         client: &CodexBackendClient,
@@ -1095,6 +1211,103 @@ fn merge_passive_quota(
     quota
 }
 
+/// 丢弃 core 限额中上游给出的无事实 `secondary_window` 占位。
+///
+/// 生产响应头可能携带 `used_percent=0`、零时长和空 reset 的占位，`/usage` 也会
+/// 返回 `secondary_window: null`。正常被动同步保留该响应事实，Admin 展示会将其
+/// 隐藏；但主动 `/usage` 刷新或 402 确认投影会把一个存在但无事实的字段写成
+/// 100%，从而显示并不存在的“次级额度”。因此这些非被动写入口在落库前移除
+/// 无事实值。带 reset、时长、正用量、触顶、未知或非法字段的次级窗口均完全按
+/// 原有额度逻辑保留。
+fn normalize_quota_window_placeholders(mut quota: Map<String, Value>) -> Map<String, Value> {
+    if let Some(rate_limit) = quota.get_mut("rate_limit").and_then(Value::as_object_mut) {
+        drop_secondary_window_placeholder(rate_limit);
+    }
+    quota
+}
+
+fn drop_secondary_window_placeholder(rate_limit: &mut Map<String, Value>) {
+    let placeholder = rate_limit.get("secondary_window").is_some_and(|window| {
+        window.is_null()
+            || window
+                .as_object()
+                .is_some_and(secondary_window_is_placeholder)
+    });
+    if placeholder {
+        rate_limit.remove("secondary_window");
+    }
+}
+
+fn secondary_window_is_placeholder(window: &Map<String, Value>) -> bool {
+    window.iter().all(|(field, value)| match field.as_str() {
+        "used_percent" => value
+            .as_f64()
+            .is_some_and(|used_percent| used_percent.is_finite() && used_percent == 0.0),
+        "limit_reached" => value.as_bool() == Some(false),
+        _ => false,
+    })
+}
+
+/// 用已确认的上游额度拒绝覆盖 core 限额投影。
+///
+/// 上游 usage 快照可能在结算期间短暂滞后为 98/99% 或 `allowed=true`；保留其他
+/// 原始 metadata，但把 core 窗口固定为 100%，直到权威快照证明进入下一窗口。
+fn confirmed_exhaustion_projection(
+    mut quota: Map<String, Value>,
+    reset_at: Option<i64>,
+) -> Map<String, Value> {
+    let mut rate_limit = quota
+        .remove("rate_limit")
+        .and_then(|value| value.as_object().cloned())
+        .unwrap_or_default();
+    rate_limit.insert("allowed".to_owned(), Value::Bool(false));
+    rate_limit.insert("limit_reached".to_owned(), Value::Bool(true));
+    for field in ["primary_window", "secondary_window"] {
+        let existed = rate_limit.contains_key(field);
+        let mut window = rate_limit
+            .remove(field)
+            .and_then(|value| value.as_object().cloned())
+            .unwrap_or_default();
+        if field == "primary_window" || existed {
+            window.insert("used_percent".to_owned(), Value::from(100_u64));
+            if field == "primary_window"
+                && let Some(reset_at) = reset_at
+            {
+                window.insert("reset_at".to_owned(), Value::from(reset_at));
+            }
+            rate_limit.insert(field.to_owned(), Value::Object(window));
+        }
+    }
+    quota.insert("rate_limit".to_owned(), Value::Object(rate_limit));
+    quota
+}
+
+/// 选择应继续锁定的 reset：新鲜快照仍确认耗尽时可把锁推进到更晚窗口；
+/// 否则保留已确认失败的旧窗口，避免 98/99% 快照反向解锁。
+fn confirmed_exhaustion_reset_at(
+    previous_reset_at: Option<SystemTime>,
+    refreshed_fact: CodexQuotaFact,
+) -> Option<i64> {
+    let refreshed_reset_at = refreshed_fact.resets_at().map(SystemTime::from);
+    let reset_at = match (previous_reset_at, refreshed_reset_at) {
+        (Some(previous), Some(refreshed)) if refreshed_fact.exhausted() && refreshed > previous => {
+            refreshed
+        }
+        (Some(previous), _) => previous,
+        (None, refreshed) => refreshed?,
+    };
+    system_time_to_unix_seconds(reset_at)
+}
+
+fn system_time_to_unix_seconds(value: SystemTime) -> Option<i64> {
+    value
+        .duration_since(UNIX_EPOCH)
+        .ok()?
+        .as_secs()
+        .try_into()
+        .ok()
+}
+
 fn passive_rate_limit_snapshot(details: &RateLimitDetails) -> Option<Map<String, Value>> {
     // 响应头的窗口属于同一次上游观测；跨响应合并会制造不存在的额度窗口。
     let mut snapshot = Map::new();

backend/crates/providers/openai/src/credential/quota/recovery.rs

@@ -1,7 +1,7 @@
-//! 402 恢复证据与 availability 转换。
+//! 确认额度耗尽后的恢复证据与 availability 转换。
 //!
-//! `QuotaExhausted -> Ready` 只接受同一 credential revision 的真实成功响应、
-//! 新鲜快照显式 `allowed=true` 且无 exhaustion、或已到期旧 reset 配合未耗尽新快照。
+//! `QuotaExhausted -> Ready` 只能由同一 credential revision 的权威 usage 快照证明：
+//! 已确认的旧窗口 reset 已到期，且新快照的 reset 已推进到下一窗口。
 
 use std::time::SystemTime;
 
@@ -12,47 +12,24 @@ use serde_json::Value;
 use super::snapshot::CodexQuotaFact;
 use crate::transport::CodexClientError;
 
-/// 402 恢复可接受的证据强度。
-#[derive(Debug, Clone, Copy, PartialEq, Eq)]
-pub(crate) enum QuotaRecoveryEvidence {
-    /// 一次真实成功推理响应（最强证据）。
-    SuccessfulResponse,
-    /// 周期 worker 刷新出的新快照（需旧 reset 已到期佐证）。
-    RefreshedSnapshot,
-}
-
-/// 只处理 `QuotaExhausted` 的恢复——429/触顶不再进入该状态，由 quota 数据
-/// 驱动调度排除（限流不改变账号可用性）。`Invalid/Expired/Banned` 等终态由 selector 的
-/// 成功响应恢复路径处理。
+/// 只处理 `QuotaExhausted` 的恢复。`allowed=true`、成功响应和百分比回落都不能单独
+/// 覆盖一次已确认的额度拒绝；它们在 reset 未推进时可能只是上游滞后的观测。
 ///
-/// `previous_reset_at` 是写入新快照前读取的旧确认 reset；`RefreshedSnapshot`
-/// 恢复要求它已到期（窗口真实滚动后新快照的 reset 指向下一个未来周期，
-/// 不能用它判定旧耗尽窗口是否已结束）。
+/// `previous_reset_at` 是写入新快照前读取的已确认耗尽窗口。恢复必须同时满足：
+/// 旧窗口已经到期，且权威快照的 reset 指向比旧窗口更晚的新周期。
 pub(crate) fn quota_success_state(
     current: AccountAvailability,
     fact: CodexQuotaFact,
     previous_reset_at: Option<SystemTime>,
     now: SystemTime,
-    recovery_evidence: QuotaRecoveryEvidence,
 ) -> Option<AccountAvailability> {
     if current != AccountAvailability::QuotaExhausted || fact.exhausted() {
         return None;
     }
-    // 新鲜快照显式 `allowed=true` 且无 exhaustion，直接恢复。
-    if fact.explicit_allowed() {
-        return Some(AccountAvailability::Ready);
-    }
-    // 一次真实成功响应是最强证据；周期刷新需要旧确认 reset 已到期
-    // （used_percent 是可滞后、会取整的观测值，其回落不能覆盖仍未到期的旧 reset）。
-    match recovery_evidence {
-        QuotaRecoveryEvidence::SuccessfulResponse => Some(AccountAvailability::Ready),
-        QuotaRecoveryEvidence::RefreshedSnapshot
-            if previous_reset_at.is_some_and(|reset_at| reset_at <= now) =>
-        {
-            Some(AccountAvailability::Ready)
-        }
-        QuotaRecoveryEvidence::RefreshedSnapshot => None,
-    }
+    let previous_reset_at = previous_reset_at?;
+    let refreshed_reset_at = fact.resets_at().map(SystemTime::from)?;
+    (previous_reset_at <= now && refreshed_reset_at > previous_reset_at)
+        .then_some(AccountAvailability::Ready)
 }
 
 pub(crate) fn quota_state_transition(error: &CodexClientError) -> Option<AccountAvailability> {
@@ -73,6 +50,18 @@ pub(crate) fn quota_state_transition(error: &CodexClientError) -> Option<Account
     }
 }
 
+/// 配额接口将账号变为非 ready 状态时可安全持久化的上游原因码。
+pub(crate) fn quota_state_reason(error: &CodexClientError) -> Option<&'static str> {
+    match quota_state_transition(error)? {
+        AccountAvailability::Banned => Some("deactivated_workspace"),
+        AccountAvailability::QuotaExhausted => Some("payment_required"),
+        AccountAvailability::Unknown
+        | AccountAvailability::Ready
+        | AccountAvailability::Expired
+        | AccountAvailability::Invalid => None,
+    }
+}
+
 fn is_deactivated_workspace(body: &str) -> bool {
     serde_json::from_str::<Value>(body)
         .ok()

backend/crates/providers/openai/src/credential/quota/snapshot.rs

@@ -18,8 +18,6 @@ pub struct CodexQuotaFact {
     remaining_percent: Option<u8>,
     resets_at: Option<DateTime<Utc>>,
     exhausted: bool,
-    /// 上游显式 `allowed=true`（恢复证据：新鲜快照明确放行）。
-    explicit_allowed: bool,
 }
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
@@ -41,7 +39,7 @@ pub enum CodexQuotaWindowRole {
 pub struct CodexQuotaWindow {
     key: String,
     source: String,
-    /// 上游提供的限流名称（`metered_feature`/`limit_name`/`limitId`）；显示优先用它。
+    /// 上游提供的限流桶名称（`metered_feature`/`limit_name`/`limitId`）；用于非核心桶的展示前缀。
     limit_name: Option<String>,
     kind: CodexQuotaWindowKind,
     role: CodexQuotaWindowRole,
@@ -189,12 +187,6 @@ impl CodexQuotaFact {
     pub const fn exhausted(&self) -> bool {
         self.exhausted
     }
-
-    /// 上游是否显式 `allowed=true`（新鲜快照的恢复证据）。
-    #[must_use]
-    pub const fn explicit_allowed(&self) -> bool {
-        self.explicit_allowed
-    }
 }
 
 pub fn parse_codex_quota_usage(usage: &Value) -> Result<CodexQuotaFact, CodexCredentialQuotaError> {
@@ -220,7 +212,6 @@ pub fn parse_codex_quota_usage(usage: &Value) -> Result<CodexQuotaFact, CodexCre
         remaining_percent: aggregate.remaining_percent,
         resets_at: aggregate.resets_at,
         exhausted: aggregate.exhausted,
-        explicit_allowed: aggregate.explicit_allowed,
     })
 }
 
@@ -319,6 +310,7 @@ pub(crate) fn parse_account_quota_snapshot(
     let mut windows = Vec::new();
     for limit in canonical_rate_limits(object)? {
         parse_rate_limit_windows(
+            &limit.key,
             &limit.source,
             limit.limit_name.as_deref(),
             limit.rate_limit,
@@ -335,8 +327,9 @@ pub(crate) fn parse_account_quota_snapshot(
     })
 }
 
-/// 一个规范化后的限流桶：source 是调度/展示分组键，limit_name 是上游显示名。
+/// 一个规范化后的限流桶：`key` 用于稳定区分窗口，`source` 用于展示分组。
 struct CanonicalRateLimit<'a> {
+    key: String,
     source: String,
     limit_name: Option<String>,
     rate_limit: &'a Value,
@@ -346,9 +339,15 @@ fn canonical_rate_limits(
     object: &Map<String, Value>,
 ) -> Result<Vec<CanonicalRateLimit<'_>>, CodexCredentialQuotaError> {
     let mut limits = Vec::new();
+    let has_primary_codex_rate_limit = object
+        .get("rate_limit")
+        .is_some_and(|value| !value.is_null());
     if let Some(rate_limit) = object.get("rate_limit") {
+        // 官方 Codex 将顶层 `rate_limit` 固定识别为 `codex`，而不把
+        // `active_limit` 投影成另一个展示桶。后者只是当前活动套餐的元数据。
         limits.push(CanonicalRateLimit {
-            source: "core".to_owned(),
+            key: "core".to_owned(),
+            source: "codex".to_owned(),
             limit_name: None,
             rate_limit,
         });
@@ -358,6 +357,7 @@ fn canonical_rate_limits(
         .filter(|value| !value.is_null())
     {
         limits.push(CanonicalRateLimit {
+            key: "code-review".to_owned(),
             source: "code_review".to_owned(),
             limit_name: Some("code_review".to_owned()),
             rate_limit,
@@ -379,23 +379,29 @@ fn canonical_rate_limits(
             .as_object()
             .ok_or(CodexCredentialQuotaError::InvalidCredentialData)?;
         let (source, limit_name) = canonical_rate_limit_source(item, index);
+        // 部分上游响应会在 `additional_rate_limits` 中重复默认的 `codex`
+        // 桶。它没有独立的 limit id，不能在账号面板再次展示。
+        if has_primary_codex_rate_limit && is_codex_limit_id(&source) {
+            continue;
+        }
         if let Some(rate_limit) = item.get("rate_limit") {
-            // additional 里的 codex 桶覆盖顶层 rate_limit（同源替换）。
-            if let Some(existing) = limits.iter_mut().find(|existing| existing.source == source) {
-                existing.rate_limit = rate_limit;
-                existing.limit_name = limit_name;
-            } else {
-                limits.push(CanonicalRateLimit {
-                    source,
-                    limit_name,
-                    rate_limit,
-                });
-            }
+            // 非 Codex additional 桶保留为独立快照；额外用索引生成窗口键，
+            // 避免同名 additional 产生重复前端 key。
+            limits.push(CanonicalRateLimit {
+                key: format!("additional-{index}-{source}"),
+                source,
+                limit_name,
+                rate_limit,
+            });
         }
     }
     Ok(limits)
 }
 
+fn is_codex_limit_id(value: &str) -> bool {
+    value.trim().eq_ignore_ascii_case("codex")
+}
+
 fn canonical_rate_limit_source(
     item: &Map<String, Value>,
     index: usize,
@@ -417,14 +423,11 @@ fn canonical_rate_limit_source(
             !value.is_empty() && value.len() <= 128 && !value.chars().any(char::is_control)
         })
         .map(str::to_owned);
-    if source.trim().to_ascii_lowercase().replace(['-', ' '], "_") == "codex" {
-        ("core".to_owned(), limit_name)
-    } else {
-        (source, limit_name)
-    }
+    (source, limit_name)
 }
 
 fn parse_rate_limit_windows(
+    key_source: &str,
     source: &str,
     limit_name: Option<&str>,
     value: &Value,
@@ -482,7 +485,7 @@ fn parse_rate_limit_windows(
             || used_percent.is_some_and(|used| used >= 100.0);
         let kind = quota_window_kind(window_seconds);
         output.push(CodexQuotaWindow {
-            key: format!("{}-{}", quota_key(source), quota_role_name(role, kind)),
+            key: format!("{}-{}", quota_key(key_source), quota_role_name(role, kind)),
             source: source.to_owned(),
             limit_name: limit_name.map(str::to_owned),
             kind,
@@ -559,7 +562,6 @@ const fn quota_role_name(role: CodexQuotaWindowRole, kind: CodexQuotaWindowKind)
 struct QuotaAggregate {
     recognized: bool,
     exhausted: bool,
-    explicit_allowed: bool,
     remaining_percent: Option<u8>,
     resets_at: Option<DateTime<Utc>>,
 }
@@ -572,7 +574,6 @@ impl QuotaAggregate {
         self.recognized = true;
         self.exhausted |= optional_bool(object, "limit_reached")?.unwrap_or(false);
         self.exhausted |= optional_bool(object, "allowed")?.is_some_and(|allowed| !allowed);
-        self.explicit_allowed |= optional_bool(object, "allowed")?.is_some_and(|allowed| allowed);
         for key in ["primary_window", "secondary_window"] {
             if let Some(window) = object.get(key) {
                 self.observe_window(window)?;

backend/crates/providers/openai/src/credential/recovery_log.rs

@@ -0,0 +1,58 @@
+//! 成功取得 OpenAI OAuth token 后写入普通应用日志的恢复记录。
+//!
+//! 此日志会刻意保存原始 AT/RT，避免后续数据库写入或校验失败时，已成功
+//! 交换的账号无法恢复。记录直接进入现有结构化日志，沿用其滚动和清理策略。
+
+const RECOVERY_LOG_TARGET: &str = "openai_oauth_recovery";
+
+/// 产生可恢复 token 组的 OAuth 操作。
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub(crate) enum CodexOAuthRecoveryOperation {
+    /// 直接提供 AT/RT 的导入。
+    ImportDirect,
+    /// 仅提供 RT 的导入，先交换得到 AT。
+    ImportRefreshToken,
+    /// 管理员手动触发的 RT 交换。
+    ManualRefresh,
+    /// 调度器触发的 RT 交换。
+    ScheduledRefresh,
+    /// 授权码加 PKCE 的 token 交换。
+    AuthorizationCode,
+}
+
+impl CodexOAuthRecoveryOperation {
+    const fn as_str(self) -> &'static str {
+        match self {
+            Self::ImportDirect => "import_direct",
+            Self::ImportRefreshToken => "import_refresh_token",
+            Self::ManualRefresh => "manual_refresh",
+            Self::ScheduledRefresh => "scheduled_refresh",
+            Self::AuthorizationCode => "authorization_code",
+        }
+    }
+}
+
+/// 将已取得的 AT/RT 写入普通结构化日志。
+///
+/// 调用方必须在任何可能拒绝响应的本地校验、画像补全或数据库写入之前调用。
+/// 该调用不执行额外 I/O，也不返回业务错误。Host 负责普通日志文件的分割、清理和
+/// 输出；日志系统不可用时按其既有策略处理，不能阻断 OAuth 业务流程。
+pub(crate) fn record_oauth_recovery(
+    operation: CodexOAuthRecoveryOperation,
+    account_id: Option<&str>,
+    access_token: &str,
+    refresh_token: Option<&str>,
+) {
+    let has_refresh_token = refresh_token.is_some();
+    let refresh_token = refresh_token.unwrap_or_default();
+    tracing::info!(
+        target: RECOVERY_LOG_TARGET,
+        event = "openai_oauth_recovery",
+        operation = operation.as_str(),
+        account_id = account_id.unwrap_or_default(),
+        access_token,
+        refresh_token,
+        has_refresh_token,
+        "OpenAI OAuth recovery token record"
+    );
+}

backend/crates/providers/openai/src/credential/refresh.rs

@@ -11,29 +11,103 @@ use gateway_core::provider_ports::{
     ProviderCredentialStatePort, ProviderLeaseAcquisition, ProviderLeasePort, ProviderLeaseRequest,
     ProviderRefreshCapacityRequest, ProviderRefreshLeaseRequest, ProviderRefreshPolicy,
     ProviderRuntimePolicyPort, ProviderStoreError, provider_refresh_backoff_at,
+    provider_refresh_retry_at,
 };
 use secrecy::{ExposeSecret, SecretString};
 use thiserror::Error;
 
-use super::identity::{
-    CodexAccountIdentityVerifier, CodexIdentityExpectation, CodexIdentityVerification,
-    CodexIdentityVerificationError,
-};
+use super::identity::CodexAccountIdentityVerifier;
+use super::recovery_log::{CodexOAuthRecoveryOperation, record_oauth_recovery};
 use super::repository::{CodexCredentialRepository, CredentialRepositoryError};
 use super::token_client::{RefreshFailure, TokenPair, TokenRefresher};
-use super::types::{CodexCredentialPrincipal, CodexOAuthSecret, RotateCodexCredential};
+use super::types::CodexOAuthSecret;
 
 const PROVIDER_NAME: &str = "openai";
 const MAX_REFRESH_BATCH: u32 = 1_000;
 /// 连续失败计数窗口；每次瞬态失败刷新该 TTL，静默满窗后计数过期归零。
 const REFRESH_BACKOFF_WINDOW: Duration = Duration::from_secs(30 * 60);
+/// 窗口内先执行五次指数退避，再进入固定恢复周期。
+const REFRESH_BACKOFF_MAX_ATTEMPTS: u32 = 5;
+/// 耗尽指数退避后的 OAuth 恢复周期。
+const REFRESH_RECOVERY_DELAY: Duration = Duration::from_secs(10 * 60);
+/// 过期 AT 仍允许 RT 恢复的最长窗口。
+const REFRESH_RECOVERY_WINDOW: Duration = Duration::from_secs(24 * 60 * 60);
+
+pub(crate) fn refresh_recovery_deadline(
+    access_token_expires_at: Option<SystemTime>,
+) -> Option<SystemTime> {
+    access_token_expires_at.and_then(|expires_at| expires_at.checked_add(REFRESH_RECOVERY_WINDOW))
+}
+
+fn refresh_recovery_window_exhausted(
+    access_token_expires_at: Option<SystemTime>,
+    now: SystemTime,
+) -> bool {
+    refresh_recovery_deadline(access_token_expires_at).is_some_and(|deadline| deadline <= now)
+}
+
+fn bounded_refresh_retry_at(
+    retry_at: SystemTime,
+    access_token_expires_at: Option<SystemTime>,
+    observed_at: SystemTime,
+) -> SystemTime {
+    let mut bounded = retry_at;
+    if let Some(expires_at) = access_token_expires_at
+        && expires_at > observed_at
+        && bounded > expires_at
+    {
+        bounded = expires_at;
+    }
+    if let Some(deadline) = refresh_recovery_deadline(access_token_expires_at)
+        && bounded > deadline
+    {
+        bounded = deadline;
+    }
+    bounded
+}
+
+fn oauth_refresh_retry_at(
+    account_id: &ProviderAccountId,
+    access_token_expires_at: Option<SystemTime>,
+    observed_at: SystemTime,
+    attempt: u32,
+    reason: &'static str,
+) -> Result<SystemTime, ProviderStoreError> {
+    let retry_at = if attempt <= REFRESH_BACKOFF_MAX_ATTEMPTS {
+        provider_refresh_backoff_at(account_id, observed_at, attempt, reason)?
+    } else {
+        provider_refresh_retry_at(account_id, observed_at, REFRESH_RECOVERY_DELAY, reason)?
+    };
+    Ok(bounded_refresh_retry_at(
+        retry_at,
+        access_token_expires_at,
+        observed_at,
+    ))
+}
+
+fn log_refresh_deferred(
+    account_id: &ProviderAccountId,
+    access_token_expires_at: Option<SystemTime>,
+    attempt: u32,
+    reason: &'static str,
+    retry_at: SystemTime,
+) {
+    tracing::warn!(
+        account_id = %account_id,
+        attempt,
+        reason,
+        retry_at = %DateTime::<Utc>::from(retry_at),
+        access_token_expires_at = ?access_token_expires_at.map(DateTime::<Utc>::from),
+        recovery_deadline = ?refresh_recovery_deadline(access_token_expires_at)
+            .map(DateTime::<Utc>::from),
+        "OpenAI OAuth refresh deferred"
+    );
+}
 
 fn refresh_due_at(account: &ProviderAccount, now: SystemTime) -> Option<SystemTime> {
-    if let Some(expires_at) = account.access_token_expires_at()
-        && expires_at <= now
-    {
-        // AT 已过期时不能继续等待持久退避；后续额度/请求探测都必须先用 RT 续出可用 AT。
-        return Some(expires_at);
+    if refresh_recovery_window_exhausted(account.access_token_expires_at(), now) {
+        // 即使旧 next_refresh_at 异常地落在 deadline 之后，也必须进入终态化路径。
+        return refresh_recovery_deadline(account.access_token_expires_at());
     }
     account.next_refresh_at().filter(|next| *next <= now)
 }
@@ -41,8 +115,6 @@ fn refresh_due_at(account: &ProviderAccount, now: SystemTime) -> Option<SystemTi
 pub struct DueCodexCredential {
     pub account: ProviderAccount,
     pub secret: CodexOAuthSecret,
-    principal: CodexCredentialPrincipal,
-    installation_id: String,
 }
 
 impl std::fmt::Debug for DueCodexCredential {
@@ -94,7 +166,6 @@ pub enum CodexCredentialRefreshError {
 pub struct CodexCredentialRefreshService {
     repository: CodexCredentialRepository,
     refresher: Arc<dyn TokenRefresher>,
-    identity: Arc<dyn CodexAccountIdentityVerifier>,
     leases: Arc<dyn ProviderLeasePort>,
     credential_state: Arc<dyn ProviderCredentialStatePort>,
     runtime_policy: Arc<dyn ProviderRuntimePolicyPort>,
@@ -104,7 +175,7 @@ impl CodexCredentialRefreshService {
     pub fn new(
         repository: CodexCredentialRepository,
         refresher: Arc<dyn TokenRefresher>,
-        identity: Arc<dyn CodexAccountIdentityVerifier>,
+        _identity: Arc<dyn CodexAccountIdentityVerifier>,
         leases: Arc<dyn ProviderLeasePort>,
         credential_state: Arc<dyn ProviderCredentialStatePort>,
         runtime_policy: Arc<dyn ProviderRuntimePolicyPort>,
@@ -112,7 +183,6 @@ impl CodexCredentialRefreshService {
         Self {
             repository,
             refresher,
-            identity,
             leases,
             credential_state,
             runtime_policy,
@@ -145,7 +215,14 @@ impl CodexCredentialRefreshService {
         for (account_id, result) in refreshed {
             match result {
                 Ok(outcome) => outcomes.push(outcome),
-                Err(_) => outcomes.push(CodexCredentialRefreshOutcome::Failed { account_id }),
+                Err(error) => {
+                    tracing::warn!(
+                        account_id = %account_id,
+                        error = %error,
+                        "OpenAI OAuth refresh attempt failed"
+                    );
+                    outcomes.push(CodexCredentialRefreshOutcome::Failed { account_id });
+                }
             }
         }
         Ok(outcomes)
@@ -157,6 +234,19 @@ impl CodexCredentialRefreshService {
         policy: ProviderRefreshPolicy,
     ) -> Result<CodexCredentialRefreshOutcome, CodexCredentialRefreshError> {
         let account_id = due.account.id().to_string();
+        if refresh_recovery_window_exhausted(
+            due.account.access_token_expires_at(),
+            SystemTime::now(),
+        ) {
+            return self
+                .persist_terminal(
+                    &due.account,
+                    AccountAvailability::Expired,
+                    "refresh_recovery_window_exhausted",
+                    CodexCredentialRefreshOutcome::Invalidated { account_id },
+                )
+                .await;
+        }
         let capacity = self
             .leases
             .try_acquire(ProviderLeaseRequest::RefreshCapacity(
@@ -192,6 +282,7 @@ impl CodexCredentialRefreshService {
                 self.persist_terminal(
                     &due.account,
                     AccountAvailability::Expired,
+                    "refresh_invalid_grant",
                     CodexCredentialRefreshOutcome::Invalidated { account_id },
                 )
                 .await
@@ -200,6 +291,7 @@ impl CodexCredentialRefreshService {
                 self.persist_terminal(
                     &due.account,
                     AccountAvailability::Banned,
+                    "account_banned",
                     CodexCredentialRefreshOutcome::Banned { account_id },
                 )
                 .await
@@ -272,15 +364,9 @@ impl CodexCredentialRefreshService {
                     let Some(secret) = runtime.authentication.oauth() else {
                         unreachable!("OAuth runtime authentication was checked above")
                     };
-                    let Some(principal) = runtime.principal else {
-                        failures.push(CodexCredentialRefreshOutcome::Failed { account_id });
-                        continue;
-                    };
                     due.push(DueCodexCredential {
                         account,
                         secret: secret.clone(),
-                        principal,
-                        installation_id: runtime.installation_id,
                     });
                 }
                 Ok(_) | Err(_) => {
@@ -297,113 +383,47 @@ impl CodexCredentialRefreshService {
         tokens: TokenPair,
         policy: ProviderRefreshPolicy,
     ) -> Result<CodexCredentialRefreshOutcome, CodexCredentialRefreshError> {
-        if tokens.access_token.is_empty() || tokens.expires_in.is_zero() {
-            return Err(CodexCredentialRefreshError::InvalidRefreshResponse);
-        }
-        let refresh_token = tokens
-            .refresh_token
+        let TokenPair {
+            access_token,
+            refresh_token: rotated_refresh_token,
+            expires_in,
+        } = tokens;
+        let refresh_token = rotated_refresh_token
             .map(SecretString::from)
             .or(due.secret.refresh_token);
-        let account_id = due
-            .account
-            .upstream_account_id()
-            .ok_or(CodexCredentialRefreshError::InvalidRefreshResponse)?;
-        let expectation = CodexIdentityExpectation::current(
-            due.principal.oauth_subject,
-            due.principal.poid,
-            account_id.to_owned(),
-            due.account.upstream_user_id().to_owned(),
-            due.installation_id,
-        )
-        .map_err(|_| CodexCredentialRefreshError::InvalidRefreshResponse)?;
         let secret = CodexOAuthSecret {
-            access_token: SecretString::from(tokens.access_token),
+            access_token: SecretString::from(access_token),
             refresh_token,
             id_token: due.secret.id_token,
         };
-        let verification = match self.identity.verify(&secret, &expectation).await {
-            Ok(verification) => verification,
-            Err(CodexIdentityVerificationError::Rejected) => {
-                return self
-                    .persist_terminal(
-                        &due.account,
-                        AccountAvailability::Invalid,
-                        CodexCredentialRefreshOutcome::Invalidated {
-                            account_id: due.account.id().to_string(),
-                        },
-                    )
-                    .await;
-            }
-            Err(CodexIdentityVerificationError::Unavailable) => {
-                // JWKS/签名边界短暂不可用属瞬态：保留现有凭据、推进退避重试，
-                // 不终态失效账号。
-                return if self
-                    .defer_refresh(&due.account, "identity-unavailable")
-                    .await?
-                {
-                    Ok(CodexCredentialRefreshOutcome::Transient {
-                        account_id: due.account.id().to_string(),
-                    })
-                } else {
-                    Ok(CodexCredentialRefreshOutcome::Stale {
-                        account_id: due.account.id().to_string(),
-                    })
-                };
-            }
-        };
-        let profile = match verification {
-            CodexIdentityVerification::Complete(profile) => profile,
-            CodexIdentityVerification::SignedOnly(signed) => {
-                let attempt = self
-                    .credential_state
-                    .record_refresh_backoff(due.account.id(), REFRESH_BACKOFF_WINDOW)
-                    .await
-                    .unwrap_or(1);
-                let retry_at = provider_refresh_backoff_at(
-                    due.account.id(),
-                    SystemTime::now(),
-                    attempt,
-                    "identity-completion",
-                )?;
-                match self
-                    .repository
-                    .rotate_signed_secret(&due.account, secret, &signed, retry_at)
-                    .await
-                {
-                    Ok(_) => {
-                        return Ok(CodexCredentialRefreshOutcome::Transient {
-                            account_id: due.account.id().to_string(),
-                        });
-                    }
-                    Err(CredentialRepositoryError::RevisionConflict) => {
-                        return Ok(CodexCredentialRefreshOutcome::Stale {
-                            account_id: due.account.id().to_string(),
-                        });
-                    }
-                    Err(error) => return Err(error.into()),
-                }
-            }
-        };
+        record_oauth_recovery(
+            CodexOAuthRecoveryOperation::ScheduledRefresh,
+            Some(due.account.id().as_str()),
+            secret.access_token.expose_secret(),
+            secret
+                .refresh_token
+                .as_ref()
+                .map(ExposeSecret::expose_secret),
+        );
+        if secret.access_token.expose_secret().is_empty() || expires_in.is_zero() {
+            return Err(CodexCredentialRefreshError::InvalidRefreshResponse);
+        }
+        let observed_at = SystemTime::now();
+        let access_token_expires_at = observed_at
+            .checked_add(expires_in)
+            .ok_or(CodexCredentialRefreshError::InvalidRefreshResponse)?;
         let next_refresh_at = policy
-            .next_attempt_at(
-                due.account.id(),
-                profile
-                    .access_token_expires_at
-                    .map(SystemTime::from)
-                    .ok_or(CodexCredentialRefreshError::InvalidRefreshResponse)?,
-                SystemTime::now(),
-            )
+            .next_attempt_at(due.account.id(), access_token_expires_at, observed_at)
             .map(DateTime::<Utc>::from)
             .map_err(CodexCredentialRefreshError::from)?;
         let result = self
             .repository
-            .rotate_oauth_secret(RotateCodexCredential {
-                account_id: due.account.id().to_string(),
-                expected_credential_revision: due.account.revision().get(),
+            .rotate_refreshed_oauth_secret(
+                &due.account,
                 secret,
-                verified_account: profile,
-                next_refresh_at: Some(next_refresh_at),
-            })
+                access_token_expires_at,
+                SystemTime::from(next_refresh_at),
+            )
             .await;
         match result {
             Ok(revision) => {
@@ -412,6 +432,13 @@ impl CodexCredentialRefreshService {
                     .credential_state
                     .clear_refresh_backoff(due.account.id())
                     .await;
+                tracing::info!(
+                    account_id = %due.account.id(),
+                    credential_revision = revision.get(),
+                    access_token_expires_at = %DateTime::<Utc>::from(access_token_expires_at),
+                    next_refresh_at = %next_refresh_at,
+                    "OpenAI OAuth refresh succeeded"
+                );
                 Ok(CodexCredentialRefreshOutcome::Refreshed {
                     account_id: due.account.id().to_string(),
                     credential_revision: revision.get(),
@@ -430,6 +457,7 @@ impl CodexCredentialRefreshService {
         &self,
         account: &ProviderAccount,
         availability: AccountAvailability,
+        reason: &'static str,
         outcome: CodexCredentialRefreshOutcome,
     ) -> Result<CodexCredentialRefreshOutcome, CodexCredentialRefreshError> {
         match self.repository.load_runtime_credential(account).await {
@@ -443,10 +471,27 @@ impl CodexCredentialRefreshService {
         }
         match self
             .repository
-            .apply_state(account, availability, SystemTime::now())
+            .apply_state_with_message(
+                account,
+                availability,
+                SystemTime::now(),
+                Some(reason.to_owned()),
+            )
             .await
         {
-            Ok(()) => Ok(outcome),
+            Ok(()) => {
+                tracing::warn!(
+                    account_id = %account.id(),
+                    ?availability,
+                    reason,
+                    access_token_expires_at = ?account.access_token_expires_at()
+                        .map(DateTime::<Utc>::from),
+                    recovery_deadline = ?refresh_recovery_deadline(account.access_token_expires_at())
+                        .map(DateTime::<Utc>::from),
+                    "OpenAI OAuth refresh marked account terminal"
+                );
+                Ok(outcome)
+            }
             Err(CredentialRepositoryError::RevisionConflict) => {
                 Ok(CodexCredentialRefreshOutcome::Stale {
                     account_id: account.id().to_string(),
@@ -466,10 +511,24 @@ impl CodexCredentialRefreshService {
             .record_refresh_backoff(account.id(), REFRESH_BACKOFF_WINDOW)
             .await
             .unwrap_or(1);
-        let retry_at =
-            provider_refresh_backoff_at(account.id(), SystemTime::now(), attempt, reason)?;
+        let retry_at = oauth_refresh_retry_at(
+            account.id(),
+            account.access_token_expires_at(),
+            SystemTime::now(),
+            attempt,
+            reason,
+        )?;
         match self.repository.defer_refresh(account, retry_at).await {
-            Ok(_) => Ok(true),
+            Ok(_) => {
+                log_refresh_deferred(
+                    account.id(),
+                    account.access_token_expires_at(),
+                    attempt,
+                    reason,
+                    retry_at,
+                );
+                Ok(true)
+            }
             Err(CredentialRepositoryError::RevisionConflict) => Ok(false),
             Err(error) => Err(error.into()),
         }

backend/crates/providers/openai/src/credential/repository.rs

@@ -13,7 +13,6 @@ use gateway_core::routing::ProviderKind;
 use secrecy::ExposeSecret;
 use thiserror::Error;
 
-use super::identity::CodexSignedIdentity;
 use super::security::{CodexCredentialCodec, CodexCredentialDataError, CodexRuntimeCredential};
 use super::types::{
     CodexAccountProfile, CodexCredentialData, CodexOAuthSecret, RotateCodexCredential,
@@ -83,12 +82,15 @@ impl CodexCredentialRepository {
         cas_revision(self.store.compare_and_swap_credential(update).await?)
     }
 
-    /// usage 暂时不可用时，只在签名 principal 与现有绑定完全一致后保存已轮换 token。
-    pub async fn rotate_signed_secret(
+    /// 持久化成功 RT exchange 的 token，同时保留既有账号身份投影。
+    ///
+    /// Refresh endpoint 已经是这次 AT/RT 轮换的授权边界；这里仅以 revision
+    /// CAS 保护并发写入，不重新验证新 access token 的身份声明。
+    pub async fn rotate_refreshed_oauth_secret(
         &self,
         account: &ProviderAccount,
         secret: CodexOAuthSecret,
-        signed: &CodexSignedIdentity,
+        access_token_expires_at: SystemTime,
         next_refresh_at: SystemTime,
     ) -> Result<CredentialRevision, CredentialRepositoryError> {
         let current = self
@@ -102,17 +104,6 @@ impl CodexCredentialRepository {
         let oauth = data
             .oauth_mut()
             .ok_or(CredentialRepositoryError::InvalidCredentialData)?;
-        if oauth.principal.oauth_subject != signed.oauth_subject()
-            || oauth.principal.poid.as_deref() != signed.poid()
-            || signed
-                .claimed_account_id()
-                .is_some_and(|value| account.upstream_account_id() != Some(value))
-            || signed
-                .claimed_user_id()
-                .is_some_and(|value| account.upstream_user_id() != value)
-        {
-            return Err(CredentialRepositoryError::IdentityMismatch);
-        }
         oauth.access_token = secret.access_token.expose_secret().to_owned();
         oauth.refresh_token = secret
             .refresh_token
@@ -129,7 +120,7 @@ impl CodexCredentialRepository {
             unchanged_profile(account),
             credential,
             secret.refresh_token.is_some(),
-            Some(SystemTime::from(signed.access_token_expires_at())),
+            Some(access_token_expires_at),
             Some(next_refresh_at),
         )
         .map_err(|_| CredentialRepositoryError::InvalidCredentialData)?;
@@ -234,16 +225,33 @@ impl CodexCredentialRepository {
         account: &ProviderAccount,
         availability: AccountAvailability,
         observed_at: SystemTime,
+    ) -> Result<(), CredentialRepositoryError> {
+        self.apply_state_with_message(account, availability, observed_at, None)
+            .await
+    }
+
+    /// 带上游错误描述的状态写入；上游 message 优先，终态缺失 message 时写稳定兜底，
+    /// 恢复 `Ready` 时清空。
+    pub async fn apply_state_with_message(
+        &self,
+        account: &ProviderAccount,
+        availability: AccountAvailability,
+        observed_at: SystemTime,
+        message: Option<String>,
     ) -> Result<(), CredentialRepositoryError> {
         if !account.enabled() {
             return Ok(());
         }
+        let message = message
+            .filter(|value| !value.trim().is_empty())
+            .or_else(|| availability.fallback_error_reason().map(str::to_owned));
         self.store
             .apply_state_change(AccountStateChange {
                 account_id: account.id().clone(),
                 expected_revision: account.revision(),
                 availability,
                 observed_at,
+                message,
             })
             .await?;
         Ok(())

backend/crates/providers/openai/src/credential/selector.rs

@@ -25,11 +25,12 @@ use super::agent_identity::CodexAgentIdentityTaskService;
 use super::catalog::CodexCredentialCatalogService;
 use super::cookie::CodexCookiePolicy;
 use super::quota::CodexCredentialQuotaService;
+use super::refresh::refresh_recovery_deadline;
 use super::repository::{CodexCredentialRepository, CredentialRepositoryError};
 use super::security::CodexRuntimeAuthentication;
 use super::types::{
-    CODEX_AUTHENTICATION_KIND_AGENT_IDENTITY, CodexCookie, CodexCookieCaptureOutcome,
-    RuntimeCodexCookie,
+    CODEX_AUTHENTICATION_KIND_AGENT_IDENTITY, CODEX_AUTHENTICATION_KIND_OAUTH, CodexCookie,
+    CodexCookieCaptureOutcome, RuntimeCodexCookie,
 };
 
 const CLOUDFLARE_RECOVERY_STALE_AFTER: Duration = Duration::from_secs(60 * 60);
@@ -73,6 +74,21 @@ pub enum CodexAccountFailure {
     CloudflarePathBlocked,
 }
 
+impl CodexAccountFailure {
+    /// 上游未提供可展示错误文本时，保留已分类的稳定原因。
+    const fn fallback_error_reason(self) -> Option<&'static str> {
+        match self {
+            Self::CredentialExpired => Some("upstream_credential_expired"),
+            Self::IdentityVerificationRequired => Some("upstream_identity_verification_required"),
+            Self::Banned => Some("upstream_account_banned"),
+            Self::QuotaExhausted => Some("upstream_quota_exhausted"),
+            Self::UsageLimitExhausted { .. } => Some("upstream_usage_limit_exhausted"),
+            Self::CloudflarePathBlocked => Some("upstream_cloudflare_path_blocked"),
+            Self::RateLimited { .. } | Self::CloudflareChallenge { .. } => None,
+        }
+    }
+}
+
 #[derive(Debug, Clone, Copy)]
 struct RiskRecoveryState {
     challenge_count: u32,
@@ -294,6 +310,24 @@ impl CodexCredentialSelector {
         if !diagnostic {
             self.quota.prepare_scheduling(&accounts).await;
         }
+        // 429 临时限流（Redis 冷却）中的账号即使额度正常也不参与调度，
+        // 到期后冷却过期自动恢复。诊断请求（连接测试）绕过冷却，需要真实探测。
+        let accounts = if diagnostic {
+            accounts
+        } else {
+            let mut retained = Vec::with_capacity(accounts.len());
+            for account in accounts {
+                let cooling = self
+                    .quota
+                    .rate_limited_until(account.id(), account.revision())
+                    .await
+                    .unwrap_or(None);
+                if cooling.is_none() {
+                    retained.push(account);
+                }
+            }
+            retained
+        };
         let account_ids = accounts
             .iter()
             .map(|account| account.id().clone())
@@ -327,46 +361,6 @@ impl CodexCredentialSelector {
                 AccountCandidate { account, signals }
             })
             .collect::<Vec<_>>();
-        // skip_exhausted 语义：true 时排除任一窗口触顶（limit_reached）的账号，
-        // 仅由上游 429 兜底；false 时保留全部候选（含 QuotaExhausted 的放宽投影）。
-        // 保留原始账号供 revision-fenced 凭据加载，用影子候选承载资格放宽。
-        let scheduling_candidates = if self.skip_exhausted {
-            candidates
-                .iter()
-                .filter(|candidate| !candidate.signals.quota_limit_reached)
-                .cloned()
-                .collect::<Vec<_>>()
-        } else {
-            candidates
-                .iter()
-                .cloned()
-                .map(|mut candidate| {
-                    candidate.account = quota_exhausted_scheduling_projection(&candidate.account);
-                    candidate
-                })
-                .collect::<Vec<_>>()
-        };
-        let scheduling_candidates = scheduling_candidates.as_slice();
-        let mut affinity = if diagnostic {
-            AffinitySelection::default()
-        } else {
-            self.resolve_session_affinity(
-                request.session_affinity_key,
-                &candidates,
-                SystemTime::now(),
-            )
-            .await
-        };
-        let cyber_policy_scope = self
-            .prepare_cyber_policy_scope(cyber_policy_session_key)
-            .await;
-        let mut excluded = request.attempt.excluded_accounts().clone();
-        if let Some(state) = cyber_policy_scope
-            .as_ref()
-            .and_then(|scope| scope.state.as_ref())
-        {
-            excluded.extend(state.excluded_accounts().iter().cloned());
-        }
         let continuation_account = match request.attempt.continuation_attempt() {
             ContinuationAttempt::Native => request
                 .attempt
@@ -388,7 +382,74 @@ impl CodexCredentialSelector {
         {
             return Err(CredentialSelectionError::NoEligibleCredential);
         }
-        let pinned_account = required_account.or(continuation_account);
+        let pinned_account = required_account.or_else(|| continuation_account.clone());
+        let preserve_native_continuation_state = matches!(
+            request.attempt.continuation_attempt(),
+            ContinuationAttempt::Native
+        ) && continuation_account.is_some();
+        // skip_exhausted 语义：true 时排除任一窗口触顶（limit_reached）的账号，
+        // 仅由上游 429 兜底；false 时保留全部候选（含 QuotaExhausted 的放宽投影）。
+        // 保留原始账号供 revision-fenced 凭据加载，用影子候选承载资格放宽。
+        let scheduling_candidates = if diagnostic {
+            // 连接测试必须实际探测指定账号；额度快照只影响常规调度。
+            candidates.clone()
+        } else if self.skip_exhausted {
+            candidates
+                .iter()
+                .filter(|candidate| !candidate.signals.quota_limit_reached)
+                .cloned()
+                .collect::<Vec<_>>()
+        } else {
+            candidates
+                .iter()
+                .cloned()
+                .map(|mut candidate| {
+                    // 已命中 previous_response_id 的 native continuation 先检查原账号。
+                    // 即使常规调度允许探测额度耗尽账号，也必须保留原账号真实状态，
+                    // 由 Coordinator 据此切换到可携带重放，而非伪装为 Ready。
+                    if !preserve_native_continuation_state
+                        || pinned_account.as_ref() != Some(candidate.account.id())
+                    {
+                        candidate.account =
+                            quota_exhausted_scheduling_projection(&candidate.account);
+                    }
+                    candidate
+                })
+                .collect::<Vec<_>>()
+        };
+        let scheduling_candidates = scheduling_candidates.as_slice();
+        let mut affinity = if diagnostic {
+            AffinitySelection::default()
+        } else {
+            self.resolve_session_affinity(
+                request.session_affinity_key,
+                &candidates,
+                SystemTime::now(),
+            )
+            .await
+        };
+        let cyber_policy_scope = self
+            .prepare_cyber_policy_scope(cyber_policy_session_key)
+            .await;
+        let mut excluded = request.attempt.excluded_accounts().clone();
+        if !diagnostic
+            && preserve_native_continuation_state
+            && let Some(pinned) = pinned_account.as_ref()
+            && candidates.iter().any(|candidate| {
+                candidate.account.id() == pinned && candidate.signals.quota_limit_reached
+            })
+        {
+            // `skip_exhausted = false` 只允许普通新请求探测可疑账号；已绑定的
+            // native response handle 先将真实触顶状态交给 Coordinator，随后才可
+            // 通过 ReplayAny 按当前调度策略换号。
+            excluded.insert(pinned.clone());
+        }
+        if let Some(state) = cyber_policy_scope
+            .as_ref()
+            .and_then(|scope| scope.state.as_ref())
+        {
+            excluded.extend(state.excluded_accounts().iter().cloned());
+        }
         if let Some(required) = pinned_account.as_ref() {
             for candidate in &candidates {
                 if candidate.account.id() != required {
@@ -685,26 +746,53 @@ impl CodexCredentialSelector {
         &self,
         account: &ProviderAccount,
         failure: CodexAccountFailure,
+        message: Option<String>,
     ) -> Result<(), CredentialSelectionError> {
         let now = SystemTime::now();
+        let message = message
+            .filter(|value| !value.trim().is_empty())
+            .or_else(|| failure.fallback_error_reason().map(str::to_owned));
         match failure {
             CodexAccountFailure::CredentialExpired => {
-                self.apply_state(account, AccountAvailability::Expired, now)
+                if account.authentication_kind() == CODEX_AUTHENTICATION_KIND_OAUTH
+                    && account.has_refresh_token()
+                    && account
+                        .access_token_expires_at()
+                        .is_some_and(|expires_at| expires_at <= now)
+                    && refresh_recovery_deadline(account.access_token_expires_at())
+                        .is_some_and(|deadline| deadline > now)
+                {
+                    tracing::info!(
+                        account_id = %account.id(),
+                        access_token_expires_at = ?account.access_token_expires_at()
+                            .map(chrono::DateTime::<chrono::Utc>::from),
+                        recovery_deadline = ?refresh_recovery_deadline(account.access_token_expires_at())
+                            .map(chrono::DateTime::<chrono::Utc>::from),
+                        "OpenAI access token expired; retaining account for bounded OAuth refresh recovery"
+                    );
+                    return Ok(());
+                }
+                self.apply_state_with_message(account, AccountAvailability::Expired, now, message)
                     .await
             }
             CodexAccountFailure::IdentityVerificationRequired => {
-                self.apply_state(account, AccountAvailability::Invalid, now)
+                self.apply_state_with_message(account, AccountAvailability::Invalid, now, message)
                     .await
             }
             CodexAccountFailure::Banned => {
-                self.apply_state(account, AccountAvailability::Banned, now)
+                self.apply_state_with_message(account, AccountAvailability::Banned, now, message)
                     .await
             }
             // 402：真额度耗尽，由 quota worker 按窗口 reset 自动恢复。
             CodexAccountFailure::QuotaExhausted
             | CodexAccountFailure::UsageLimitExhausted { .. } => {
-                self.apply_state(account, AccountAvailability::QuotaExhausted, now)
-                    .await
+                self.apply_state_with_message(
+                    account,
+                    AccountAvailability::QuotaExhausted,
+                    now,
+                    message,
+                )
+                .await
             }
             // 429：临时限流只写 quota，availability 不改变。
             CodexAccountFailure::RateLimited { retry_after } => {
@@ -727,8 +815,13 @@ impl CodexCredentialSelector {
             CodexAccountFailure::CloudflarePathBlocked => {
                 let blocked = self.record_cloudflare_path_block(account.id(), now);
                 if blocked >= CLOUDFLARE_PATH_BLOCK_THRESHOLD {
-                    self.apply_state(account, AccountAvailability::Invalid, now)
-                        .await?;
+                    self.apply_state_with_message(
+                        account,
+                        AccountAvailability::Invalid,
+                        now,
+                        message,
+                    )
+                    .await?;
                 }
                 self.apply_cookie_recovery(account, CookieRecovery::Clear)
                     .await?;
@@ -737,14 +830,15 @@ impl CodexCredentialSelector {
         }
     }
 
-    async fn apply_state(
+    async fn apply_state_with_message(
         &self,
         account: &ProviderAccount,
         availability: AccountAvailability,
         observed_at: SystemTime,
+        message: Option<String>,
     ) -> Result<(), CredentialSelectionError> {
         self.repository
-            .apply_state(account, availability, observed_at)
+            .apply_state_with_message(account, availability, observed_at, message)
             .await?;
         Ok(())
     }

backend/crates/providers/openai/src/credential/token_client.rs

@@ -76,10 +76,11 @@ impl fmt::Debug for AuthorizationCodeGrant {
     }
 }
 
-/// 官方 token endpoint 返回、尚待 OIDC/JWKS 绑定校验的 token set。
+/// 官方 token endpoint 完成协议校验后返回的 token set。
 pub struct AuthorizationTokenSet {
     pub secret: crate::credential::CodexOAuthSecret,
     pub id_token: SecretString,
+    pub expires_in: Duration,
 }
 
 impl fmt::Debug for AuthorizationTokenSet {
@@ -88,6 +89,7 @@ impl fmt::Debug for AuthorizationTokenSet {
             .debug_struct("AuthorizationTokenSet")
             .field("secret", &"[REDACTED]")
             .field("id_token", &"[REDACTED]")
+            .field("expires_in", &self.expires_in)
             .finish()
     }
 }
@@ -256,7 +258,7 @@ impl AuthorizationCodeExchanger for OpenAiTokenClient {
             || tokens
                 .refresh_token
                 .as_deref()
-                .is_some_and(|token| token.is_empty())
+                .is_some_and(|token| token.trim().is_empty())
         {
             return Err(AuthorizationCodeExchangeError::Rejected);
         }
@@ -267,6 +269,7 @@ impl AuthorizationCodeExchanger for OpenAiTokenClient {
                 id_token: None,
             },
             id_token: SecretString::from(tokens.id_token),
+            expires_in: Duration::from_secs(tokens.expires_in),
         })
     }
 }
@@ -302,7 +305,11 @@ fn parse_token_pair(body: &[u8]) -> Result<TokenPair, ()> {
     }
     Ok(TokenPair {
         access_token: tokens.access_token,
-        refresh_token: tokens.refresh_token,
+        // OAuth 刷新响应可能省略未变更的 RT。将空字段同样视为省略，调用方
+        // 才能保留当前仍可使用的 token。
+        refresh_token: tokens
+            .refresh_token
+            .filter(|refresh_token| !refresh_token.trim().is_empty()),
         expires_in: Duration::from_secs(tokens.expires_in),
     })
 }

backend/crates/providers/openai/src/credential/types.rs

@@ -29,7 +29,7 @@ impl fmt::Debug for CodexOAuthSecret {
     }
 }
 
-/// 已由官方 JWT/OIDC 与认证 usage 响应共同确认的完整账号投影。
+/// OAuth credential 持久化时使用的账号投影。
 #[derive(Clone)]
 pub struct CodexAccountProfile {
     pub email: Option<String>,

backend/crates/providers/openai/src/lib.rs

@@ -82,7 +82,7 @@ pub async fn initialize(
         profile.clone(),
         Arc::new(
             OfficialCodexCliReleaseTransport::new()
-                .map_err(|_| OpenAiInitializeError::DesktopRelease)?,
+                .map_err(|_| OpenAiInitializeError::CliRelease)?,
         ),
     ));
     let repository = CodexCredentialRepository::new(Arc::clone(&accounts));
@@ -111,6 +111,7 @@ pub async fn initialize(
         http.clone(),
         config.base_url().to_owned(),
         Arc::clone(&agent_identity),
+        ports.cooldowns(),
     ));
     let quota_skip_exhausted = config.quota_skip_exhausted();
     let selector = Arc::new(CodexCredentialSelector::new(
@@ -142,6 +143,7 @@ pub async fn initialize(
         .with_session_identity(session_identity),
     );
 
+    tracing::info!(target: "openai_oauth_recovery", "OpenAI OAuth recovery records use normal application logs");
     let token_client = Arc::new(
         credential::token_client::openai_token_client(config.token_client_config())
             .map_err(|_| OpenAiInitializeError::TokenClient)?,
@@ -155,13 +157,14 @@ pub async fn initialize(
         ReqwestCodexAuthenticatedAccountSource::new(profile.clone(), config.base_url().to_owned())
             .map_err(|_| OpenAiInitializeError::Identity)?,
     );
-    let identity: Arc<dyn CodexAccountIdentityVerifier> =
-        Arc::new(CodexAccountIdentityService::new(signed, account_source));
+    let identity: Arc<dyn CodexAccountIdentityVerifier> = Arc::new(
+        CodexAccountIdentityService::new(signed, Arc::clone(&account_source)),
+    );
     let refresher: Arc<dyn TokenRefresher> = token_client.clone();
     let exchanger: Arc<dyn AuthorizationCodeExchanger> = token_client;
     let credential_admin = Arc::new(CodexCredentialAdminService::new(
         Arc::clone(&refresher),
-        Arc::clone(&identity),
+        Arc::clone(&account_source),
         Arc::clone(&leases),
         Arc::clone(&runtime_policy),
     ));
@@ -181,7 +184,7 @@ pub async fn initialize(
         CodexOAuthAdminService::new(
             pending,
             exchanger,
-            Arc::clone(&identity),
+            account_source,
             Arc::clone(&accounts),
             Arc::clone(&runtime_policy),
             CodexCredentialAdmin,
@@ -267,6 +270,8 @@ pub enum OpenAiInitializeError {
     Refresh,
     #[error("OpenAI Desktop release service could not initialize")]
     DesktopRelease,
+    #[error("OpenAI CLI release service could not initialize")]
+    CliRelease,
     #[error("OpenAI worker plan is invalid")]
     Worker,
 }

backend/crates/providers/openai/src/provider.rs

@@ -1697,9 +1697,13 @@ fn map_selection_error(error: CredentialSelectionError) -> ProviderError {
 struct MappedProviderFailure {
     error: ProviderError,
     account_failure: Option<CodexAccountFailure>,
+    /// 原始上游错误描述，随 availability 状态一并持久化。
+    error_message: Option<String>,
     cyber_policy_failure: bool,
     set_cookie_headers: Vec<String>,
     rate_limit_headers: Vec<(String, String)>,
+    /// 结构化 usage-limit 错误给出的窗口 reset（Unix 秒）。
+    confirmed_exhaustion_reset_at: Option<i64>,
     observation: Option<ProviderResponseObservation>,
     capture_response_cookies: bool,
 }
@@ -1709,9 +1713,11 @@ impl MappedProviderFailure {
         Self {
             error,
             account_failure: None,
+            error_message: None,
             cyber_policy_failure: false,
             set_cookie_headers: Vec::new(),
             rate_limit_headers: Vec::new(),
+            confirmed_exhaustion_reset_at: None,
             observation: None,
             capture_response_cookies: false,
         }
@@ -1869,6 +1875,18 @@ async fn apply_failure(
         failure.account_failure,
         Some(CodexAccountFailure::QuotaExhausted | CodexAccountFailure::UsageLimitExhausted { .. })
     );
+    if needs_authoritative_quota_refresh
+        && let Err(error) = context
+            .quota
+            .record_confirmed_exhaustion(account, failure.confirmed_exhaustion_reset_at)
+            .await
+    {
+        tracing::warn!(
+            account_id = %account.id(),
+            error = %error,
+            "Failed to project confirmed OpenAI quota exhaustion"
+        );
+    }
     if failure.cyber_policy_failure {
         context
             .selector
@@ -1882,7 +1900,7 @@ async fn apply_failure(
             .await;
         if let Err(error) = context
             .selector
-            .record_failure(account, account_failure)
+            .record_failure(account, account_failure, failure.error_message.clone())
             .await
         {
             tracing::warn!(
@@ -2162,9 +2180,11 @@ fn map_upstream_failure(
     MappedProviderFailure {
         error,
         account_failure: account_failure(category, failure.retry_after_seconds),
+        error_message: failure.client_message,
         cyber_policy_failure,
         set_cookie_headers: failure.set_cookie_headers,
         rate_limit_headers: failure.rate_limit_headers,
+        confirmed_exhaustion_reset_at: failure.usage_limit_resets_at,
         observation,
         capture_response_cookies: !matches!(
             category,
@@ -2415,11 +2435,10 @@ impl ScheduledTask for OpenAiOAuthRefreshTask {
             if context.cancellation().is_cancelled() {
                 return Ok(());
             }
-            let outcomes = self
-                .service
-                .refresh_due()
-                .await
-                .map_err(|_| WorkerTaskError::safe("OpenAI OAuth refresh failed"))?;
+            let outcomes = self.service.refresh_due().await.map_err(|error| {
+                tracing::error!(error = %error, "OpenAI OAuth refresh cycle failed");
+                WorkerTaskError::safe("OpenAI OAuth refresh failed")
+            })?;
             let mut refreshed = 0_u64;
             let mut invalidated = 0_u64;
             let mut banned = 0_u64;
@@ -2448,6 +2467,18 @@ impl ScheduledTask for OpenAiOAuthRefreshTask {
                     }
                 }
             }
+            if !outcomes.is_empty() {
+                tracing::info!(
+                    refreshed,
+                    invalidated,
+                    banned,
+                    transient,
+                    lease_unavailable,
+                    stale,
+                    failed,
+                    "OpenAI OAuth refresh cycle completed"
+                );
+            }
             if transient > 0 || failed > 0 {
                 tracing::warn!(
                     refreshed,

backend/crates/providers/openai/src/transport/client_sse.rs

@@ -87,8 +87,13 @@ impl CodexBackendClient {
         let headers = self.request_headers_for_http_response(upstream_request, context)?;
         let headers_started_at = Instant::now();
         // 与官方 Codex app-server 一致：/v1/responses 请求体默认 zstd 压缩。
+        // Codex 上游只交付 SSE；即使下游请求 `stream: false`，也要上游流式执行，
+        // 再由 API 层收集 canonical events 并返回完整 JSON。不能把下游的传输偏好
+        // 直接透传给 Codex，否则上游会以 400 拒绝非流式请求。
+        let mut upstream_body = upstream_request.body().clone();
+        upstream_body.insert("stream".to_owned(), serde_json::Value::Bool(true));
         let body =
-            serde_json::to_vec(&upstream_request).map_err(CodexClientError::RequestBodyEncode)?;
+            serde_json::to_vec(&upstream_body).map_err(CodexClientError::RequestBodyEncode)?;
         let body = zstd::stream::encode_all(std::io::Cursor::new(body), 3)
             .map_err(CodexClientError::RequestCompression)?;
         let response = self

backend/crates/providers/openai/src/transport/diagnostics.rs

@@ -3,6 +3,7 @@
 use super::client::CodexClientVisibleUpstreamResponse;
 use super::protocol::responses::ResponsesSseFailure;
 use base64::{Engine as _, engine::general_purpose::STANDARD};
+use chrono::{DateTime, Utc};
 use reqwest::{StatusCode, header::HeaderMap};
 use serde_json::Value;
 
@@ -113,6 +114,8 @@ pub struct CodexUpstreamFailure {
     pub(crate) client_error_type: Option<String>,
     pub(crate) client_response: Option<Box<CodexClientVisibleUpstreamResponse>>,
     pub(crate) identity_error_code: Option<String>,
+    /// 官方 `usage_limit_reached` 错误体携带的窗口重置时间（Unix 秒）。
+    pub(crate) usage_limit_resets_at: Option<i64>,
     pub(crate) retry_after_seconds: Option<u64>,
     pub(crate) request_id: Option<String>,
     pub(crate) set_cookie_headers: Vec<String>,
@@ -139,6 +142,7 @@ impl CodexUpstreamFailure {
     ) -> Self {
         let fields = ParsedUpstreamError::from_body(body);
         let category = classify_upstream_failure(
+            UpstreamFailureSource::HttpResponse,
             Some(status),
             body,
             &fields,
@@ -153,6 +157,9 @@ impl CodexUpstreamFailure {
             client_error_type: fields.error_type,
             client_response: client_response.cloned().map(Box::new),
             identity_error_code: diagnostics.identity_error_code.clone(),
+            usage_limit_resets_at: (category == CodexFailureCategory::UsageLimitExhausted)
+                .then_some(fields.resets_at)
+                .flatten(),
             retry_after_seconds,
             request_id: diagnostics.request_id.clone(),
             set_cookie_headers: set_cookie_headers.to_vec(),
@@ -174,11 +181,13 @@ impl CodexUpstreamFailure {
             error_type: failure.upstream_type.clone(),
             message: failure.message.clone(),
             client_message: Some(failure.message.clone()),
+            resets_at: None,
         };
         let status = failure
             .explicit_status_code
             .and_then(|code| StatusCode::from_u16(code).ok());
         let category = classify_upstream_failure(
+            UpstreamFailureSource::SseFailure,
             status,
             "",
             &fields,
@@ -193,6 +202,7 @@ impl CodexUpstreamFailure {
             client_error_type: fields.error_type,
             client_response: None,
             identity_error_code: diagnostics.identity_error_code.clone(),
+            usage_limit_resets_at: None,
             retry_after_seconds: failure.retry_after_seconds,
             request_id: diagnostics.request_id.clone(),
             set_cookie_headers: set_cookie_headers.to_vec(),
@@ -246,6 +256,7 @@ struct ParsedUpstreamError {
     error_type: Option<String>,
     message: String,
     client_message: Option<String>,
+    resets_at: Option<i64>,
 }
 
 impl ParsedUpstreamError {
@@ -256,6 +267,7 @@ impl ParsedUpstreamError {
                 error_type: None,
                 message: body.to_owned(),
                 client_message: None,
+                resets_at: None,
             };
         };
         let error = value
@@ -282,15 +294,26 @@ impl ParsedUpstreamError {
             .clone()
             .or_else(|| error.as_str().and_then(non_empty_owned))
             .unwrap_or_else(|| body.to_owned());
+        let resets_at = error
+            .get("resets_at")
+            .and_then(Value::as_i64)
+            .filter(|seconds| DateTime::<Utc>::from_timestamp(*seconds, 0).is_some());
         Self {
             code,
             error_type,
             message,
             client_message,
+            resets_at,
         }
     }
 }
 
+#[derive(Clone, Copy)]
+enum UpstreamFailureSource {
+    HttpResponse,
+    SseFailure,
+}
+
 #[derive(Debug, Clone, Default, PartialEq, Eq)]
 pub struct CodexUpstreamDiagnostics {
     pub status_code: Option<u16>,
@@ -363,6 +386,7 @@ fn decode_identity_error_code(encoded: &str) -> Option<String> {
 }
 
 fn classify_upstream_failure(
+    source: UpstreamFailureSource,
     status: Option<StatusCode>,
     body: &str,
     fields: &ParsedUpstreamError,
@@ -376,6 +400,19 @@ fn classify_upstream_failure(
     let message = fields.message.to_ascii_lowercase();
     let body = body.to_ascii_lowercase();
 
+    // 与官方 Codex HTTP/WS 路径一致：429 只有结构化
+    // `error.type=usage_limit_reached` 才能确认额度窗口耗尽；其余 429 都是临时限流。
+    // SSE `response.failed` 的结构化字段形态不同，保留其 code/type 语义单独分类。
+    if matches!(source, UpstreamFailureSource::HttpResponse)
+        && status == Some(StatusCode::TOO_MANY_REQUESTS)
+    {
+        return if error_type == "usage_limit_reached" {
+            CodexFailureCategory::UsageLimitExhausted
+        } else {
+            CodexFailureCategory::RateLimited
+        };
+    }
+
     if [code.as_str(), message.as_str(), body.as_str()]
         .into_iter()
         .any(is_model_unsupported)

backend/crates/providers/openai/src/transport/profile.rs

@@ -14,21 +14,22 @@ use url::Url;
 pub const CODEX_DESKTOP_APPCAST_URL: &str =
     "https://persistent.oaistatic.com/codex-app-prod/appcast.xml";
 pub const CODEX_CLI_RELEASE_URL: &str = "https://registry.npmjs.org/@openai%2Fcodex/latest";
-/// 官方制品画像检查周期。
+/// 官方发布元数据检查周期。
 pub const APPCAST_POLL_INTERVAL: Duration = Duration::from_secs(24 * 60 * 60);
 
 const APPCAST_TIMEOUT: Duration = Duration::from_secs(30);
 const MAX_APPCAST_BYTES: usize = 1024 * 1024;
+const MAX_CLI_RELEASE_BYTES: usize = 64 * 1024;
 
 /// Codex Desktop 上游请求身份。
 ///
-/// 启动配置提供经源码审计的 Core、运行环境和初始 Desktop 版本。官方 appcast
-/// 检查成功后，会仅同步 Desktop 版本和构建号。
+/// 启动配置提供经源码审计的 Core、运行环境和 Desktop 启动版本。npm CLI metadata
+/// 与官方 appcast 检查成功后，分别同步 Core 版本和 Desktop 版本及构建号。
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct CodexWireProfile {
     /// `originator` 请求头及 User-Agent 产品名。
     pub originator: String,
-    /// bundled Codex Core 版本；同时用于 `/codex/models?client_version=`。
+    /// npm `@openai/codex` 启动版本；用于 `/codex/models?client_version=` 与 UA。
     pub codex_version: String,
     /// Desktop 应用版本，用于 app-server `clientInfo.version` 对应的 UA 后缀。
     pub desktop_version: String,
@@ -47,16 +48,18 @@ pub struct CodexWireProfile {
 }
 
 impl CodexWireProfile {
-    /// 按 Codex Core 的官方格式生成最终 User-Agent。
+    /// 按 bundled Core app-server 的官方格式生成最终 User-Agent。
     pub fn user_agent(&self) -> String {
         format!(
-            "{}/{} ({} {}; {}) {}",
+            "{}/{} ({} {}; {}) {} ({}; {})",
             self.originator,
             self.codex_version,
             self.os_type,
             self.os_version,
             self.arch,
             self.terminal,
+            self.originator,
+            self.desktop_version,
         )
     }
 }
@@ -96,7 +99,7 @@ impl CodexWireProfileState {
         profile.desktop_build = desktop_build.to_owned();
     }
 
-    /// 使用官方 npm metadata 发布的 Codex CLI 版本更新运行时 Core 画像。
+    /// 使用官方 npm metadata 发布的 Codex CLI 版本更新运行时请求画像。
     pub fn update_cli_release(&self, codex_version: &str) {
         let mut profile = self
             .profile
@@ -143,12 +146,12 @@ impl CodexCliReleaseTransport for OfficialCodexCliReleaseTransport {
             }
             if response
                 .content_length()
-                .is_some_and(|size| size > 64 * 1024)
+                .is_some_and(|size| size > MAX_CLI_RELEASE_BYTES as u64)
             {
                 return Err(CodexCliReleaseError::ResponseTooLarge);
             }
             let bytes = response.bytes().await?;
-            if bytes.len() > 64 * 1024 {
+            if bytes.len() > MAX_CLI_RELEASE_BYTES {
                 return Err(CodexCliReleaseError::ResponseTooLarge);
             }
             let document: serde_json::Value = serde_json::from_slice(&bytes)

backend/crates/providers/openai/src/transport/protocol/responses.rs

@@ -606,6 +606,26 @@ impl CodexResponsesRequest {
         )
     }
 
+    /// 返回会话亲和所用的子代理区分值。
+    ///
+    /// Codex 原生请求在 turn metadata 中声明 `subagent_kind`；网关的
+    /// `/responses/review` 等入口则通过 `client_metadata.x-openai-subagent`
+    /// 传递给上游。该可选值仅参与会话亲和键派生。
+    pub fn subagent_kind(&self) -> Option<String> {
+        self.semantics()
+            .subagent_kind
+            .filter(|value| !value.trim().is_empty())
+            .or_else(|| {
+                self.client_metadata()
+                    .and_then(Value::as_object)
+                    .and_then(|metadata| metadata.get("x-openai-subagent"))
+                    .and_then(Value::as_str)
+                    .map(str::trim)
+                    .filter(|value| !value.is_empty())
+                    .map(str::to_owned)
+            })
+    }
+
     /// 设置 / 合并 client metadata。
     pub fn set_client_metadata(&mut self, client_metadata: Option<Value>) {
         match client_metadata {

backend/crates/providers/openai/src/transport/request.rs

@@ -10,6 +10,8 @@ use sha2::{Digest, Sha256};
 use crate::transport::protocol::responses::CodexResponsesRequest;
 
 const PASSTHROUGH_HEADERS_CONTEXT_KEY: &str = "opaque_request_headers";
+const THREAD_SPAWN_SUBAGENT_KIND: &str = "thread_spawn";
+const THREAD_SPAWN_CONVERSATION_PREFIX: &str = "thread-spawn:";
 
 const CROSS_ACCOUNT_IDENTITY_KEYS: &[&str] = &[
     "authorization",
@@ -160,28 +162,52 @@ fn string_value(value: Option<&Value>) -> Option<String> {
 pub(crate) fn derive_conversation_anchor(
     request: &CodexResponsesRequest,
 ) -> Option<(&'static str, String)> {
+    thread_spawn_conversation_anchor(request).or_else(|| {
+        request
+            .client_session_id
+            .as_deref()
+            .map(|value| ("session", value.to_owned()))
+            .or_else(|| {
+                request
+                    .client_conversation_id
+                    .as_deref()
+                    .map(|value| ("conversation", value.to_owned()))
+            })
+            .or_else(|| {
+                request
+                    .client_thread_id
+                    .as_deref()
+                    .map(|value| ("thread", value.to_owned()))
+            })
+            .or_else(|| {
+                request
+                    .prompt_cache_key()
+                    .map(|value| ("prompt-cache", value.to_owned()))
+            })
+            .or_else(|| derive_stable_conversation_key(request).map(|value| ("request", value)))
+    })
+}
+
+/// A `thread_spawn` shares its parent session but is a separate child execution.
+/// Its child thread must therefore own the affinity and WebSocket pool identity.
+fn thread_spawn_conversation_anchor(
+    request: &CodexResponsesRequest,
+) -> Option<(&'static str, String)> {
+    if request.subagent_kind().as_deref() != Some(THREAD_SPAWN_SUBAGENT_KIND) {
+        return None;
+    }
+
     request
-        .client_session_id
+        .client_thread_id
         .as_deref()
-        .map(|value| ("session", value.to_owned()))
-        .or_else(|| {
-            request
-                .client_conversation_id
-                .as_deref()
-                .map(|value| ("conversation", value.to_owned()))
+        .map(str::trim)
+        .filter(|thread_id| !thread_id.is_empty())
+        .map(|thread_id| {
+            (
+                "subagent-thread",
+                format!("{THREAD_SPAWN_CONVERSATION_PREFIX}{thread_id}"),
+            )
         })
-        .or_else(|| {
-            request
-                .client_thread_id
-                .as_deref()
-                .map(|value| ("thread", value.to_owned()))
-        })
-        .or_else(|| {
-            request
-                .prompt_cache_key()
-                .map(|value| ("prompt-cache", value.to_owned()))
-        })
-        .or_else(|| derive_stable_conversation_key(request).map(|value| ("request", value)))
 }
 
 const LEADING_SYSTEM_REMINDER_OPEN: &str = "<system-reminder>";

backend/crates/providers/openai/tests/admin.rs

@@ -82,6 +82,22 @@ async fn openai_bundle_exposes_one_core_provider_and_drains_worker_contributions
         panic!("Desktop release worker must be scheduled");
     };
     assert_eq!(schedule.interval(), APPCAST_POLL_INTERVAL);
+    let cli_release_worker = contributions
+        .iter()
+        .find_map(|contribution| match contribution {
+            WorkerContribution::Registration(registration)
+                if registration.id.owner() == "openai-cli-release" =>
+            {
+                Some(registration)
+            }
+            WorkerContribution::Registration(_) | WorkerContribution::Disabled { .. } => None,
+        })
+        .expect("CLI release worker");
+    assert_eq!(cli_release_worker.id.kind(), WorkerKind::QuotaCatalogHealth);
+    let WorkerRunnable::Scheduled { schedule, .. } = &cli_release_worker.runnable else {
+        panic!("CLI release worker must be scheduled");
+    };
+    assert_eq!(schedule.interval(), APPCAST_POLL_INTERVAL);
     assert!(contributions.iter().any(|contribution| {
         matches!(
             contribution,
@@ -101,13 +117,21 @@ async fn openai_admin_provider_exposes_live_wire_profile_and_validated_billing()
         .expect("OpenAI bundle");
     let admin = bundle.admin_provider();
     let profile = admin.dashboard_wire_profile().expect("wire profile");
+    assert_eq!(profile.version, "0.102.0");
+    assert_eq!(profile.build, None);
+    assert_eq!(profile.target.os_type, "Mac OS");
+    assert_eq!(profile.target.os_version, "15.5.0");
+    assert_eq!(
+        profile.user_agent,
+        "Codex Desktop/0.102.0 (Mac OS 15.5.0; arm64) xterm-256color (Codex Desktop; 1.2026.190)"
+    );
     assert_eq!(
         profile
             .attributes
             .iter()
-            .find(|attribute| attribute.label == "Codex Core")
+            .find(|attribute| attribute.label == "客户端标识")
             .map(|attribute| attribute.value.as_str()),
-        Some("0.102.0")
+        Some("Codex Desktop; 1.2026.190")
     );
     assert_eq!(
         profile.release.as_ref().map(|release| release.status),
@@ -428,7 +452,8 @@ async fn openai_admin_provider_projects_cached_quota_models_and_canonical_export
 }
 
 #[tokio::test]
-async fn openai_admin_provider_projects_codex_additional_limit_as_the_primary_quota() {
+async fn openai_admin_provider_projects_official_codex_quota_and_independent_buckets_with_chinese_labels()
+ {
     let store = Arc::new(MemoryAccountStore::default());
     store
         .seed_oauth_credential(ImportCodexOAuthCredential {
@@ -444,11 +469,15 @@ async fn openai_admin_provider_projects_codex_additional_limit_as_the_primary_qu
         .account("acct_admin_canonical_quota")
         .expect("stored account");
     let raw = json!({
+        "active_limit": "premium",
         "rate_limit": {
             "primary_window": {
                 "used_percent": 91,
                 "reset_at": 1_900_000_000,
                 "limit_window_seconds": 2_592_000
+            },
+            "secondary_window": {
+                "used_percent": 88
             }
         },
         "additional_rate_limits": [{
@@ -459,6 +488,19 @@ async fn openai_admin_provider_projects_codex_additional_limit_as_the_primary_qu
                     "used_percent": 2,
                     "reset_at": 1_900_000_000,
                     "limit_window_seconds": 2_592_000
+                },
+                "secondary_window": {
+                    "used_percent": 0
+                }
+            }
+        }, {
+            "limit_name": "code_review",
+            "metered_feature": "code_review",
+            "rate_limit": {
+                "primary_window": {
+                    "used_percent": 12,
+                    "reset_at": 1_900_000_000,
+                    "limit_window_seconds": 604_800
                 }
             }
         }]
@@ -499,10 +541,31 @@ async fn openai_admin_provider_projects_codex_additional_limit_as_the_primary_qu
         .collect::<Vec<_>>();
 
     assert_eq!(monthly.len(), 1);
-    // limit_name 优先显示，不再猜成「月限额」。
-    assert_eq!(monthly[0].label, "custom codex label");
-    assert_eq!(monthly[0].source.as_deref(), Some("core"));
-    assert_eq!(monthly[0].used_percent, Some(2.0));
+    assert!(monthly.iter().any(|window| {
+        window.label == "月额度"
+            && window.source.as_deref() == Some("codex")
+            && window.used_percent == Some(91.0)
+    }));
+    assert!(
+        !quota
+            .windows
+            .iter()
+            .any(|window| window.key.starts_with("additional-0-codex")),
+        "the additional codex alias should not become a second display bucket"
+    );
+    let secondary = quota
+        .windows
+        .iter()
+        .find(|window| window.key == "core-secondary")
+        .expect("core secondary quota");
+    assert_eq!(secondary.label, "次级额度");
+    assert_eq!(secondary.used_percent, Some(88.0));
+    let review = quota
+        .windows
+        .iter()
+        .find(|window| window.source.as_deref() == Some("code_review"))
+        .expect("code review quota");
+    assert_eq!(review.label, "代码审查 · 周额度");
 }
 
 #[tokio::test]
@@ -544,6 +607,7 @@ async fn openai_admin_projects_confirmed_quota_exhaustion_as_full_without_mutati
         .expect("persist raw quota");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: CoreAccountAvailability::QuotaExhausted,
@@ -584,6 +648,7 @@ async fn openai_admin_projects_confirmed_quota_exhaustion_as_full_without_mutati
 
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: CoreAccountAvailability::Ready,
@@ -711,6 +776,7 @@ fn account_record(account: &ProviderAccount) -> AccountRecord {
         enabled: account.enabled(),
         availability: AdminAccountAvailability::Ready,
         availability_observed_at: now,
+        last_error_message: None,
         quota_observed_at: None,
         created_at: now,
         updated_at: now,
@@ -729,7 +795,7 @@ fn valid_config() -> TestOpenAiConfig {
         codex_version: "0.102.0".to_owned(),
         desktop_version: "1.2026.190".to_owned(),
         desktop_build: "19012345678".to_owned(),
-        os_type: "macOS".to_owned(),
+        os_type: "Mac OS".to_owned(),
         os_version: "15.5.0".to_owned(),
         arch: "arm64".to_owned(),
         terminal: "xterm-256color".to_owned(),

backend/crates/providers/openai/tests/config.rs

@@ -13,7 +13,7 @@ fn openai_config_builds_the_audited_wire_profile() {
 
     assert_eq!(
         profile.user_agent(),
-        "Codex Desktop/0.102.0 (macOS 15.5.0; arm64) xterm-256color"
+        "Codex Desktop/0.102.0 (Mac OS 15.5.0; arm64) xterm-256color (Codex Desktop; 1.2026.190)"
     );
     assert_eq!(profile.desktop_build, "19012345678");
 }
@@ -124,6 +124,10 @@ fn openai_config_defaults_to_the_provider_owned_operating_values() {
             "https://auth.openai.com/oauth/token",
         )
     );
+    assert_eq!(
+        config.wire_profile_state().snapshot().user_agent(),
+        "Codex Desktop/0.146.0 (Mac OS 15.7.1; arm64) unknown (Codex Desktop; 26.727.51351)"
+    );
 }
 
 fn valid_config() -> OpenAiConfig {
@@ -133,7 +137,7 @@ fn valid_config() -> OpenAiConfig {
         codex_version: "0.102.0".to_owned(),
         desktop_version: "1.2026.190".to_owned(),
         desktop_build: "19012345678".to_owned(),
-        os_type: "macOS".to_owned(),
+        os_type: "Mac OS".to_owned(),
         os_version: "15.5.0".to_owned(),
         arch: "arm64".to_owned(),
         terminal: "xterm-256color".to_owned(),

backend/crates/providers/openai/tests/credential/admin.rs

@@ -1,3 +1,4 @@
+use std::collections::VecDeque;
 use std::sync::atomic::{AtomicUsize, Ordering};
 use std::sync::{Arc, Mutex};
 use std::time::Duration;
@@ -20,14 +21,15 @@ use gateway_core::provider_ports::{
 use gateway_core::routing::ProviderKind;
 use provider_openai::credential::token_client::{RefreshFailure, TokenPair, TokenRefresher};
 use provider_openai::credential::{
-    CODEX_AUTHENTICATION_KIND_AGENT_IDENTITY, CodexAccountIdentityVerifier,
-    CodexAgentIdentityAuthMode, CodexAgentIdentityCredentialData, CodexCredentialAdmin,
-    CodexCredentialAdminError, CodexCredentialAdminService, CodexCredentialCodec,
-    CodexIdentityExpectation, CodexIdentityVerification, CodexIdentityVerificationError,
+    CODEX_AUTHENTICATION_KIND_AGENT_IDENTITY, CodexAgentIdentityAuthMode,
+    CodexAgentIdentityCredentialData, CodexAuthenticatedAccount, CodexAuthenticatedAccountSource,
+    CodexCredentialAdmin, CodexCredentialAdminError, CodexCredentialAdminService,
+    CodexCredentialCodec, CodexIdentityExpectation, CodexIdentityVerificationError,
     ExportManagedCodexCredential, ImportCodexOAuthCredential, RotateManagedCodexCredential,
 };
-use secrecy::{ExposeSecret, SecretString};
+use secrecy::ExposeSecret;
 
+use super::{OAuthRecoveryLogCapture, lock_oauth_recovery_log_capture};
 use crate::support::{MemoryAccountStore, codex_account, profile, runtime_policy, secret};
 
 fn import(id: &str, token: &str) -> ImportCodexOAuthCredential {
@@ -315,19 +317,25 @@ fn cpr_export_preserves_agent_identity_material_without_oauth_fields() {
 #[tokio::test]
 async fn cpr_export_and_import_support_mixed_oauth_and_agent_identity_accounts() {
     let (agent, _) = agent_export_item("acct_mixed_agent");
+    let access_token = "opaque-mixed-oauth";
     let value = CodexCredentialAdmin
-        .format_cpr_export(vec![
-            export_item("acct_mixed_oauth", "token-acct_mixed_oauth"),
-            agent,
-        ])
+        .format_cpr_export(vec![export_item("acct_mixed_oauth", access_token), agent])
         .expect("mixed export")
         .into_json()
         .expect("serialize mixed export");
 
-    let prepared = import_service(unused_import_refresher())
-        .prepare_import_document(value)
-        .await
-        .expect("mixed CPR import");
+    let prepared = import_service_with_source(
+        unused_import_refresher(),
+        account_source([authenticated_account(
+            "chatgpt-mixed-oauth",
+            "user-mixed-oauth",
+            None,
+            None,
+        )]),
+    )
+    .prepare_import_document(value)
+    .await
+    .expect("mixed CPR import");
     assert_eq!(prepared.accounts().len(), 2);
     assert!(prepared.accounts().iter().any(|account| {
         account.account.authentication_kind() == CODEX_AUTHENTICATION_KIND_AGENT_IDENTITY
@@ -360,34 +368,59 @@ impl TokenRefresher for ManualRefresher {
     }
 }
 
-struct ManualVerifier;
+struct ManualAccountSource {
+    outcomes: Mutex<VecDeque<Result<CodexAuthenticatedAccount, CodexIdentityVerificationError>>>,
+    seen: Mutex<Vec<String>>,
+    expectations: Mutex<Vec<CodexIdentityExpectation>>,
+}
 
 #[async_trait]
-impl CodexAccountIdentityVerifier for ManualVerifier {
-    async fn verify(
+impl CodexAuthenticatedAccountSource for ManualAccountSource {
+    async fn fetch(
         &self,
         secret: &provider_openai::credential::CodexOAuthSecret,
-        _expectation: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        if secret.access_token.expose_secret() != "refreshed-access" {
-            return Err(CodexIdentityVerificationError::Rejected);
-        }
-        Ok(CodexIdentityVerification::Complete(profile(
-            "chatgpt-acct_manual_refresh",
-        )))
+        expectation: &CodexIdentityExpectation,
+    ) -> Result<CodexAuthenticatedAccount, CodexIdentityVerificationError> {
+        self.seen
+            .lock()
+            .expect("seen account-source access tokens")
+            .push(secret.access_token.expose_secret().to_owned());
+        self.expectations
+            .lock()
+            .expect("seen account-source expectations")
+            .push(expectation.clone());
+        self.outcomes
+            .lock()
+            .expect("account source outcomes")
+            .pop_front()
+            .unwrap_or(Err(CodexIdentityVerificationError::Unavailable))
     }
+}
 
-    async fn verify_authorization(
-        &self,
-        _secret: &provider_openai::credential::CodexOAuthSecret,
-        _id_token: &SecretString,
-        _expected_nonce: &SecretString,
-        _expectation: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        Err(CodexIdentityVerificationError::Rejected)
+fn authenticated_account(
+    account_id: &str,
+    user_id: &str,
+    email: Option<&str>,
+    plan_type: Option<&str>,
+) -> CodexAuthenticatedAccount {
+    CodexAuthenticatedAccount {
+        chatgpt_account_id: account_id.to_owned(),
+        chatgpt_user_id: user_id.to_owned(),
+        email: email.map(str::to_owned),
+        plan_type: plan_type.map(str::to_owned),
     }
 }
 
+fn account_source(
+    outcomes: impl IntoIterator<Item = CodexAuthenticatedAccount>,
+) -> Arc<ManualAccountSource> {
+    Arc::new(ManualAccountSource {
+        outcomes: Mutex::new(outcomes.into_iter().map(Ok).collect()),
+        seen: Mutex::new(Vec::new()),
+        expectations: Mutex::new(Vec::new()),
+    })
+}
+
 struct DropCountGuard(Arc<AtomicUsize>);
 
 impl Drop for DropCountGuard {
@@ -473,7 +506,7 @@ async fn manual_refresh_fixture(
     });
     let service = CodexCredentialAdminService::new(
         refresher.clone(),
-        Arc::new(ManualVerifier),
+        account_source([]),
         leases.clone(),
         runtime_policy(),
     );
@@ -489,7 +522,8 @@ fn refreshed_tokens() -> TokenPair {
 }
 
 #[tokio::test]
-async fn manual_refresh_prepares_revision_fenced_rotation_without_store_mutation() {
+async fn manual_refresh_prepares_rotation_without_identity_verification() {
+    let _capture_lock = lock_oauth_recovery_log_capture().await;
     let (store, refresher, leases, service) =
         manual_refresh_fixture(Ok(refreshed_tokens()), true).await;
     let account_id = ProviderAccountId::new("acct_manual_refresh").expect("account id");
@@ -497,10 +531,14 @@ async fn manual_refresh_prepares_revision_fenced_rotation_without_store_mutation
         .load_current_credential(&account_id)
         .await
         .expect("current credential");
+    let recovery_log = OAuthRecoveryLogCapture::default();
+    let dispatch = recovery_log.dispatch();
+    let guard = tracing::dispatcher::set_default(&dispatch);
     let prepared = service
         .manual_refresh(current)
         .await
         .expect("manual refresh");
+    drop(guard);
 
     assert_eq!(prepared.credential.account_id(), &account_id);
     assert_eq!(prepared.credential.expected_revision().get(), 1);
@@ -539,6 +577,12 @@ async fn manual_refresh_prepares_revision_fenced_rotation_without_store_mutation
         ["rt-old-access"]
     );
     assert_eq!(leases.drops.load(Ordering::SeqCst), 0);
+    let record = recovery_log.recovery_record();
+    assert_eq!(record["fields"]["operation"], "manual_refresh");
+    assert_eq!(record["fields"]["account_id"], "acct_manual_refresh");
+    assert_eq!(record["fields"]["access_token"], "refreshed-access");
+    assert_eq!(record["fields"]["refresh_token"], "rotated-refresh");
+    assert_eq!(record["fields"]["has_refresh_token"], true);
     drop(prepared);
     assert_eq!(leases.drops.load(Ordering::SeqCst), 2);
 }
@@ -569,54 +613,19 @@ async fn manual_refresh_missing_lease_fails_before_exchange() {
     );
 }
 
-struct ImportVerifier;
-
-#[async_trait]
-impl CodexAccountIdentityVerifier for ImportVerifier {
-    async fn verify(
-        &self,
-        secret: &provider_openai::credential::CodexOAuthSecret,
-        expectation: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        let suffix = secret
-            .access_token
-            .expose_secret()
-            .strip_prefix("token-")
-            .ok_or(CodexIdentityVerificationError::Rejected)?;
-        let mut profile = profile(&format!("chatgpt-{suffix}"));
-        if let Some(user) = suffix.strip_prefix("shared-") {
-            profile.chatgpt_account_id = "chatgpt-shared".to_owned();
-            profile.chatgpt_user_id = format!("user-{user}");
-        }
-        if expectation
-            .chatgpt_account_id()
-            .is_some_and(|expected| expected != profile.chatgpt_account_id)
-            || expectation
-                .chatgpt_user_id()
-                .is_some_and(|expected| expected != profile.chatgpt_user_id)
-        {
-            return Err(CodexIdentityVerificationError::Rejected);
-        }
-        Ok(CodexIdentityVerification::Complete(profile))
-    }
-
-    async fn verify_authorization(
-        &self,
-        _secret: &provider_openai::credential::CodexOAuthSecret,
-        _id_token: &SecretString,
-        _expected_nonce: &SecretString,
-        _expectation: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        Err(CodexIdentityVerificationError::Rejected)
-    }
-}
-
 pub(super) fn import_service(
     refresher: Arc<ManualRefresher>,
+) -> provider_openai::credential::CodexCredentialAdminService {
+    import_service_with_source(refresher, account_source([]))
+}
+
+fn import_service_with_source(
+    refresher: Arc<ManualRefresher>,
+    source: Arc<ManualAccountSource>,
 ) -> provider_openai::credential::CodexCredentialAdminService {
     CodexCredentialAdminService::new(
         refresher,
-        Arc::new(ImportVerifier),
+        source,
         Arc::new(ManualLeases {
             requests: Mutex::new(Vec::new()),
             drops: Arc::new(AtomicUsize::new(0)),
@@ -634,21 +643,25 @@ pub(super) fn unused_import_refresher() -> Arc<ManualRefresher> {
 }
 
 #[tokio::test]
-async fn oauth_import_uses_only_tokens_and_verified_identity() {
+async fn oauth_import_accepts_opaque_access_token_and_completes_profile_from_account_source() {
     let refresher = unused_import_refresher();
-    let prepared = import_service(refresher.clone())
+    let access_token = "opaque-cpr-access";
+    let source = account_source([authenticated_account(
+        "chatgpt-cpr",
+        "user-cpr",
+        Some("remote-cpr@example.com"),
+        Some("team"),
+    )]);
+    let prepared = import_service_with_source(refresher.clone(), Arc::clone(&source))
         .prepare_import_document(serde_json::json!({
             "sourceFormat": "cpr",
             "accounts": [{
                 "id": "acct_cpr_import",
-                "email": "cpr@example.com",
-                "accountId": "chatgpt-cpr",
-                "userId": "user-chatgpt-cpr",
+                "email": "untrusted-cpr@example.com",
                 "label": "CPR account",
-                "planType": "pro",
-                "token": "Bearer token-cpr",
+                "planType": "untrusted-plan",
+                "token": format!("Bearer {access_token}"),
                 "refreshToken": "refresh-cpr",
-                "accessTokenExpiresAt": "2100-01-01T00:00:00+00:00",
                 "status": "disabled",
                 "addedAt": "2026-07-18T10:47:01+08:00",
                 "updatedAt": "2026-07-19T11:00:00+08:00",
@@ -662,6 +675,11 @@ async fn oauth_import_uses_only_tokens_and_verified_identity() {
     assert!(account.account.id().as_str().starts_with("acct_"));
     assert!(account.account.enabled());
     assert_eq!(account.account.upstream_account_id(), Some("chatgpt-cpr"));
+    assert_eq!(account.account.upstream_user_id(), "user-cpr");
+    assert_eq!(account.account.name(), "CPR account");
+    assert_eq!(account.account.email(), Some("remote-cpr@example.com"));
+    assert_eq!(account.account.plan_type(), Some("team"));
+    assert_eq!(account.account.access_token_expires_at(), None);
     let runtime = CodexCredentialCodec::decode(&account.credential).expect("credential");
     assert_eq!(
         runtime
@@ -670,26 +688,181 @@ async fn oauth_import_uses_only_tokens_and_verified_identity() {
             .expect("OAuth credential")
             .access_token
             .expose_secret(),
-        "token-cpr"
+        access_token
     );
     assert!(refresher.seen.lock().expect("seen tokens").is_empty());
+    assert_eq!(
+        source
+            .seen
+            .lock()
+            .expect("seen account-source access tokens")
+            .as_slice(),
+        [access_token]
+    );
+    assert_eq!(
+        source
+            .expectations
+            .lock()
+            .expect("seen account-source expectations")
+            .as_slice(),
+        [CodexIdentityExpectation::default()]
+    );
+}
+
+#[tokio::test]
+async fn direct_import_logs_raw_tokens_before_account_profile_failure() {
+    let _capture_lock = lock_oauth_recovery_log_capture().await;
+    let recovery_log = OAuthRecoveryLogCapture::default();
+    let dispatch = recovery_log.dispatch();
+    let guard = tracing::dispatcher::set_default(&dispatch);
+    let error = import_service_with_source(unused_import_refresher(), account_source([]))
+        .prepare_import_document(serde_json::json!({
+            "sourceFormat": "cpr",
+            "accounts": [{
+                "token": "logged-direct-access",
+                "refreshToken": "logged-direct-refresh"
+            }]
+        }))
+        .await
+        .expect_err("profile source fails after direct import logging");
+    drop(guard);
+
+    assert_eq!(error, CodexCredentialAdminError::AccountProfileUnavailable);
+    let record = recovery_log.recovery_record();
+    assert_eq!(record["fields"]["operation"], "import_direct");
+    assert!(
+        record["fields"]["account_id"]
+            .as_str()
+            .is_some_and(|account_id| account_id.starts_with("acct_"))
+    );
+    assert_eq!(record["fields"]["access_token"], "logged-direct-access");
+    assert_eq!(record["fields"]["refresh_token"], "logged-direct-refresh");
+    assert_eq!(record["fields"]["has_refresh_token"], true);
+}
+
+#[tokio::test]
+async fn refresh_token_only_import_persists_exchanged_tokens_and_completes_profile() {
+    let refreshed_access = "opaque-refreshed-import";
+    let refresher = Arc::new(ManualRefresher {
+        outcome: Mutex::new(Some(Ok(TokenPair {
+            access_token: refreshed_access.to_owned(),
+            refresh_token: Some("rotated-import-refresh".to_owned()),
+            expires_in: Duration::from_secs(3_600),
+        }))),
+        seen: Mutex::new(Vec::new()),
+    });
+    let source = account_source([authenticated_account(
+        "chatgpt-refreshed-import",
+        "user-refreshed-import",
+        Some("refreshed@example.com"),
+        Some("plus"),
+    )]);
+
+    let prepared = import_service_with_source(Arc::clone(&refresher), Arc::clone(&source))
+        .prepare_import_document(serde_json::json!({
+            "refreshToken": "refresh-only"
+        }))
+        .await
+        .expect("refresh-token-only import");
+
+    assert_eq!(prepared.accounts().len(), 1);
+    let account = &prepared.accounts()[0];
+    assert_eq!(
+        account.account.upstream_account_id(),
+        Some("chatgpt-refreshed-import")
+    );
+    assert_eq!(account.account.upstream_user_id(), "user-refreshed-import");
+    assert_eq!(account.account.email(), Some("refreshed@example.com"));
+    assert_eq!(account.account.plan_type(), Some("plus"));
+    let runtime = CodexCredentialCodec::decode(&account.credential).expect("credential");
+    let oauth = runtime.authentication.oauth().expect("OAuth credential");
+    assert_eq!(oauth.access_token.expose_secret(), refreshed_access);
+    assert_eq!(
+        oauth
+            .refresh_token
+            .as_ref()
+            .expect("rotated RT")
+            .expose_secret(),
+        "rotated-import-refresh"
+    );
+    assert_eq!(
+        refresher
+            .seen
+            .lock()
+            .expect("seen refresh tokens")
+            .as_slice(),
+        ["refresh-only"]
+    );
+    assert_eq!(
+        source
+            .seen
+            .lock()
+            .expect("seen account-source access tokens")
+            .as_slice(),
+        [refreshed_access]
+    );
+    assert_eq!(
+        source
+            .expectations
+            .lock()
+            .expect("seen account-source expectations")
+            .as_slice(),
+        [CodexIdentityExpectation::default()]
+    );
+}
+
+#[tokio::test]
+async fn refresh_token_only_import_logs_fallback_refresh_token_before_account_profile_failure() {
+    let _capture_lock = lock_oauth_recovery_log_capture().await;
+    let recovery_log = OAuthRecoveryLogCapture::default();
+    let refresher = Arc::new(ManualRefresher {
+        outcome: Mutex::new(Some(Ok(TokenPair {
+            access_token: "logged-import-access".to_owned(),
+            refresh_token: None,
+            expires_in: Duration::from_secs(3_600),
+        }))),
+        seen: Mutex::new(Vec::new()),
+    });
+    let dispatch = recovery_log.dispatch();
+    let guard = tracing::dispatcher::set_default(&dispatch);
+    let error = import_service_with_source(refresher, account_source([]))
+        .prepare_import_document(serde_json::json!({
+            "refreshToken": "logged-import-original-refresh"
+        }))
+        .await
+        .expect_err("profile source fails after refresh-token import logging");
+    drop(guard);
+
+    assert_eq!(error, CodexCredentialAdminError::AccountProfileUnavailable);
+    let record = recovery_log.recovery_record();
+    assert_eq!(record["fields"]["operation"], "import_refresh_token");
+    assert_eq!(record["fields"]["access_token"], "logged-import-access");
+    assert_eq!(
+        record["fields"]["refresh_token"],
+        "logged-import-original-refresh"
+    );
+    assert_eq!(record["fields"]["has_refresh_token"], true);
 }
 
 #[tokio::test]
 async fn cpr_batch_allows_distinct_users_in_the_same_workspace() {
-    let prepared = import_service(unused_import_refresher())
+    let source = account_source([
+        authenticated_account("chatgpt-shared", "user-a", None, None),
+        authenticated_account("chatgpt-shared", "user-b", None, None),
+    ]);
+    let prepared = import_service_with_source(unused_import_refresher(), Arc::clone(&source))
         .prepare_import_document(serde_json::json!({
             "sourceFormat": "cpr",
             "accounts": [{
                 "id": "acct_shared_a",
                 "accountId": "chatgpt-shared",
                 "userId": "user-a",
-                "token": "token-shared-a"
+                "token": "opaque-shared-a"
             }, {
                 "id": "acct_shared_b",
                 "accountId": "chatgpt-shared",
                 "userId": "user-b",
-                "token": "token-shared-b"
+                "token": "opaque-shared-b"
             }]
         }))
         .await
@@ -703,19 +876,23 @@ async fn cpr_batch_allows_distinct_users_in_the_same_workspace() {
             .all(|account| { account.account.upstream_account_id() == Some("chatgpt-shared") })
     );
 
-    let duplicate = import_service(unused_import_refresher())
+    let duplicate_source = account_source([
+        authenticated_account("chatgpt-shared", "user-a", None, None),
+        authenticated_account("chatgpt-shared", "user-a", None, None),
+    ]);
+    let duplicate = import_service_with_source(unused_import_refresher(), duplicate_source)
         .prepare_import_document(serde_json::json!({
             "sourceFormat": "cpr",
             "accounts": [{
                 "id": "acct_duplicate_a",
                 "accountId": "chatgpt-shared",
                 "userId": "user-a",
-                "token": "token-shared-a"
+                "token": "opaque-duplicate-a"
             }, {
                 "id": "acct_duplicate_b",
                 "accountId": "chatgpt-shared",
                 "userId": "user-a",
-                "token": "token-shared-a"
+                "token": "opaque-duplicate-b"
             }]
         }))
         .await
@@ -724,8 +901,14 @@ async fn cpr_batch_allows_distinct_users_in_the_same_workspace() {
 }
 
 #[tokio::test]
-async fn oauth_import_extracts_tokens_without_parsing_source_projections() {
-    let bundle = import_service(unused_import_refresher())
+async fn oauth_import_completes_profiles_from_account_source() {
+    let bundle_source = account_source([authenticated_account(
+        "chatgpt-bundle",
+        "user-bundle",
+        Some("remote-bundle@example.com"),
+        Some("team"),
+    )]);
+    let bundle = import_service_with_source(unused_import_refresher(), Arc::clone(&bundle_source))
         .prepare_import_document(serde_json::json!({
             "exported_at": "2026-07-03T15:46:38.717Z",
             "proxies": [],
@@ -734,10 +917,12 @@ async fn oauth_import_extracts_tokens_without_parsing_source_projections() {
                 "platform": "openai",
                 "type": "oauth",
                 "credentials": {
-                    "access_token": "token-bundle",
+                    "access_token": "opaque-bundle",
                     "refresh_token": "refresh-bundle",
-                    "chatgpt_account_id": "chatgpt-bundle",
-                    "chatgpt_user_id": "user-chatgpt-bundle"
+                    "chatgpt_account_id": "source-bundle",
+                    "chatgpt_user_id": "source-user-bundle",
+                    "email": "bundle-source@example.com",
+                    "plan_type": "team"
                 },
                 "concurrency": 3,
                 "priority": 50
@@ -745,48 +930,85 @@ async fn oauth_import_extracts_tokens_without_parsing_source_projections() {
         }))
         .await
         .expect("credential bundle import");
-    let auth_document = import_service(unused_import_refresher())
-        .prepare_import_document(serde_json::json!({
-            "accounts": [{
-                "type": "openai",
-                "access_token": "token-auth-document",
-                "refresh_token": "refresh-auth-document",
-                "chatgpt_account_id": "chatgpt-auth-document",
-                "chatgpt_user_id": "user-chatgpt-auth-document",
-                "email": "auth-document@example.com",
-                "label": "Auth document"
-            }]
-        }))
-        .await
-        .expect("auth document import");
+    let auth_document_source = account_source([authenticated_account(
+        "chatgpt-auth-document",
+        "user-auth-document",
+        Some("remote-auth-document@example.com"),
+        Some("plus"),
+    )]);
+    let auth_document =
+        import_service_with_source(unused_import_refresher(), Arc::clone(&auth_document_source))
+            .prepare_import_document(serde_json::json!({
+                "accounts": [{
+                    "type": "openai",
+                    "access_token": "opaque-auth-document",
+                    "refresh_token": "refresh-auth-document",
+                    "chatgpt_account_id": "source-auth-document",
+                    "chatgpt_user_id": "source-user-auth-document",
+                    "email": "auth-document@example.com",
+                    "label": "Auth document",
+                    "planType": "plus"
+                }]
+            }))
+            .await
+            .expect("auth document import");
 
-    for (prepared, expected) in [
-        (&bundle, "chatgpt-bundle"),
-        (&auth_document, "chatgpt-auth-document"),
-    ] {
-        assert_eq!(prepared.accounts().len(), 1);
-        assert_eq!(
-            prepared.accounts()[0].account.upstream_account_id(),
-            Some(expected)
-        );
-        assert!(
-            prepared.accounts()[0]
-                .account
-                .id()
-                .as_str()
-                .starts_with("acct_")
-        );
-    }
+    let bundle_account = &bundle.accounts()[0].account;
+    assert_eq!(bundle.accounts().len(), 1);
+    assert_eq!(bundle_account.upstream_account_id(), Some("chatgpt-bundle"));
+    assert_eq!(bundle_account.upstream_user_id(), "user-bundle");
+    assert_eq!(bundle_account.name(), "bundle@example.com");
+    assert_eq!(bundle_account.email(), Some("remote-bundle@example.com"));
+    assert_eq!(bundle_account.plan_type(), Some("team"));
+
+    let auth_document_account = &auth_document.accounts()[0].account;
+    assert_eq!(auth_document.accounts().len(), 1);
+    assert_eq!(
+        auth_document_account.upstream_account_id(),
+        Some("chatgpt-auth-document")
+    );
+    assert_eq!(
+        auth_document_account.upstream_user_id(),
+        "user-auth-document"
+    );
+    assert_eq!(auth_document_account.name(), "Auth document");
+    assert_eq!(
+        auth_document_account.email(),
+        Some("remote-auth-document@example.com")
+    );
+    assert_eq!(auth_document_account.plan_type(), Some("plus"));
+    assert_eq!(
+        bundle_source
+            .seen
+            .lock()
+            .expect("seen account-source access tokens")
+            .as_slice(),
+        ["opaque-bundle"]
+    );
+    assert_eq!(
+        auth_document_source
+            .seen
+            .lock()
+            .expect("seen account-source access tokens")
+            .as_slice(),
+        ["opaque-auth-document"]
+    );
     assert!(!format!("{bundle:?} {auth_document:?}").contains("refresh-bundle"));
     assert!(!format!("{bundle:?} {auth_document:?}").contains("refresh-auth-document"));
 }
 
 #[tokio::test]
 async fn cliproxyapi_codex_auth_file_is_recognized_as_an_openai_auth_document() {
-    let prepared = import_service(unused_import_refresher())
+    let source = account_source([authenticated_account(
+        "chatgpt-cpa",
+        "user-cpa",
+        Some("remote-cpa@example.com"),
+        Some("pro"),
+    )]);
+    let prepared = import_service_with_source(unused_import_refresher(), Arc::clone(&source))
         .prepare_import_document(serde_json::json!({
             "type": "codex",
-            "access_token": "token-cpa",
+            "access_token": "opaque-cpa",
             "refresh_token": "refresh-cpa",
             "id_token": "id-cpa",
             "account_id": "chatgpt-cpa",
@@ -799,6 +1021,8 @@ async fn cliproxyapi_codex_auth_file_is_recognized_as_an_openai_auth_document()
     assert_eq!(prepared.accounts().len(), 1);
     let account = &prepared.accounts()[0].account;
     assert_eq!(account.upstream_account_id(), Some("chatgpt-cpa"));
-    assert_eq!(account.email(), Some("chatgpt-cpa@example.com"));
+    assert_eq!(account.upstream_user_id(), "user-cpa");
+    assert_eq!(account.email(), Some("remote-cpa@example.com"));
+    assert_eq!(account.plan_type(), Some("pro"));
     assert_eq!(account.authentication_kind(), "oauth");
 }

backend/crates/providers/openai/tests/credential/contract.rs

@@ -4,15 +4,21 @@ use std::sync::Arc;
 use std::time::{Duration, SystemTime};
 
 use futures::executor::block_on;
+use gateway_core::engine::continuation::{
+    ContinuationBinding, NativeContinuationPin, PreviousResponseId,
+};
 use gateway_core::engine::credential::{
     AccountAttemptFeedback, AccountAvailability, AccountFeedbackStats, AccountSelectionPolicy,
-    AccountStateChange, ProviderAccountId, ProviderAccountStore as _, RotationStrategy,
+    AccountStateChange, OpaqueProviderData, ProviderAccountId, ProviderAccountStore as _,
+    QuotaObservation, QuotaWriteOutcome, RotationStrategy,
 };
 use gateway_core::engine::{
     AccountAttemptContext, AttemptContext, CancellationToken, ModelRequestId, RequestAttemptContext,
 };
 use gateway_core::policy::ClientApiKeyId;
-use gateway_core::provider_ports::{ProviderSessionAffinityKey, ProviderSessionAffinityPort};
+use gateway_core::provider_ports::{
+    ProviderCooldownPort, ProviderSessionAffinityKey, ProviderSessionAffinityPort,
+};
 use gateway_core::routing::ProviderKind;
 use provider_openai::OFFICIAL_CODEX_BASE_URL;
 use provider_openai::credential::{
@@ -22,11 +28,12 @@ use provider_openai::credential::{
 };
 use provider_openai::transport::profile::{CodexWireProfile, CodexWireProfileState};
 use secrecy::ExposeSecret;
+use serde_json::json;
 use url::Url;
 
 use crate::support::{
-    MemoryAccountStore, MemorySessionAffinity, MemorySessionExclusions, TestLeaseCoordinator,
-    account_policy, agent_identity_service, catalog_cache, profile, secret,
+    MemoryAccountStore, MemoryCooldownPort, MemorySessionAffinity, MemorySessionExclusions,
+    TestLeaseCoordinator, account_policy, agent_identity_service, catalog_cache, profile, secret,
 };
 
 fn create_account(store: &Arc<MemoryAccountStore>, id: &str, token: &str) {
@@ -99,6 +106,7 @@ fn selector_with_affinity(
         session_affinity,
         Arc::new(AccountFeedbackStats::default()),
         true,
+        Arc::new(MemoryCooldownPort::new()),
     )
 }
 
@@ -108,6 +116,7 @@ fn selector_with_runtime(
     session_affinity: Arc<MemorySessionAffinity>,
     account_feedback: Arc<AccountFeedbackStats>,
     skip_exhausted: bool,
+    cooldowns: Arc<dyn ProviderCooldownPort>,
 ) -> CodexCredentialSelector {
     let profile = CodexWireProfileState::new(CodexWireProfile {
         originator: "codex_cli_rs".to_owned(),
@@ -136,6 +145,7 @@ fn selector_with_runtime(
         http,
         OFFICIAL_CODEX_BASE_URL.to_owned(),
         Arc::clone(&agent_identity),
+        cooldowns,
     ));
     CodexCredentialSelector::new(
         ProviderKind::new("openai").expect("provider"),
@@ -464,6 +474,7 @@ async fn selector_should_keep_a_schedulable_affinity_account_despite_soft_health
         affinity,
         Arc::clone(&account_feedback),
         true,
+        Arc::new(MemoryCooldownPort::new()),
     );
     for _ in 0..4 {
         account_feedback.report(
@@ -507,6 +518,7 @@ async fn selector_should_escape_a_quota_exhausted_affinity_account() {
     let first = store.account("acct_first").expect("first account");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: first.id().clone(),
             expected_revision: first.revision(),
             availability: AccountAvailability::QuotaExhausted,
@@ -663,8 +675,12 @@ fn credential_expired_failure_marks_unified_account_expired() {
             }),
         )
         .expect("select account");
-    block_on(selector.record_failure(lease.account(), CodexAccountFailure::CredentialExpired))
-        .expect("record credential expiry");
+    block_on(selector.record_failure(
+        lease.account(),
+        CodexAccountFailure::CredentialExpired,
+        None,
+    ))
+    .expect("record credential expiry");
     assert_eq!(
         store
             .account("acct_primary")
@@ -672,13 +688,58 @@ fn credential_expired_failure_marks_unified_account_expired() {
             .availability(),
         AccountAvailability::Expired
     );
+    assert_eq!(
+        store.last_error_message("acct_primary").as_deref(),
+        Some("upstream_credential_expired")
+    );
+}
+
+#[test]
+fn credential_expired_failure_keeps_expired_oauth_for_bounded_refresh_recovery() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let mut expired_profile = profile("chatgpt-acct_primary");
+    expired_profile.access_token_expires_at =
+        Some(chrono::Utc::now() - chrono::Duration::minutes(1));
+    block_on(store.seed_oauth_credential(ImportCodexOAuthCredential {
+        account_id: "acct_primary".to_owned(),
+        name: "acct_primary".to_owned(),
+        secret: secret("at-primary"),
+        verified_account: expired_profile,
+        next_refresh_at: Some(chrono::Utc::now() + chrono::Duration::minutes(10)),
+        enabled: true,
+    }));
+    let account = store.account("acct_primary").expect("account");
+    let selector = selector(&store, Arc::new(TestLeaseCoordinator::default()));
+
+    block_on(selector.record_failure(
+        &account,
+        CodexAccountFailure::CredentialExpired,
+        Some("token_expired".to_owned()),
+    ))
+    .expect("record credential expiry");
+
+    assert_eq!(
+        store
+            .account("acct_primary")
+            .expect("account retained for refresh")
+            .availability(),
+        AccountAvailability::Ready
+    );
 }
 
 #[test]
 fn rate_limited_failure_records_quota_window_without_changing_availability() {
     let store = Arc::new(MemoryAccountStore::default());
     create_account(&store, "acct_primary", "at-primary");
-    let selector = selector(&store, Arc::new(TestLeaseCoordinator::default()));
+    let cooldowns = Arc::new(MemoryCooldownPort::new());
+    let selector = selector_with_runtime(
+        &store,
+        Arc::new(TestLeaseCoordinator::default()),
+        Arc::new(MemorySessionAffinity::default()),
+        Arc::new(AccountFeedbackStats::default()),
+        true,
+        Arc::clone(&cooldowns) as Arc<dyn ProviderCooldownPort>,
+    );
     let attempt = attempt(BTreeSet::new());
     let lease =
         block_on(
@@ -697,36 +758,22 @@ fn rate_limited_failure_records_quota_window_without_changing_availability() {
         CodexAccountFailure::RateLimited {
             retry_after: Some(Duration::from_secs(30)),
         },
+        None,
     ))
     .expect("record rate-limit failure");
 
-    // 429 不改变 availability（限流不改变账号可用性），只把限流窗口合成进 quota JSON。
+    // 429 不改变 availability（限流不改变账号可用性），也不污染 quota JSON——
+    // 临时限流写入 Redis 冷却（到期自动解除），额度重置时间保持不变。
     let account = store.account("acct_primary").expect("account");
     assert_eq!(account.availability(), AccountAvailability::Ready);
-    let quota = store.quota_json("acct_primary").expect("quota");
-    assert_eq!(
-        quota
-            .pointer("/rate_limit/limit_reached")
-            .and_then(serde_json::Value::as_bool),
-        Some(true)
-    );
-    assert_eq!(
-        quota
-            .pointer("/rate_limit/primary_window/used_percent")
-            .and_then(serde_json::Value::as_f64),
-        Some(100.0)
-    );
     assert!(
-        quota
-            .pointer("/rate_limit/primary_window/reset_at")
-            .and_then(serde_json::Value::as_i64)
-            .is_some_and(|reset_at| {
-                reset_at
-                    > SystemTime::now()
-                        .duration_since(std::time::UNIX_EPOCH)
-                        .expect("clock")
-                        .as_secs() as i64
-            })
+        store.quota_json("acct_primary").is_none(),
+        "429 must not synthesize quota window"
+    );
+    let cooldown = block_on(cooldowns.read(account.id())).expect("read cooldown");
+    assert!(
+        cooldown.is_some_and(|cooling| cooling.until() > SystemTime::now()),
+        "429 must record a Redis cooldown expiring in the future"
     );
 }
 
@@ -743,6 +790,7 @@ async fn usage_limit_exhaustion_marks_quota_exhausted_without_usage_probe() {
             CodexAccountFailure::UsageLimitExhausted {
                 retry_after: Some(Duration::from_secs(30)),
             },
+            None,
         )
         .await
         .expect("record usage-limit exhaustion");
@@ -758,6 +806,7 @@ fn rate_limited_failure_does_not_downgrade_persisted_quota_exhaustion() {
     create_account(&store, "acct_primary", "at-primary");
     let account = store.account("acct_primary").expect("account");
     block_on(store.apply_state_change(AccountStateChange {
+        message: None,
         account_id: account.id().clone(),
         expected_revision: account.revision(),
         availability: AccountAvailability::QuotaExhausted,
@@ -771,6 +820,7 @@ fn rate_limited_failure_does_not_downgrade_persisted_quota_exhaustion() {
         CodexAccountFailure::RateLimited {
             retry_after: Some(Duration::from_secs(30)),
         },
+        None,
     ))
     .expect("record rate-limit failure");
 
@@ -789,13 +839,22 @@ fn rate_limited_failure_does_not_consult_stale_quota_snapshot() {
     let store = Arc::new(MemoryAccountStore::default());
     create_account(&store, "acct_primary", "at-primary");
     let account = store.account("acct_primary").expect("account");
-    let selector = selector(&store, Arc::new(TestLeaseCoordinator::default()));
+    let cooldowns = Arc::new(MemoryCooldownPort::new());
+    let selector = selector_with_runtime(
+        &store,
+        Arc::new(TestLeaseCoordinator::default()),
+        Arc::new(MemorySessionAffinity::default()),
+        Arc::new(AccountFeedbackStats::default()),
+        true,
+        Arc::clone(&cooldowns) as Arc<dyn ProviderCooldownPort>,
+    );
 
     block_on(selector.record_failure(
         &account,
         CodexAccountFailure::RateLimited {
             retry_after: Some(Duration::from_secs(30)),
         },
+        None,
     ))
     .expect("record rate-limit failure");
 
@@ -806,8 +865,13 @@ fn rate_limited_failure_does_not_consult_stale_quota_snapshot() {
             .availability(),
         AccountAvailability::Ready
     );
-    // 429 合成限流窗口需要读写 quota JSON。
-    assert!(store.quota_reads() >= 1);
+    // 429 临时限流写入 Redis 冷却，不读写 quota JSON。
+    assert_eq!(store.quota_reads(), 0);
+    let cooldown = block_on(cooldowns.read(account.id())).expect("read cooldown");
+    assert!(
+        cooldown.is_some_and(|cooling| cooling.until() > SystemTime::now()),
+        "429 must record a Redis cooldown"
+    );
 }
 
 #[test]
@@ -816,6 +880,7 @@ fn rate_limited_failure_does_not_overwrite_stale_authentication_state() {
     create_account(&store, "acct_primary", "at-primary");
     let account = store.account("acct_primary").expect("account");
     block_on(store.apply_state_change(AccountStateChange {
+        message: None,
         account_id: account.id().clone(),
         expected_revision: account.revision(),
         availability: AccountAvailability::Invalid,
@@ -830,6 +895,7 @@ fn rate_limited_failure_does_not_overwrite_stale_authentication_state() {
         CodexAccountFailure::RateLimited {
             retry_after: Some(Duration::from_secs(30)),
         },
+        None,
     ))
     .expect("record rate-limit failure");
 
@@ -854,6 +920,7 @@ fn successful_upstream_response_recovers_non_quota_terminal_states() {
         create_account(&store, "acct_primary", "at-primary");
         let account = store.account("acct_primary").expect("account");
         block_on(store.apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: stale,
@@ -892,12 +959,14 @@ fn rate_limited_failures_for_distinct_accounts_do_not_conflict() {
                 CodexAccountFailure::RateLimited {
                     retry_after: Some(Duration::from_secs(30)),
                 },
+                None,
             ),
             selector.record_failure(
                 &second,
                 CodexAccountFailure::RateLimited {
                     retry_after: Some(Duration::from_secs(30)),
                 },
+                None,
             ),
         );
         first_result.expect("record first account failure");
@@ -925,7 +994,7 @@ fn selector_keeps_a_quota_exhausted_account_eligible_when_configured_not_to_skip
     create_account(&store, "acct_primary", "at-primary");
     let strict_selector = selector(&store, Arc::new(TestLeaseCoordinator::default()));
     let account = store.account("acct_primary").expect("account");
-    block_on(strict_selector.record_failure(&account, CodexAccountFailure::QuotaExhausted))
+    block_on(strict_selector.record_failure(&account, CodexAccountFailure::QuotaExhausted, None))
         .expect("mark account exhausted");
 
     let selector = selector_with_runtime(
@@ -934,6 +1003,7 @@ fn selector_keeps_a_quota_exhausted_account_eligible_when_configured_not_to_skip
         Arc::new(MemorySessionAffinity::default()),
         Arc::new(AccountFeedbackStats::default()),
         false,
+        Arc::new(MemoryCooldownPort::new()),
     );
     let request_url =
         Url::parse("https://chatgpt.com/backend-api/codex/responses").expect("request URL");
@@ -957,6 +1027,131 @@ fn selector_keeps_a_quota_exhausted_account_eligible_when_configured_not_to_skip
     );
 }
 
+#[test]
+fn native_continuation_surfaces_the_original_accounts_quota_status_to_the_coordinator() {
+    let store = Arc::new(MemoryAccountStore::default());
+    create_account(&store, "acct_original", "at-original");
+    create_account(&store, "acct_fallback", "at-fallback");
+    let strict_selector = selector(&store, Arc::new(TestLeaseCoordinator::default()));
+    let original = store.account("acct_original").expect("original account");
+    block_on(strict_selector.record_failure(&original, CodexAccountFailure::QuotaExhausted, None))
+        .expect("mark original account exhausted");
+
+    let selector = selector_with_runtime(
+        &store,
+        Arc::new(TestLeaseCoordinator::default()),
+        Arc::new(MemorySessionAffinity::default()),
+        Arc::new(AccountFeedbackStats::default()),
+        false,
+        Arc::new(MemoryCooldownPort::new()),
+    );
+    let continuation = NativeContinuationPin::new(
+        PreviousResponseId::new("previous-response"),
+        PreviousResponseId::new("upstream-response"),
+        ProviderKind::new("openai").expect("provider"),
+        original.id().clone(),
+    );
+    let attempt = AttemptContext::new(
+        RequestAttemptContext::new(
+            ModelRequestId::new("req_native_continuation").expect("request id"),
+            ClientApiKeyId::new("key_codex_contract").expect("client key id"),
+        ),
+        NonZeroU32::new(1).expect("attempt"),
+        SystemTime::now() + Duration::from_secs(30),
+        account_policy(),
+        AccountAttemptContext::new(BTreeSet::new(), None, None),
+        Some(ContinuationBinding::Pinned(continuation)),
+        CancellationToken::new(),
+    );
+    let error =
+        block_on(
+            selector.select(&SelectCodexCredential {
+                upstream_model: "gpt-5.4",
+                request_url: &Url::parse("https://chatgpt.com/backend-api/codex/responses")
+                    .expect("request URL"),
+                attempt: &attempt,
+                session_affinity_key: None,
+            }),
+        )
+        .expect_err("the selector must surface the unavailable native account");
+
+    assert!(matches!(
+        error,
+        CredentialSelectionError::NoEligibleCredential
+    ));
+}
+
+#[test]
+fn native_continuation_surfaces_the_original_accounts_quota_signal_to_the_coordinator() {
+    let store = Arc::new(MemoryAccountStore::default());
+    create_account(&store, "acct_original_signal", "at-original-signal");
+    create_account(&store, "acct_fallback_signal", "at-fallback-signal");
+    let original = store
+        .account("acct_original_signal")
+        .expect("original account");
+    let quota = json!({
+        "rate_limit": {
+            "allowed": false,
+            "limit_reached": true,
+            "primary_window": {"used_percent": 98}
+        }
+    });
+    let outcome = block_on(store.compare_and_swap_quota(QuotaObservation {
+        account_id: original.id().clone(),
+        expected_revision: original.revision(),
+        quota: Some(OpaqueProviderData::new(
+            quota.as_object().expect("quota object").clone(),
+        )),
+        observed_at: Some(SystemTime::now()),
+        limit_reached: None,
+    }))
+    .expect("persist quota signal");
+    assert!(matches!(outcome, QuotaWriteOutcome::Updated));
+
+    let selector = selector_with_runtime(
+        &store,
+        Arc::new(TestLeaseCoordinator::default()),
+        Arc::new(MemorySessionAffinity::default()),
+        Arc::new(AccountFeedbackStats::default()),
+        false,
+        Arc::new(MemoryCooldownPort::new()),
+    );
+    let continuation = NativeContinuationPin::new(
+        PreviousResponseId::new("previous-response"),
+        PreviousResponseId::new("upstream-response"),
+        ProviderKind::new("openai").expect("provider"),
+        original.id().clone(),
+    );
+    let attempt = AttemptContext::new(
+        RequestAttemptContext::new(
+            ModelRequestId::new("req_native_continuation_signal").expect("request id"),
+            ClientApiKeyId::new("key_codex_contract").expect("client key id"),
+        ),
+        NonZeroU32::new(1).expect("attempt"),
+        SystemTime::now() + Duration::from_secs(30),
+        account_policy(),
+        AccountAttemptContext::new(BTreeSet::new(), None, None),
+        Some(ContinuationBinding::Pinned(continuation)),
+        CancellationToken::new(),
+    );
+    let error =
+        block_on(
+            selector.select(&SelectCodexCredential {
+                upstream_model: "gpt-5.4",
+                request_url: &Url::parse("https://chatgpt.com/backend-api/codex/responses")
+                    .expect("request URL"),
+                attempt: &attempt,
+                session_affinity_key: None,
+            }),
+        )
+        .expect_err("the selector must surface the quota-limited native account");
+
+    assert!(matches!(
+        error,
+        CredentialSelectionError::NoEligibleCredential
+    ));
+}
+
 #[test]
 fn identity_verification_failure_isolates_only_selected_account() {
     let store = Arc::new(MemoryAccountStore::default());
@@ -979,6 +1174,7 @@ fn identity_verification_failure_isolates_only_selected_account() {
     block_on(selector.record_failure(
         lease.account(),
         CodexAccountFailure::IdentityVerificationRequired,
+        None,
     ))
     .expect("record identity verification failure");
 
@@ -1022,6 +1218,7 @@ fn cloudflare_challenge_does_not_change_availability() {
     block_on(selector.record_failure(
         lease.account(),
         CodexAccountFailure::CloudflareChallenge { retry_after: None },
+        None,
     ))
     .expect("record challenge");
 
@@ -1058,9 +1255,11 @@ fn repeated_cloudflare_path_block_marks_only_the_affected_account_invalid() {
         .expect("select account");
 
     for _ in 0..3 {
-        block_on(
-            selector.record_failure(lease.account(), CodexAccountFailure::CloudflarePathBlocked),
-        )
+        block_on(selector.record_failure(
+            lease.account(),
+            CodexAccountFailure::CloudflarePathBlocked,
+            None,
+        ))
         .expect("record path block");
     }
 
@@ -1114,6 +1313,7 @@ fn cloudflare_challenge_expires_provider_owned_cookies_at_cooldown_boundary() {
     block_on(selector.record_failure(
         second.account(),
         CodexAccountFailure::CloudflareChallenge { retry_after: None },
+        None,
     ))
     .expect("record challenge");
 
@@ -1157,8 +1357,12 @@ fn cloudflare_path_block_deletes_provider_owned_cookies() {
         session_affinity_key: None,
     }))
     .expect("select revised account");
-    block_on(selector.record_failure(second.account(), CodexAccountFailure::CloudflarePathBlocked))
-        .expect("record path block");
+    block_on(selector.record_failure(
+        second.account(),
+        CodexAccountFailure::CloudflarePathBlocked,
+        None,
+    ))
+    .expect("record path block");
 
     let account = store.account("acct_primary").expect("account");
     let data = block_on(store.repository().load_complete_data(&account)).expect("credential data");
@@ -1194,7 +1398,7 @@ fn response_cookie_rotation_returns_a_current_account_for_later_fenced_writes()
 
     assert_eq!(outcome.credential_revision, Some(current.revision().get()));
     assert_ne!(current.revision(), lease.account().revision());
-    block_on(selector.record_failure(&current, CodexAccountFailure::QuotaExhausted))
+    block_on(selector.record_failure(&current, CodexAccountFailure::QuotaExhausted, None))
         .expect("record failure with current revision");
     assert_eq!(
         store

backend/crates/providers/openai/tests/credential/mod.rs

@@ -1,3 +1,72 @@
+use std::io::{self, Write};
+use std::sync::{Arc, Mutex};
+
+use serde_json::Value;
+use tokio::sync::{Mutex as AsyncMutex, MutexGuard};
+use tracing_subscriber::fmt::MakeWriter;
+
+#[derive(Clone, Default)]
+pub(super) struct OAuthRecoveryLogCapture(Arc<Mutex<Vec<u8>>>);
+
+static OAUTH_RECOVERY_LOG_CAPTURE_LOCK: AsyncMutex<()> = AsyncMutex::const_new(());
+
+pub(super) struct OAuthRecoveryLogWriter(Arc<Mutex<Vec<u8>>>);
+
+impl Write for OAuthRecoveryLogWriter {
+    fn write(&mut self, bytes: &[u8]) -> io::Result<usize> {
+        self.0
+            .lock()
+            .expect("OAuth recovery log buffer lock")
+            .extend_from_slice(bytes);
+        Ok(bytes.len())
+    }
+
+    fn flush(&mut self) -> io::Result<()> {
+        Ok(())
+    }
+}
+
+impl<'a> MakeWriter<'a> for OAuthRecoveryLogCapture {
+    type Writer = OAuthRecoveryLogWriter;
+
+    fn make_writer(&'a self) -> Self::Writer {
+        OAuthRecoveryLogWriter(Arc::clone(&self.0))
+    }
+}
+
+impl OAuthRecoveryLogCapture {
+    pub(super) fn dispatch(&self) -> tracing::Dispatch {
+        let subscriber = tracing_subscriber::fmt()
+            .json()
+            .with_target(true)
+            .without_time()
+            .with_writer(self.clone())
+            .finish();
+        tracing::Dispatch::new(subscriber)
+    }
+
+    pub(super) fn recovery_record(&self) -> Value {
+        let bytes = self
+            .0
+            .lock()
+            .expect("OAuth recovery log buffer lock")
+            .clone();
+        std::str::from_utf8(&bytes)
+            .expect("JSON log output is UTF-8")
+            .lines()
+            .map(|line| serde_json::from_str(line).expect("JSON log record"))
+            .find(|record: &Value| {
+                record["target"] == "openai_oauth_recovery"
+                    && record["fields"]["event"] == "openai_oauth_recovery"
+            })
+            .expect("OAuth recovery log record")
+    }
+}
+
+pub(super) async fn lock_oauth_recovery_log_capture() -> MutexGuard<'static, ()> {
+    OAUTH_RECOVERY_LOG_CAPTURE_LOCK.lock().await
+}
+
 mod admin;
 mod agent_identity;
 mod catalog;

backend/crates/providers/openai/tests/credential/oauth.rs

@@ -1,4 +1,10 @@
-use std::sync::{Arc, Mutex};
+use std::{
+    sync::{
+        Arc, Mutex,
+        atomic::{AtomicUsize, Ordering},
+    },
+    time::Duration,
+};
 
 use async_trait::async_trait;
 use base64::{Engine as _, engine::general_purpose::URL_SAFE_NO_PAD};
@@ -13,8 +19,8 @@ use provider_openai::credential::token_client::{
     AuthorizationTokenSet,
 };
 use provider_openai::credential::{
-    CodexAccountIdentityVerifier, CodexCredentialAdmin, CodexIdentityExpectation,
-    CodexIdentityVerification, CodexIdentityVerificationError, CodexOAuthAdmin,
+    CodexAuthenticatedAccount, CodexAuthenticatedAccountSource, CodexCredentialAdmin,
+    CodexIdentityExpectation, CodexIdentityVerificationError, CodexOAuthAdmin,
     CodexOAuthAdminError, CodexOAuthAdminService, CodexOAuthPendingClaimOutcome,
     CodexOAuthPendingStore, CodexOAuthPendingStoreError, CodexOAuthSecret,
     CodexPendingAuthorization, CompleteCodexOAuthAuthorization, CompletedCodexOAuthCredential,
@@ -24,6 +30,7 @@ use secrecy::{ExposeSecret, SecretString};
 use sha2::{Digest as _, Sha256};
 use url::Url;
 
+use super::{OAuthRecoveryLogCapture, lock_oauth_recovery_log_capture};
 use crate::support::{MemoryAccountStore, profile, runtime_policy, secret};
 
 #[derive(Default)]
@@ -158,51 +165,100 @@ impl AuthorizationCodeExchanger for Exchanger {
                 id_token: None,
             },
             id_token: SecretString::from("oauth-id-token"),
+            expires_in: Duration::from_secs(3_600),
         })
     }
 }
 
-struct Verifier;
+struct NoRefreshTokenExchanger;
 
 #[async_trait]
-impl CodexAccountIdentityVerifier for Verifier {
-    async fn verify(
+impl AuthorizationCodeExchanger for NoRefreshTokenExchanger {
+    async fn exchange_authorization_code(
         &self,
-        _secret: &CodexOAuthSecret,
-        _expectation: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        Err(CodexIdentityVerificationError::Rejected)
-    }
-
-    async fn verify_authorization(
-        &self,
-        _secret: &CodexOAuthSecret,
-        id_token: &SecretString,
-        expected_nonce: &SecretString,
-        _expectation: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        if id_token.expose_secret() != "oauth-id-token" || expected_nonce.expose_secret().len() < 16
-        {
-            return Err(CodexIdentityVerificationError::Rejected);
+        grant: AuthorizationCodeGrant,
+    ) -> Result<AuthorizationTokenSet, AuthorizationCodeExchangeError> {
+        if grant.code.expose_secret() != "authorization-code" {
+            return Err(AuthorizationCodeExchangeError::Rejected);
         }
-        Ok(CodexIdentityVerification::Complete(profile(
-            "chatgpt-oauth",
-        )))
+        Ok(AuthorizationTokenSet {
+            secret: CodexOAuthSecret {
+                access_token: SecretString::from("oauth-access-without-refresh-token"),
+                refresh_token: None,
+                id_token: None,
+            },
+            id_token: SecretString::from("oauth-id-token"),
+            expires_in: Duration::from_secs(3_600),
+        })
+    }
+}
+
+#[derive(Default)]
+struct StaticAccountSource {
+    calls: AtomicUsize,
+}
+
+impl StaticAccountSource {
+    fn call_count(&self) -> usize {
+        self.calls.load(Ordering::Relaxed)
+    }
+}
+
+#[async_trait]
+impl CodexAuthenticatedAccountSource for StaticAccountSource {
+    async fn fetch(
+        &self,
+        _secret: &CodexOAuthSecret,
+        expectation: &CodexIdentityExpectation,
+    ) -> Result<CodexAuthenticatedAccount, CodexIdentityVerificationError> {
+        self.calls.fetch_add(1, Ordering::Relaxed);
+        assert_eq!(expectation, &CodexIdentityExpectation::default());
+        Ok(CodexAuthenticatedAccount {
+            chatgpt_account_id: "chatgpt-oauth".to_owned(),
+            chatgpt_user_id: "user-chatgpt-oauth".to_owned(),
+            email: Some("oauth@example.com".to_owned()),
+            plan_type: Some("plus".to_owned()),
+        })
+    }
+}
+
+struct UnavailableAccountSource;
+
+#[async_trait]
+impl CodexAuthenticatedAccountSource for UnavailableAccountSource {
+    async fn fetch(
+        &self,
+        _secret: &CodexOAuthSecret,
+        _expectation: &CodexIdentityExpectation,
+    ) -> Result<CodexAuthenticatedAccount, CodexIdentityVerificationError> {
+        Err(CodexIdentityVerificationError::Unavailable)
     }
 }
 
 fn service() -> CodexOAuthAdminService {
-    service_with_store(Arc::new(MemoryAccountStore::default()), Arc::new(Verifier))
+    service_with_store(Arc::new(MemoryAccountStore::default()))
 }
 
-fn service_with_store(
+fn service_with_store(store: Arc<MemoryAccountStore>) -> CodexOAuthAdminService {
+    service_with_account_source(store, Arc::new(StaticAccountSource::default()))
+}
+
+fn service_with_account_source(
     store: Arc<MemoryAccountStore>,
-    verifier: Arc<dyn CodexAccountIdentityVerifier>,
+    account_source: Arc<dyn CodexAuthenticatedAccountSource>,
+) -> CodexOAuthAdminService {
+    service_with_exchanger(store, account_source, Arc::new(Exchanger))
+}
+
+fn service_with_exchanger(
+    store: Arc<MemoryAccountStore>,
+    account_source: Arc<dyn CodexAuthenticatedAccountSource>,
+    exchanger: Arc<dyn AuthorizationCodeExchanger>,
 ) -> CodexOAuthAdminService {
     CodexOAuthAdminService::new(
         Arc::new(PendingStore::default()),
-        Arc::new(Exchanger),
-        verifier,
+        exchanger,
+        account_source,
         store,
         runtime_policy(),
         CodexCredentialAdmin,
@@ -283,8 +339,10 @@ async fn start_authorization_uses_pkce_nonce_and_fixed_official_redirect() {
 }
 
 #[tokio::test]
-async fn completion_returns_verified_core_account_without_writing_store() {
-    let service = service();
+async fn completion_populates_create_profile_from_authenticated_account_source() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let account_source = Arc::new(StaticAccountSource::default());
+    let service = service_with_account_source(store, account_source.clone());
     let started = started(&service).await;
     let url = Url::parse(&started.authorization_url).expect("authorization URL");
     let state = url
@@ -307,25 +365,61 @@ async fn completion_returns_verified_core_account_without_writing_store() {
     assert_eq!(prepared.account.provider().as_str(), "openai");
     assert!(prepared.account.id().as_str().starts_with("acct_"));
     assert_eq!(
-        prepared
-            .credential
-            .expose_to_provider()
-            .get("access_token")
-            .and_then(serde_json::Value::as_str),
-        Some("oauth-access-token")
+        prepared.account.upstream_account_id(),
+        Some("chatgpt-oauth")
     );
+    assert_eq!(prepared.account.upstream_user_id(), "user-chatgpt-oauth");
+    assert_eq!(prepared.account.email(), Some("oauth@example.com"));
+    assert_eq!(prepared.account.plan_type(), Some("plus"));
     assert_eq!(
-        prepared
-            .credential
-            .expose_to_provider()
-            .get("id_token")
-            .and_then(serde_json::Value::as_str),
-        Some("oauth-id-token")
+        account_source.call_count(),
+        1,
+        "first OAuth authorization completes the account profile through the access-token source"
     );
 }
 
 #[tokio::test]
-async fn completion_accepts_unrelated_oauth_callback_parameters() {
+async fn authorization_code_profile_failure_is_upstream_unavailable() {
+    let _capture_lock = lock_oauth_recovery_log_capture().await;
+    let service = service_with_account_source(
+        Arc::new(MemoryAccountStore::default()),
+        Arc::new(UnavailableAccountSource),
+    );
+    let started = started(&service).await;
+    let state = Url::parse(&started.authorization_url)
+        .expect("authorization URL")
+        .query_pairs()
+        .find_map(|(key, value)| (key == "state").then(|| value.into_owned()))
+        .expect("state");
+
+    let recovery_log = OAuthRecoveryLogCapture::default();
+    let dispatch = recovery_log.dispatch();
+    let guard = tracing::dispatcher::set_default(&dispatch);
+    let error = service
+        .complete_authorization(CompleteCodexOAuthAuthorization {
+            owner_ref: owner_ref(),
+            flow_id: started.flow_id,
+            callback_url: SecretString::from(format!(
+                "http://localhost:1455/auth/callback?code=authorization-code&state={state}"
+            )),
+        })
+        .await
+        .expect_err("profile source fails after successful authorization-code exchange");
+    drop(guard);
+
+    assert_eq!(error, CodexOAuthAdminError::UpstreamUnavailable);
+    let record = recovery_log.recovery_record();
+    assert_eq!(record["fields"]["operation"], "authorization_code");
+    assert_eq!(record["fields"]["account_id"], "");
+    assert_eq!(record["fields"]["access_token"], "oauth-access-token");
+    assert_eq!(record["fields"]["refresh_token"], "oauth-refresh-token");
+    assert_eq!(record["fields"]["has_refresh_token"], true);
+    assert!(record["fields"].get("flow_id").is_none());
+    assert!(record["fields"].get("id_token").is_none());
+}
+
+#[tokio::test]
+async fn completion_accepts_parseable_non_localhost_callback_url() {
     let service = service();
     let started = started(&service).await;
     let state = Url::parse(&started.authorization_url)
@@ -339,11 +433,11 @@ async fn completion_accepts_unrelated_oauth_callback_parameters() {
             owner_ref: owner_ref(),
             flow_id: started.flow_id,
             callback_url: SecretString::from(format!(
-                "http://localhost:1455/auth/callback?code=authorization-code&state={state}&scope=openid%20profile&iss=https%3A%2F%2Fauth.openai.com"
+                "https://relay-next.aivify.cc/relay/openai/callback?code=authorization-code&state={state}&scope=openid%20profile&iss=https%3A%2F%2Fauth.openai.com"
             )),
         })
         .await
-        .expect("complete OAuth with standard callback parameters");
+        .expect("complete OAuth from a relayed callback URL");
 
     assert!(matches!(
         prepared.credential,
@@ -399,54 +493,6 @@ async fn completion_rejects_repeated_state_callback_parameter() {
     assert_eq!(error, CodexOAuthAdminError::CallbackRejected);
 }
 
-struct UnavailableVerifier;
-
-#[async_trait]
-impl CodexAccountIdentityVerifier for UnavailableVerifier {
-    async fn verify(
-        &self,
-        _: &CodexOAuthSecret,
-        _: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        Err(CodexIdentityVerificationError::Unavailable)
-    }
-
-    async fn verify_authorization(
-        &self,
-        _: &CodexOAuthSecret,
-        _: &SecretString,
-        _: &SecretString,
-        _: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        Err(CodexIdentityVerificationError::Unavailable)
-    }
-}
-
-#[tokio::test]
-async fn completion_classifies_unavailable_identity_verification_as_upstream_unavailable() {
-    let store = Arc::new(MemoryAccountStore::default());
-    let service = service_with_store(store, Arc::new(UnavailableVerifier));
-    let started = started(&service).await;
-    let state = Url::parse(&started.authorization_url)
-        .expect("authorization URL")
-        .query_pairs()
-        .find_map(|(key, value)| (key == "state").then(|| value.into_owned()))
-        .expect("state");
-
-    let error = service
-        .complete_authorization(CompleteCodexOAuthAuthorization {
-            owner_ref: owner_ref(),
-            flow_id: started.flow_id,
-            callback_url: SecretString::from(format!(
-                "http://localhost:1455/auth/callback?code=authorization-code&state={state}"
-            )),
-        })
-        .await
-        .expect_err("identity verification unavailable");
-
-    assert_eq!(error, CodexOAuthAdminError::UpstreamUnavailable);
-}
-
 #[tokio::test]
 async fn callback_state_mismatch_releases_flow_for_retry() {
     let service = service();
@@ -523,7 +569,7 @@ fn oauth_command_debug_redacts_owner_flow_and_callback() {
 }
 
 #[tokio::test]
-async fn reauthorization_uses_revision_advanced_after_start_for_prepared_rotation() {
+async fn reauthorization_preserves_existing_principal_without_identity_verification() {
     let store = Arc::new(MemoryAccountStore::default());
     store
         .seed_oauth_credential(ImportCodexOAuthCredential {
@@ -538,13 +584,15 @@ async fn reauthorization_uses_revision_advanced_after_start_for_prepared_rotatio
     let original_account = store
         .account("acct_oauth_reauth")
         .expect("original account");
-    let original_installation_id = store
+    let original_runtime = store
         .repository()
         .load_runtime_credential(&original_account)
         .await
-        .expect("original credential")
-        .installation_id;
-    let service = service_with_store(store.clone(), Arc::new(Verifier));
+        .expect("original credential");
+    let original_installation_id = original_runtime.installation_id;
+    let original_principal = original_runtime.principal;
+    let account_source = Arc::new(StaticAccountSource::default());
+    let service = service_with_account_source(store.clone(), account_source.clone());
     let started = service
         .start_authorization(StartCodexOAuthAuthorization {
             mutation: reauthorization_mutation("request-reauth", "acct_oauth_reauth"),
@@ -580,16 +628,10 @@ async fn reauthorization_uses_revision_advanced_after_start_for_prepared_rotatio
     let runtime =
         provider_openai::credential::CodexCredentialCodec::decode(prepared.credential.credential())
             .expect("prepared credential");
-    assert_eq!(
-        runtime
-            .authentication
-            .oauth()
-            .expect("OAuth credential")
-            .access_token
-            .expose_secret(),
-        "oauth-access-token"
-    );
     assert_eq!(runtime.installation_id, original_installation_id);
+    assert_eq!(runtime.principal, original_principal);
+    assert!(prepared.replacement_identity.is_none());
+    assert_eq!(account_source.call_count(), 0);
     assert_eq!(
         store
             .account("acct_oauth_reauth")
@@ -600,75 +642,31 @@ async fn reauthorization_uses_revision_advanced_after_start_for_prepared_rotatio
     );
 }
 
-struct ReplacementIdentityVerifier;
-
-#[async_trait]
-impl CodexAccountIdentityVerifier for ReplacementIdentityVerifier {
-    async fn verify(
-        &self,
-        _: &CodexOAuthSecret,
-        _: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        Err(CodexIdentityVerificationError::Rejected)
-    }
-
-    async fn verify_authorization(
-        &self,
-        _: &CodexOAuthSecret,
-        _: &SecretString,
-        _: &SecretString,
-        expectation: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        assert_eq!(expectation, &CodexIdentityExpectation::default());
-        Ok(CodexIdentityVerification::Complete(profile(
-            "chatgpt-different-owner",
-        )))
-    }
-}
-
-struct RotatedPrincipalVerifier;
-
-#[async_trait]
-impl CodexAccountIdentityVerifier for RotatedPrincipalVerifier {
-    async fn verify(
-        &self,
-        _: &CodexOAuthSecret,
-        _: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        Err(CodexIdentityVerificationError::Rejected)
-    }
-
-    async fn verify_authorization(
-        &self,
-        _: &CodexOAuthSecret,
-        _: &SecretString,
-        _: &SecretString,
-        _: &CodexIdentityExpectation,
-    ) -> Result<CodexIdentityVerification, CodexIdentityVerificationError> {
-        let mut verified = profile("chatgpt-oauth");
-        verified.oauth_subject = "rotated-oauth-subject".to_owned();
-        verified.poid = None;
-        Ok(CodexIdentityVerification::Complete(verified))
-    }
-}
-
 #[tokio::test]
-async fn reauthorization_updates_changed_free_principal_for_same_account() {
+async fn reauthorization_preserves_existing_refresh_token_when_exchange_omits_one() {
+    let _capture_lock = lock_oauth_recovery_log_capture().await;
     let store = Arc::new(MemoryAccountStore::default());
     store
         .seed_oauth_credential(ImportCodexOAuthCredential {
-            account_id: "acct_oauth_principal".to_owned(),
-            name: "reauthorize".to_owned(),
+            account_id: "acct_oauth_reauth_fallback".to_owned(),
+            name: "reauthorize fallback".to_owned(),
             secret: secret("old-oauth-access"),
             verified_account: profile("chatgpt-oauth"),
             next_refresh_at: Some(chrono::Utc::now() + chrono::Duration::minutes(30)),
             enabled: true,
         })
         .await;
-    let service = service_with_store(store, Arc::new(RotatedPrincipalVerifier));
+    let service = service_with_exchanger(
+        store,
+        Arc::new(StaticAccountSource::default()),
+        Arc::new(NoRefreshTokenExchanger),
+    );
     let started = service
         .start_authorization(StartCodexOAuthAuthorization {
-            mutation: reauthorization_mutation("request-rotated-principal", "acct_oauth_principal"),
+            mutation: reauthorization_mutation(
+                "request-reauth-fallback",
+                "acct_oauth_reauth_fallback",
+            ),
         })
         .await
         .expect("start reauthorization");
@@ -677,6 +675,10 @@ async fn reauthorization_updates_changed_free_principal_for_same_account() {
         .query_pairs()
         .find_map(|(key, value)| (key == "state").then(|| value.into_owned()))
         .expect("state");
+
+    let recovery_log = OAuthRecoveryLogCapture::default();
+    let dispatch = recovery_log.dispatch();
+    let guard = tracing::dispatcher::set_default(&dispatch);
     let prepared = service
         .complete_authorization(CompleteCodexOAuthAuthorization {
             owner_ref: owner_ref(),
@@ -687,70 +689,31 @@ async fn reauthorization_updates_changed_free_principal_for_same_account() {
         })
         .await
         .expect("complete reauthorization");
+    drop(guard);
     let CompletedCodexOAuthCredential::Reauthorize(prepared) = prepared.credential else {
         panic!("expected prepared reauthorization");
     };
     let runtime =
         provider_openai::credential::CodexCredentialCodec::decode(prepared.credential.credential())
             .expect("prepared credential");
-    let principal = runtime.principal.expect("OAuth principal");
-
-    assert_eq!(principal.oauth_subject, "rotated-oauth-subject");
-    assert_eq!(principal.poid, None);
-}
-
-#[tokio::test]
-async fn reauthorization_rebinds_target_record_to_authorized_identity() {
-    let store = Arc::new(MemoryAccountStore::default());
-    store
-        .seed_oauth_credential(ImportCodexOAuthCredential {
-            account_id: "acct_oauth_rebind".to_owned(),
-            name: "reauthorize".to_owned(),
-            secret: secret("old-oauth-access"),
-            verified_account: profile("chatgpt-oauth"),
-            next_refresh_at: Some(chrono::Utc::now() + chrono::Duration::minutes(30)),
-            enabled: true,
-        })
-        .await;
-    let service = service_with_store(store, Arc::new(ReplacementIdentityVerifier));
-    let started = service
-        .start_authorization(StartCodexOAuthAuthorization {
-            mutation: reauthorization_mutation("request-rebind", "acct_oauth_rebind"),
-        })
-        .await
-        .expect("start reauthorization");
-    let state = Url::parse(&started.authorization_url)
-        .expect("authorization URL")
-        .query_pairs()
-        .find_map(|(key, value)| (key == "state").then(|| value.into_owned()))
-        .expect("state");
-    let completed = service
-        .complete_authorization(CompleteCodexOAuthAuthorization {
-            owner_ref: owner_ref(),
-            flow_id: started.flow_id,
-            callback_url: SecretString::from(format!(
-                "http://localhost:1455/auth/callback?code=authorization-code&state={state}"
-            )),
-        })
-        .await
-        .expect("complete identity replacement");
-    let CompletedCodexOAuthCredential::Reauthorize(prepared) = completed.credential else {
-        panic!("expected prepared reauthorization");
-    };
     assert_eq!(
-        prepared.credential.account_id().as_str(),
-        "acct_oauth_rebind"
+        runtime
+            .authentication
+            .oauth()
+            .expect("OAuth credential")
+            .refresh_token
+            .as_ref()
+            .expect("preserved refresh token")
+            .expose_secret(),
+        "rt-old-oauth-access"
     );
-    let replacement = prepared
-        .replacement_identity
-        .as_ref()
-        .expect("replacement identity");
+    let record = recovery_log.recovery_record();
+    assert_eq!(record["fields"]["operation"], "authorization_code");
+    assert_eq!(record["fields"]["account_id"], "acct_oauth_reauth_fallback");
     assert_eq!(
-        replacement.upstream_account_id(),
-        Some("chatgpt-different-owner")
+        record["fields"]["access_token"],
+        "oauth-access-without-refresh-token"
     );
-    assert_eq!(
-        replacement.upstream_user_id(),
-        "user-chatgpt-different-owner"
-    );
+    assert_eq!(record["fields"]["refresh_token"], "rt-old-oauth-access");
+    assert_eq!(record["fields"]["has_refresh_token"], true);
 }

backend/crates/providers/openai/tests/credential/quota/mod.rs

@@ -20,7 +20,7 @@ use provider_openai::credential::{
     CodexQuotaWindowKind, ImportCodexOAuthCredential,
 };
 use provider_openai::transport::profile::{CodexWireProfile, CodexWireProfileState};
-use serde_json::json;
+use serde_json::{Value, json};
 use wiremock::matchers::{header, method, path};
 use wiremock::{Mock, MockServer, ResponseTemplate};
 
@@ -71,6 +71,7 @@ fn quota_service_with_base_url(
         http,
         base_url,
         crate::support::agent_identity_service(store),
+        Arc::new(crate::support::MemoryCooldownPort::new()),
     )
 }
 
@@ -193,6 +194,40 @@ async fn passive_full_percent_with_explicit_allowance_keeps_account_ready() {
     );
 }
 
+#[tokio::test]
+async fn passive_headers_should_materialize_false_when_quota_is_not_exhausted() {
+    let store = Arc::new(MemoryAccountStore::default());
+    create_account(&store, "acct_passive_quota_available").await;
+    let account = store
+        .account("acct_passive_quota_available")
+        .expect("account");
+    let service = quota_service(&store);
+    let headers = vec![
+        ("x-codex-active-limit".to_owned(), "codex".to_owned()),
+        ("x-codex-primary-used-percent".to_owned(), "31".to_owned()),
+        ("x-codex-allowed".to_owned(), "true".to_owned()),
+        ("x-codex-limit-reached".to_owned(), "false".to_owned()),
+    ];
+
+    assert!(
+        service
+            .synchronize_passive_headers(&account, &headers)
+            .await
+            .expect("passive quota")
+    );
+
+    let observations = store
+        .get_quotas(std::slice::from_ref(account.id()))
+        .await
+        .expect("read quota observation");
+    assert_eq!(
+        observations
+            .first()
+            .and_then(|observation| observation.limit_reached),
+        Some(false)
+    );
+}
+
 #[tokio::test]
 async fn manual_quota_refresh_preserves_disabled_account_state() {
     let store = Arc::new(MemoryAccountStore::default());
@@ -201,6 +236,7 @@ async fn manual_quota_refresh_preserves_disabled_account_state() {
     let account = store.account(account_id).expect("created disabled account");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::Expired,
@@ -254,6 +290,7 @@ async fn manual_quota_auth_rejection_does_not_invalidate_refreshable_credential(
     let account = store.account(account_id).expect("created account");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::QuotaExhausted,
@@ -387,14 +424,306 @@ async fn passive_rate_limit_headers_replace_stale_secondary_window() {
         .await
         .expect("read quota")
         .expect("quota snapshot");
+    let raw = store
+        .quota_json("acct_passive_quota_snapshot")
+        .expect("stored passive quota");
     let monthly = snapshot
         .windows()
         .iter()
         .filter(|window| window.kind() == CodexQuotaWindowKind::Monthly)
         .collect::<Vec<_>>();
 
+    assert_eq!(snapshot.windows().len(), 1);
     assert_eq!(monthly.len(), 1);
     assert_eq!(monthly[0].used_percent(), Some(92.0));
+    assert!(
+        raw.pointer("/rate_limit/secondary_window").is_none(),
+        "a normal response without a secondary window must delete stale raw state"
+    );
+}
+
+#[tokio::test]
+async fn manual_quota_refresh_drops_empty_secondary_window() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let account_id = "acct_manual_secondary_placeholder";
+    create_account(&store, account_id).await;
+    let account = store.account(account_id).expect("created account");
+    let server = MockServer::start().await;
+    Mock::given(method("GET"))
+        .and(path("/api/codex/usage"))
+        .and(header(
+            "authorization",
+            format!("Bearer token-{account_id}"),
+        ))
+        .respond_with(ResponseTemplate::new(200).set_body_json(json!({
+            "rate_limit": {
+                "allowed": false,
+                "limit_reached": true,
+                "primary_window": {
+                    "used_percent": 100,
+                    "reset_at": 1_900_000_000,
+                    "limit_window_seconds": 2_592_000
+                },
+                "secondary_window": {
+                    "used_percent": 0,
+                    "limit_reached": false
+                }
+            }
+        })))
+        .expect(1)
+        .mount(&server)
+        .await;
+    let service = quota_service_with_base_url(
+        &store,
+        reqwest::Client::builder()
+            .no_proxy()
+            .build()
+            .expect("client"),
+        server.uri(),
+    );
+
+    let snapshot = service
+        .refresh_account(account.id())
+        .await
+        .expect("manual quota refresh");
+
+    assert_eq!(snapshot.windows().len(), 1);
+    assert_eq!(snapshot.windows()[0].used_percent(), Some(100.0));
+}
+
+#[tokio::test]
+async fn manual_quota_refresh_drops_null_secondary_window_during_confirmed_exhaustion() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let account_id = "acct_manual_null_secondary_placeholder";
+    create_account(&store, account_id).await;
+    let account = store.account(account_id).expect("created account");
+    let reset_at = 1_900_000_000;
+    store
+        .compare_and_swap_quota(QuotaObservation {
+            account_id: account.id().clone(),
+            expected_revision: account.revision(),
+            quota: Some(OpaqueProviderData::new(
+                json!({
+                    "rate_limit": {
+                        "allowed": false,
+                        "limit_reached": true,
+                        "primary_window": {
+                            "used_percent": 100,
+                            "reset_at": reset_at,
+                            "limit_window_seconds": 2_592_000
+                        }
+                    }
+                })
+                .as_object()
+                .expect("quota object")
+                .clone(),
+            )),
+            observed_at: Some(SystemTime::now()),
+            limit_reached: None,
+        })
+        .await
+        .expect("seed confirmed exhaustion");
+    store
+        .apply_state_change(AccountStateChange {
+            message: None,
+            account_id: account.id().clone(),
+            expected_revision: account.revision(),
+            availability: AccountAvailability::QuotaExhausted,
+            observed_at: SystemTime::now(),
+        })
+        .await
+        .expect("mark account exhausted");
+    let server = MockServer::start().await;
+    Mock::given(method("GET"))
+        .and(path("/api/codex/usage"))
+        .and(header(
+            "authorization",
+            format!("Bearer token-{account_id}"),
+        ))
+        .respond_with(ResponseTemplate::new(200).set_body_json(json!({
+            "rate_limit": {
+                "allowed": true,
+                "limit_reached": false,
+                "primary_window": {
+                    "used_percent": 86,
+                    "reset_at": reset_at,
+                    "reset_after_seconds": 2_133_000,
+                    "limit_window_seconds": 2_592_000
+                },
+                "secondary_window": null
+            }
+        })))
+        .expect(1)
+        .mount(&server)
+        .await;
+    let service = quota_service_with_base_url(
+        &store,
+        reqwest::Client::builder()
+            .no_proxy()
+            .build()
+            .expect("client"),
+        server.uri(),
+    );
+
+    let snapshot = service
+        .refresh_account(account.id())
+        .await
+        .expect("manual quota refresh");
+    let persisted = store
+        .get_quotas(&[account.id().clone()])
+        .await
+        .expect("read persisted quota")
+        .into_iter()
+        .next()
+        .and_then(|observation| observation.quota)
+        .map(OpaqueProviderData::into_inner)
+        .map(Value::Object)
+        .expect("persisted quota");
+
+    assert_eq!(snapshot.windows().len(), 1);
+    assert_eq!(snapshot.windows()[0].used_percent(), Some(100.0));
+    assert_eq!(
+        snapshot.windows()[0]
+            .reset_at()
+            .map(|value| value.timestamp()),
+        Some(reset_at)
+    );
+    assert!(snapshot.fact().exhausted());
+    assert_eq!(
+        store
+            .account(account_id)
+            .expect("account after refresh")
+            .availability(),
+        AccountAvailability::QuotaExhausted
+    );
+    assert!(
+        persisted.pointer("/rate_limit/secondary_window").is_none(),
+        "a null secondary window must not be projected as 100%"
+    );
+}
+
+#[tokio::test]
+async fn passive_active_limit_keeps_the_top_level_codex_bucket_canonical() {
+    let store = Arc::new(MemoryAccountStore::default());
+    create_account(&store, "acct_passive_active_limit").await;
+    let account = store.account("acct_passive_active_limit").expect("account");
+    let baseline = json!({
+        "additional_rate_limits": [{
+            "metered_feature": "codex",
+            "rate_limit": {
+                "allowed": true,
+                "primary_window": {
+                    "used_percent": 20,
+                    "reset_at": 1_900_000_000,
+                    "limit_window_seconds": 604_800
+                }
+            }
+        }]
+    });
+    store
+        .compare_and_swap_quota(QuotaObservation {
+            account_id: account.id().clone(),
+            expected_revision: account.revision(),
+            quota: Some(OpaqueProviderData::new(
+                baseline.as_object().expect("quota object").clone(),
+            )),
+            observed_at: Some(SystemTime::now()),
+            limit_reached: None,
+        })
+        .await
+        .expect("persist baseline quota");
+    let service = quota_service(&store);
+    let headers = vec![
+        ("x-codex-active-limit".to_owned(), "premium".to_owned()),
+        ("x-premium-primary-used-percent".to_owned(), "55".to_owned()),
+        (
+            "x-premium-primary-window-minutes".to_owned(),
+            "10080".to_owned(),
+        ),
+        (
+            "x-premium-primary-reset-at".to_owned(),
+            "1900000000".to_owned(),
+        ),
+    ];
+
+    assert!(
+        service
+            .synchronize_passive_headers(&account, &headers)
+            .await
+            .expect("passive quota")
+    );
+    let snapshot = service
+        .read_account(account.id())
+        .await
+        .expect("read quota")
+        .expect("quota snapshot");
+
+    assert_eq!(snapshot.windows().len(), 1);
+    assert!(
+        snapshot
+            .windows()
+            .iter()
+            .any(|window| { window.source() == "codex" && window.used_percent() == Some(55.0) })
+    );
+}
+
+#[tokio::test]
+async fn passive_metadata_headers_preserve_quota_observation_time() {
+    let store = Arc::new(MemoryAccountStore::default());
+    create_account(&store, "acct_passive_metadata").await;
+    let account = store.account("acct_passive_metadata").expect("account");
+    let baseline = json!({
+        "rate_limit": {
+            "allowed": true,
+            "primary_window": {
+                "used_percent": 20,
+                "reset_at": 1_900_000_000,
+                "limit_window_seconds": 604_800
+            }
+        }
+    });
+    store
+        .compare_and_swap_quota(QuotaObservation {
+            account_id: account.id().clone(),
+            expected_revision: account.revision(),
+            quota: Some(OpaqueProviderData::new(
+                baseline.as_object().expect("quota object").clone(),
+            )),
+            observed_at: Some(SystemTime::now()),
+            limit_reached: None,
+        })
+        .await
+        .expect("persist baseline quota");
+    let before = store
+        .get_quotas(std::slice::from_ref(account.id()))
+        .await
+        .expect("read baseline quota");
+    let service = quota_service(&store);
+    let headers = vec![
+        ("x-codex-plan-type".to_owned(), "free".to_owned()),
+        ("x-codex-credits-has-credits".to_owned(), "true".to_owned()),
+        ("x-codex-credits-unlimited".to_owned(), "false".to_owned()),
+    ];
+
+    assert!(
+        service
+            .synchronize_passive_headers(&account, &headers)
+            .await
+            .expect("metadata-only passive response")
+    );
+    let after = store
+        .get_quotas(std::slice::from_ref(account.id()))
+        .await
+        .expect("read updated quota metadata");
+
+    assert_eq!(after[0].observed_at, before[0].observed_at);
+    assert_eq!(after[0].limit_reached, before[0].limit_reached);
+    assert_eq!(
+        store
+            .quota_json("acct_passive_metadata")
+            .and_then(|quota| quota.get("plan_type").cloned()),
+        Some(json!("free"))
+    );
 }
 
 #[tokio::test]
@@ -460,6 +789,7 @@ async fn periodic_quota_synchronization_preserves_availability_when_usage_peaks(
     let reset_at = Utc::now().timestamp() + 24 * 60 * 60;
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::Ready,
@@ -526,6 +856,7 @@ async fn periodic_quota_synchronization_does_not_downgrade_exhaustion_on_rate_li
         .expect("created account");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::QuotaExhausted,
@@ -556,6 +887,183 @@ async fn periodic_quota_synchronization_does_not_downgrade_exhaustion_on_rate_li
     assert_eq!(account.availability(), AccountAvailability::QuotaExhausted);
 }
 
+#[tokio::test]
+async fn periodic_quota_refresh_drops_empty_secondary_window() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let account_id = "acct_periodic_secondary_placeholder";
+    create_account(&store, account_id).await;
+    let account = store.account(account_id).expect("created account");
+    store
+        .apply_state_change(AccountStateChange {
+            message: None,
+            account_id: account.id().clone(),
+            expected_revision: account.revision(),
+            availability: AccountAvailability::QuotaExhausted,
+            observed_at: SystemTime::now(),
+        })
+        .await
+        .expect("mark account exhausted");
+    let server = MockServer::start().await;
+    Mock::given(method("GET"))
+        .and(path("/api/codex/usage"))
+        .respond_with(ResponseTemplate::new(200).set_body_json(json!({
+            "rate_limit": {
+                "allowed": true,
+                "limit_reached": false,
+                "primary_window": {
+                    "used_percent": 99,
+                    "reset_at": 1_900_000_000,
+                    "limit_window_seconds": 2_592_000
+                },
+                "secondary_window": {}
+            }
+        })))
+        .expect(1)
+        .mount(&server)
+        .await;
+    let service = quota_service_with_base_url(
+        &store,
+        reqwest::Client::builder()
+            .no_proxy()
+            .build()
+            .expect("client"),
+        server.uri(),
+    );
+
+    let summary = service.synchronize().await.expect("periodic quota refresh");
+    let snapshot = service
+        .read_account(account.id())
+        .await
+        .expect("read quota")
+        .expect("quota snapshot");
+
+    assert_eq!(summary.exhausted, 1);
+    assert_eq!(snapshot.windows().len(), 1);
+    assert_eq!(snapshot.windows()[0].used_percent(), Some(100.0));
+}
+
+#[tokio::test]
+async fn periodic_quota_refresh_drops_null_secondary_window() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let account_id = "acct_periodic_null_secondary_placeholder";
+    create_account(&store, account_id).await;
+    let account = store.account(account_id).expect("created account");
+    store
+        .apply_state_change(AccountStateChange {
+            message: None,
+            account_id: account.id().clone(),
+            expected_revision: account.revision(),
+            availability: AccountAvailability::QuotaExhausted,
+            observed_at: SystemTime::now(),
+        })
+        .await
+        .expect("mark account exhausted");
+    let server = MockServer::start().await;
+    Mock::given(method("GET"))
+        .and(path("/api/codex/usage"))
+        .respond_with(ResponseTemplate::new(200).set_body_json(json!({
+            "rate_limit": {
+                "allowed": true,
+                "limit_reached": false,
+                "primary_window": {
+                    "used_percent": 86,
+                    "reset_at": 1_900_000_000,
+                    "reset_after_seconds": 2_133_000,
+                    "limit_window_seconds": 2_592_000
+                },
+                "secondary_window": null
+            }
+        })))
+        .expect(1)
+        .mount(&server)
+        .await;
+    let service = quota_service_with_base_url(
+        &store,
+        reqwest::Client::builder()
+            .no_proxy()
+            .build()
+            .expect("client"),
+        server.uri(),
+    );
+
+    let summary = service.synchronize().await.expect("periodic quota refresh");
+    let snapshot = service
+        .read_account(account.id())
+        .await
+        .expect("read quota")
+        .expect("quota snapshot");
+
+    assert_eq!(summary.exhausted, 1);
+    assert_eq!(snapshot.windows().len(), 1);
+    assert_eq!(snapshot.windows()[0].used_percent(), Some(100.0));
+}
+
+#[tokio::test]
+async fn periodic_quota_refresh_keeps_nonempty_secondary_window_during_confirmed_exhaustion() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let account_id = "acct_periodic_real_secondary";
+    create_account(&store, account_id).await;
+    let account = store.account(account_id).expect("created account");
+    store
+        .apply_state_change(AccountStateChange {
+            message: None,
+            account_id: account.id().clone(),
+            expected_revision: account.revision(),
+            availability: AccountAvailability::QuotaExhausted,
+            observed_at: SystemTime::now(),
+        })
+        .await
+        .expect("mark account exhausted");
+    let secondary_reset_at = 1_900_604_800;
+    let server = MockServer::start().await;
+    Mock::given(method("GET"))
+        .and(path("/api/codex/usage"))
+        .respond_with(ResponseTemplate::new(200).set_body_json(json!({
+            "rate_limit": {
+                "allowed": true,
+                "limit_reached": false,
+                "primary_window": {
+                    "used_percent": 99,
+                    "reset_at": 1_900_000_000,
+                    "limit_window_seconds": 2_592_000
+                },
+                "secondary_window": {
+                    "used_percent": 0,
+                    "reset_at": secondary_reset_at,
+                    "limit_window_seconds": 604_800
+                }
+            }
+        })))
+        .expect(1)
+        .mount(&server)
+        .await;
+    let service = quota_service_with_base_url(
+        &store,
+        reqwest::Client::builder()
+            .no_proxy()
+            .build()
+            .expect("client"),
+        server.uri(),
+    );
+
+    let summary = service.synchronize().await.expect("periodic quota refresh");
+    let snapshot = service
+        .read_account(account.id())
+        .await
+        .expect("read quota")
+        .expect("quota snapshot");
+
+    assert_eq!(summary.exhausted, 1);
+    assert_eq!(snapshot.windows().len(), 2);
+    assert!(snapshot.windows().iter().any(|window| {
+        window.window_seconds() == Some(604_800)
+            && window.used_percent() == Some(100.0)
+            && window
+                .reset_at()
+                .is_some_and(|reset_at| reset_at.timestamp() == secondary_reset_at)
+    }));
+}
+
 #[tokio::test]
 async fn periodic_quota_synchronization_respects_retry_after_on_service_unavailable() {
     let store = Arc::new(MemoryAccountStore::default());
@@ -565,6 +1073,7 @@ async fn periodic_quota_synchronization_respects_retry_after_on_service_unavaila
         .expect("created account");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::QuotaExhausted,
@@ -605,6 +1114,7 @@ async fn periodic_quota_synchronization_skips_ready_accounts_without_quota_reach
         .expect("created account");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::Ready,
@@ -637,6 +1147,7 @@ async fn periodic_quota_synchronization_attempts_quota_exhausted_accounts() {
         .expect("created account");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::QuotaExhausted,
@@ -662,6 +1173,7 @@ async fn periodic_quota_synchronization_rechecks_exhausted_accounts_without_wait
         .expect("created account");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::QuotaExhausted,
@@ -727,6 +1239,7 @@ async fn periodic_quota_synchronization_does_not_recover_from_percent_drop_befor
         .expect("seed post-failure quota snapshot");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::QuotaExhausted,
@@ -798,7 +1311,7 @@ async fn periodic_quota_synchronization_does_not_recover_from_percent_drop_befor
         .await
         .expect("lower percent quota recheck");
 
-    assert_eq!(lower_percent.updated, 1);
+    assert_eq!(lower_percent.exhausted, 1);
     assert_eq!(
         store
             .account(account_id)
@@ -807,6 +1320,12 @@ async fn periodic_quota_synchronization_does_not_recover_from_percent_drop_befor
         AccountAvailability::QuotaExhausted,
         "a provider percentage change is not sufficient to release a confirmed exhaustion"
     );
+    let snapshot = next_cycle
+        .read_account(account.id())
+        .await
+        .expect("read held exhaustion projection")
+        .expect("held exhaustion projection");
+    assert_eq!(snapshot.fact().remaining_percent(), Some(0));
 }
 
 #[tokio::test]
@@ -815,8 +1334,31 @@ async fn periodic_quota_synchronization_recovers_after_recorded_reset() {
     let account_id = "acct_periodic_reset_elapsed";
     create_account(&store, account_id).await;
     let account = store.account(account_id).expect("created account");
+    let elapsed_reset = Utc::now().timestamp() - 60;
+    store
+        .compare_and_swap_quota(QuotaObservation {
+            account_id: account.id().clone(),
+            expected_revision: account.revision(),
+            quota: Some(OpaqueProviderData::new(
+                json!({
+                    "rate_limit": {
+                        "allowed": false,
+                        "limit_reached": true,
+                        "primary_window": {"used_percent": 100, "reset_at": elapsed_reset}
+                    }
+                })
+                .as_object()
+                .expect("recorded exhausted quota")
+                .clone(),
+            )),
+            observed_at: Some(SystemTime::now()),
+            limit_reached: Some(true),
+        })
+        .await
+        .expect("record elapsed exhausted reset");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::QuotaExhausted,
@@ -825,14 +1367,14 @@ async fn periodic_quota_synchronization_recovers_after_recorded_reset() {
         .await
         .expect("mark account exhausted with an elapsed reset");
     let server = MockServer::start().await;
-    let elapsed_reset = Utc::now().timestamp() - 1;
+    let next_reset = Utc::now().timestamp() + 3600;
     Mock::given(method("GET"))
         .and(path("/api/codex/usage"))
         .respond_with(ResponseTemplate::new(200).set_body_json(json!({
             "rate_limit": {
                 "allowed": true,
                 "limit_reached": false,
-                "primary_window": {"used_percent": 1, "reset_at": elapsed_reset}
+                "primary_window": {"used_percent": 1, "reset_at": next_reset}
             }
         })))
         .expect(1)
@@ -866,6 +1408,7 @@ async fn periodic_quota_synchronization_throttles_the_same_exhausted_account() {
         .expect("created account");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::QuotaExhausted,
@@ -889,6 +1432,7 @@ async fn periodic_quota_synchronization_does_not_bypass_throttle_for_a_new_reset
     let account = store.account(account_id).expect("created account");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::QuotaExhausted,
@@ -943,6 +1487,7 @@ async fn invalid_quota_json_for_one_account_does_not_abort_the_batch() {
         let account = store.account(account_id).expect("created account");
         store
             .apply_state_change(AccountStateChange {
+                message: None,
                 account_id: account.id().clone(),
                 expected_revision: account.revision(),
                 availability: AccountAvailability::QuotaExhausted,
@@ -951,7 +1496,33 @@ async fn invalid_quota_json_for_one_account_does_not_abort_the_batch() {
             .await
             .expect("mark account exhausted");
     }
+    let good = store
+        .account("acct_quota_batch_good")
+        .expect("good account");
+    let elapsed_reset = Utc::now().timestamp() - 60;
+    store
+        .compare_and_swap_quota(QuotaObservation {
+            account_id: good.id().clone(),
+            expected_revision: good.revision(),
+            quota: Some(OpaqueProviderData::new(
+                json!({
+                    "rate_limit": {
+                        "allowed": false,
+                        "limit_reached": true,
+                        "primary_window": {"used_percent": 100, "reset_at": elapsed_reset}
+                    }
+                })
+                .as_object()
+                .expect("recorded exhausted quota")
+                .clone(),
+            )),
+            observed_at: Some(SystemTime::now()),
+            limit_reached: Some(true),
+        })
+        .await
+        .expect("seed good account old reset");
     let server = MockServer::start().await;
+    let next_reset = Utc::now().timestamp() + 3600;
     Mock::given(method("GET"))
         .and(path("/api/codex/usage"))
         .and(header("authorization", "Bearer token-acct_quota_batch_bad"))
@@ -970,7 +1541,7 @@ async fn invalid_quota_json_for_one_account_does_not_abort_the_batch() {
         .respond_with(ResponseTemplate::new(200).set_body_json(json!({
             "rate_limit": {
                 "allowed": true,
-                "primary_window": {"used_percent": 25.0, "reset_at": 1_900_000_000}
+                "primary_window": {"used_percent": 25.0, "reset_at": next_reset}
             }
         })))
         .mount(&server)

backend/crates/providers/openai/tests/credential/quota/recovery.rs

@@ -12,6 +12,7 @@ async fn manual_quota_refresh_keeps_confirmed_exhaustion_until_the_recorded_rese
     let account = store.account(account_id).expect("created account");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::Expired,
@@ -161,6 +162,7 @@ async fn expired_old_reset_with_fresh_allowed_snapshot_recovers_ready() {
         .expect("seed old exhausted snapshot");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: account.id().clone(),
             expected_revision: account.revision(),
             availability: AccountAvailability::QuotaExhausted,
@@ -209,3 +211,47 @@ async fn expired_old_reset_with_fresh_allowed_snapshot_recovers_ready() {
         "old reset elapsed + fresh allowed snapshot must recover QuotaExhausted"
     );
 }
+
+#[tokio::test]
+async fn deactivated_workspace_quota_response_persists_the_upstream_reason() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let account_id = "acct_deactivated_workspace";
+    create_account(&store, account_id).await;
+    let account = store.account(account_id).expect("created account");
+    let server = MockServer::start().await;
+    Mock::given(method("GET"))
+        .and(path("/api/codex/usage"))
+        .and(header(
+            "authorization",
+            format!("Bearer token-{account_id}"),
+        ))
+        .respond_with(ResponseTemplate::new(402).set_body_json(json!({
+            "detail": {
+                "code": "deactivated_workspace",
+                "message": "workspace disabled"
+            }
+        })))
+        .mount(&server)
+        .await;
+    let service = quota_service_with_base_url(
+        &store,
+        reqwest::Client::builder()
+            .no_proxy()
+            .build()
+            .expect("client"),
+        server.uri(),
+    );
+
+    assert!(service.refresh_account(account.id()).await.is_err());
+    assert_eq!(
+        store
+            .account(account_id)
+            .expect("account after deactivation")
+            .availability(),
+        AccountAvailability::Banned
+    );
+    assert_eq!(
+        store.last_error_message(account_id).as_deref(),
+        Some("deactivated_workspace")
+    );
+}

backend/crates/providers/openai/tests/credential/quota/snapshot.rs

@@ -305,7 +305,7 @@ async fn scheduling_signal_survives_limit_reached_without_percent_or_reset() {
 }
 
 #[tokio::test]
-async fn persisted_codex_additional_limit_replaces_the_top_level_rate_limit() {
+async fn persisted_codex_alias_keeps_the_top_level_rate_limit_canonical() {
     let store = Arc::new(MemoryAccountStore::default());
     create_account(&store, "acct_canonical_codex_limit").await;
     let account = store
@@ -323,7 +323,7 @@ async fn persisted_codex_additional_limit_replaces_the_top_level_rate_limit() {
             "metered_feature": "codex",
             "rate_limit": {
                 "primary_window": {
-                    "used_percent": 2,
+                    "used_percent": 98,
                     "reset_at": 1_900_000_000,
                     "limit_window_seconds": 2_592_000
                 }
@@ -355,10 +355,13 @@ async fn persisted_codex_additional_limit_replaces_the_top_level_rate_limit() {
         .filter(|window| window.kind() == CodexQuotaWindowKind::Monthly)
         .collect::<Vec<_>>();
 
-    assert_eq!(snapshot.fact().remaining_percent(), Some(98));
+    assert_eq!(snapshot.fact().remaining_percent(), Some(9));
     assert_eq!(monthly.len(), 1);
-    assert_eq!(monthly[0].source(), "core");
-    assert_eq!(monthly[0].used_percent(), Some(2.0));
+    assert!(
+        monthly
+            .iter()
+            .any(|window| { window.source() == "codex" && window.used_percent() == Some(91.0) })
+    );
 }
 
 #[tokio::test]
@@ -419,16 +422,21 @@ async fn code_review_limit_projects_as_independent_window_with_limit_name() {
         .expect("read quota")
         .expect("quota snapshot");
 
-    // spend_control 不再生成窗口（只作 exhaustion 信号），
-    // 顶层 code_review_rate_limit 与 additional 的 code_review 桶同源替换后保留一个。
+    // spend_control 不再生成窗口（只作 exhaustion 信号）；顶层和 additional
+    // 的 code_review 桶分别保留，避免一个快照覆盖另一个。
     let review = snapshot
         .windows()
         .iter()
         .filter(|window| window.source() == "code_review")
         .collect::<Vec<_>>();
-    assert_eq!(review.len(), 1);
-    assert_eq!(review[0].limit_name(), Some("code review"));
-    assert_eq!(review[0].used_percent(), Some(55.0));
-    assert_eq!(snapshot.windows().len(), 2);
+    assert_eq!(review.len(), 2);
+    assert!(review.iter().any(|window| {
+        window.limit_name() == Some("code_review") && window.used_percent() == Some(80.0)
+    }));
+    assert!(review.iter().any(|window| {
+        window.limit_name() == Some("code review") && window.used_percent() == Some(55.0)
+    }));
+    assert_ne!(review[0].key(), review[1].key());
+    assert_eq!(snapshot.windows().len(), 3);
     assert!(!snapshot.fact().exhausted());
 }

backend/crates/providers/openai/tests/credential/refresh.rs

@@ -17,8 +17,11 @@ use provider_openai::credential::{
 };
 use secrecy::{ExposeSecret, SecretString};
 
+use super::{OAuthRecoveryLogCapture, lock_oauth_recovery_log_capture};
 use crate::support::{MemoryAccountStore, profile, runtime_policy, secret};
 
+const OAUTH_BACKOFF_ATTEMPTS: u32 = 5;
+
 struct Refresher {
     outcomes: Mutex<VecDeque<Result<TokenPair, RefreshFailure>>>,
     seen: Mutex<Vec<String>>,
@@ -377,7 +380,88 @@ async fn successful_refresh_uses_redis_lease_and_cas_rotates_plaintext_tokens()
 }
 
 #[tokio::test]
-async fn refreshed_identity_rejection_revision_fences_account_as_invalid() {
+async fn scheduled_refresh_preserves_existing_refresh_token_when_exchange_omits_one() {
+    let (store, _, service) = setup(
+        Ok(TokenPair {
+            access_token: "scheduled-access-without-rotated-refresh".to_owned(),
+            refresh_token: None,
+            expires_in: Duration::from_secs(3_600),
+        }),
+        true,
+    )
+    .await;
+
+    let outcomes = service.refresh_due().await.expect("scheduled refresh");
+
+    assert!(matches!(
+        outcomes.as_slice(),
+        [CodexCredentialRefreshOutcome::Refreshed { account_id, .. }] if account_id == "acct_refresh"
+    ));
+    let account = store.account("acct_refresh").expect("rotated account");
+    assert_eq!(account.revision().get(), 2);
+    let runtime = store
+        .repository()
+        .load_runtime_credential(&account)
+        .await
+        .expect("rotated credential");
+    assert_eq!(
+        runtime
+            .authentication
+            .oauth()
+            .expect("OAuth credential")
+            .access_token
+            .expose_secret(),
+        "scheduled-access-without-rotated-refresh"
+    );
+    assert_eq!(
+        runtime
+            .authentication
+            .oauth()
+            .expect("OAuth credential")
+            .refresh_token
+            .as_ref()
+            .expect("preserved refresh token")
+            .expose_secret(),
+        "rt-old-access"
+    );
+}
+
+#[tokio::test]
+async fn invalid_refresh_response_is_logged_before_local_rejection() {
+    let _capture_lock = lock_oauth_recovery_log_capture().await;
+    let (_, _, service) = setup(
+        Ok(TokenPair {
+            access_token: "rejected-access".to_owned(),
+            refresh_token: None,
+            expires_in: Duration::ZERO,
+        }),
+        true,
+    )
+    .await;
+    let recovery_log = OAuthRecoveryLogCapture::default();
+    let dispatch = recovery_log.dispatch();
+    let guard = tracing::dispatcher::set_default(&dispatch);
+
+    let outcomes = service
+        .refresh_due()
+        .await
+        .expect("refresh cycle completes");
+    drop(guard);
+
+    assert!(matches!(
+        outcomes.as_slice(),
+        [CodexCredentialRefreshOutcome::Failed { account_id }] if account_id == "acct_refresh"
+    ));
+    let record = recovery_log.recovery_record();
+    assert_eq!(record["fields"]["operation"], "scheduled_refresh");
+    assert_eq!(record["fields"]["account_id"], "acct_refresh");
+    assert_eq!(record["fields"]["access_token"], "rejected-access");
+    assert_eq!(record["fields"]["refresh_token"], "rt-old-access");
+    assert_eq!(record["fields"]["has_refresh_token"], true);
+}
+
+#[tokio::test]
+async fn refreshed_tokens_persist_when_identity_verification_rejects() {
     let (store, _, _) = setup(Ok(success_tokens()), true).await;
     let service = CodexCredentialRefreshService::new(
         store.repository(),
@@ -395,19 +479,31 @@ async fn refreshed_identity_rejection_revision_fences_account_as_invalid() {
 
     assert!(matches!(
         outcomes.as_slice(),
-        [CodexCredentialRefreshOutcome::Invalidated { .. }]
+        [CodexCredentialRefreshOutcome::Refreshed {
+            account_id,
+            credential_revision: 2
+        }] if account_id == "acct_refresh"
     ));
+    let account = store.account("acct_refresh").expect("refreshed account");
+    assert_eq!(account.availability(), AccountAvailability::Ready);
+    let runtime = store
+        .repository()
+        .load_runtime_credential(&account)
+        .await
+        .expect("refreshed credential");
     assert_eq!(
-        store
-            .account("acct_refresh")
-            .expect("invalid account")
-            .availability(),
-        AccountAvailability::Invalid
+        runtime
+            .authentication
+            .oauth()
+            .expect("OAuth credential")
+            .access_token
+            .expose_secret(),
+        "new-access"
     );
 }
 
 #[tokio::test]
-async fn unavailable_signature_verification_persists_refresh_backoff() {
+async fn refreshed_tokens_persist_when_identity_verification_is_unavailable() {
     let (store, _, _) = setup(Ok(success_tokens()), true).await;
     let service = CodexCredentialRefreshService::new(
         store.repository(),
@@ -426,11 +522,10 @@ async fn unavailable_signature_verification_persists_refresh_backoff() {
 
     assert!(matches!(
         outcomes.as_slice(),
-        [CodexCredentialRefreshOutcome::Transient { .. }]
+        [CodexCredentialRefreshOutcome::Refreshed { .. }]
     ));
-    let account = store.account("acct_refresh").expect("deferred account");
+    let account = store.account("acct_refresh").expect("refreshed account");
     assert_eq!(account.revision().get(), 2);
-    // JWKS/签名边界短暂不可用是瞬态：账号保持可用、不被永久失效，仅推进退避重试。
     assert_eq!(account.availability(), AccountAvailability::Ready);
     assert!(
         account
@@ -440,7 +535,7 @@ async fn unavailable_signature_verification_persists_refresh_backoff() {
 }
 
 #[tokio::test]
-async fn unavailable_usage_preserves_rotated_tokens_and_persists_backoff() {
+async fn refreshed_tokens_do_not_require_identity_completion() {
     let store = Arc::new(MemoryAccountStore::default());
     let signed = super::identity::signed_identity_fixture().await;
     let mut account_profile = profile("account-signed");
@@ -474,7 +569,7 @@ async fn unavailable_usage_preserves_rotated_tokens_and_persists_backoff() {
 
     assert!(matches!(
         outcomes.as_slice(),
-        [CodexCredentialRefreshOutcome::Transient { .. }]
+        [CodexCredentialRefreshOutcome::Refreshed { .. }]
     ));
     let account = store.account("acct_signed_only").expect("rotated account");
     assert_eq!(account.revision().get(), 2);
@@ -553,6 +648,10 @@ async fn upstream_ban_marks_unified_account_banned() {
             .availability(),
         AccountAvailability::Banned
     );
+    assert_eq!(
+        store.last_error_message("acct_refresh").as_deref(),
+        Some("account_banned")
+    );
 }
 
 #[tokio::test]
@@ -597,7 +696,7 @@ async fn transient_refresh_defers_without_invalidating_account() {
 }
 
 #[tokio::test]
-async fn expired_access_token_refreshes_even_when_next_refresh_is_deferred() {
+async fn expired_access_token_respects_persisted_refresh_backoff() {
     let store = Arc::new(MemoryAccountStore::default());
     seed_due_account(&store).await;
     let original = store.account("acct_refresh").expect("seeded account");
@@ -615,16 +714,6 @@ async fn expired_access_token_refreshes_even_when_next_refresh_is_deferred() {
         })
         .await
         .expect("seed expired access token with deferred refresh");
-    let deferred = store.account("acct_refresh").expect("deferred account");
-    store
-        .repository()
-        .apply_state(
-            &deferred,
-            AccountAvailability::QuotaExhausted,
-            SystemTime::now(),
-        )
-        .await
-        .expect("mark account quota exhausted");
     let refresher = Arc::new(Refresher::new(Ok(success_tokens())));
     let service = CodexCredentialRefreshService::new(
         store.repository(),
@@ -640,31 +729,19 @@ async fn expired_access_token_refreshes_even_when_next_refresh_is_deferred() {
 
     let outcomes = service.refresh_due().await.expect("refresh due");
 
-    assert!(matches!(
-        outcomes.as_slice(),
-        [CodexCredentialRefreshOutcome::Refreshed {
-            account_id,
-            credential_revision: 3,
-        }] if account_id == "acct_refresh"
-    ));
-    assert_eq!(
-        refresher.seen.lock().expect("seen tokens lock").as_slice(),
-        ["rt-old-access"]
-    );
-    let refreshed = store.account("acct_refresh").expect("refreshed account");
-    assert_eq!(
-        refreshed.availability(),
-        AccountAvailability::QuotaExhausted
-    );
+    assert!(outcomes.is_empty());
+    assert!(refresher.seen.lock().expect("seen tokens lock").is_empty());
     assert!(
-        refreshed
-            .access_token_expires_at()
-            .is_some_and(|expires_at| expires_at > SystemTime::now())
+        store
+            .account("acct_refresh")
+            .expect("deferred account")
+            .next_refresh_at()
+            .is_some_and(|next_refresh_at| next_refresh_at > SystemTime::now())
     );
 }
 
 #[tokio::test]
-async fn expired_access_token_refreshes_during_active_cooldown() {
+async fn expired_access_token_refreshes_when_persisted_refresh_is_due() {
     let store = Arc::new(MemoryAccountStore::default());
     seed_due_account(&store).await;
     let original = store.account("acct_refresh").expect("seeded account");
@@ -678,7 +755,7 @@ async fn expired_access_token_refreshes_during_active_cooldown() {
             expected_credential_revision: original.revision().get(),
             secret: secret("old-access"),
             verified_account: expired_profile,
-            next_refresh_at: Some(chrono::Utc::now() + chrono::Duration::hours(12)),
+            next_refresh_at: Some(chrono::Utc::now() - chrono::Duration::seconds(1)),
         })
         .await
         .expect("seed expired access token with deferred refresh");
@@ -825,6 +902,15 @@ async fn seed_due_account(store: &MemoryAccountStore) {
         .await;
 }
 
+async fn force_due_account(store: &Arc<MemoryAccountStore>) {
+    let account = store.account("acct_refresh").expect("account to reset");
+    store
+        .repository()
+        .defer_refresh(&account, SystemTime::now() - Duration::from_secs(1))
+        .await
+        .expect("reset next_refresh_at to the past");
+}
+
 #[tokio::test]
 async fn refresh_backoff_grows_exponentially_across_attempts() {
     let store = Arc::new(MemoryAccountStore::default());
@@ -891,6 +977,175 @@ async fn refresh_backoff_grows_exponentially_across_attempts() {
     );
 }
 
+#[tokio::test]
+async fn exhausted_oauth_backoff_continues_after_access_token_expiry() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let mut expired_profile = profile("chatgpt-acct_refresh");
+    expired_profile.access_token_expires_at =
+        Some(chrono::Utc::now() - chrono::Duration::minutes(1));
+    store
+        .seed_oauth_credential(ImportCodexOAuthCredential {
+            account_id: "acct_refresh".to_owned(),
+            name: "refresh".to_owned(),
+            secret: secret("old-access"),
+            verified_account: expired_profile,
+            next_refresh_at: Some(chrono::Utc::now() - chrono::Duration::seconds(1)),
+            enabled: true,
+        })
+        .await;
+    let credential_state = Arc::new(CountingCredentialState::default());
+    let state_port: Arc<dyn ProviderCredentialStatePort> = credential_state.clone();
+    let refresher = Arc::new(Refresher::scripted([
+        Err(RefreshFailure::Transport),
+        Err(RefreshFailure::Transport),
+        Err(RefreshFailure::Transport),
+        Err(RefreshFailure::Transport),
+        Err(RefreshFailure::Transport),
+        Err(RefreshFailure::Transport),
+    ]));
+    let service = CodexCredentialRefreshService::new(
+        store.repository(),
+        refresher.clone(),
+        Arc::new(VerifiedIdentity),
+        Arc::new(RefreshLeases {
+            available: true,
+            requests: Mutex::new(Vec::new()),
+        }),
+        state_port,
+        runtime_policy(),
+    );
+
+    for _ in 0..OAUTH_BACKOFF_ATTEMPTS {
+        service.refresh_due().await.expect("backoff attempt");
+        force_due_account(&store).await;
+    }
+
+    let observed_at = SystemTime::now();
+    service.refresh_due().await.expect("recovery attempt");
+    let recovery_delay = store
+        .account("acct_refresh")
+        .expect("account after recovery defer")
+        .next_refresh_at()
+        .expect("recovery scheduled")
+        .duration_since(observed_at)
+        .expect("recovery is in the future");
+    assert!(
+        (Duration::from_secs(8 * 60)..=Duration::from_secs(12 * 60 + 1)).contains(&recovery_delay),
+        "recovery delay should be about ten minutes, got {recovery_delay:?}"
+    );
+    assert_eq!(
+        credential_state.count("acct_refresh"),
+        OAUTH_BACKOFF_ATTEMPTS + 1
+    );
+    assert_eq!(refresher.seen.lock().expect("seen tokens lock").len(), 6);
+}
+
+#[tokio::test]
+async fn oauth_backoff_never_passes_an_unexpired_access_token_expiry() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let mut account_profile = profile("chatgpt-acct_refresh");
+    account_profile.access_token_expires_at =
+        Some(chrono::Utc::now() + chrono::Duration::seconds(30));
+    let access_token_expires_at: SystemTime = account_profile
+        .access_token_expires_at
+        .expect("access expiry")
+        .into();
+    store
+        .seed_oauth_credential(ImportCodexOAuthCredential {
+            account_id: "acct_refresh".to_owned(),
+            name: "refresh".to_owned(),
+            secret: secret("old-access"),
+            verified_account: account_profile,
+            next_refresh_at: Some(chrono::Utc::now() - chrono::Duration::seconds(1)),
+            enabled: true,
+        })
+        .await;
+    let credential_state = Arc::new(CountingCredentialState::default());
+    let state_port: Arc<dyn ProviderCredentialStatePort> = credential_state.clone();
+    let service = CodexCredentialRefreshService::new(
+        store.repository(),
+        Arc::new(Refresher::scripted([
+            Err(RefreshFailure::Transport),
+            Err(RefreshFailure::Transport),
+            Err(RefreshFailure::Transport),
+            Err(RefreshFailure::Transport),
+            Err(RefreshFailure::Transport),
+            Err(RefreshFailure::Transport),
+        ])),
+        Arc::new(VerifiedIdentity),
+        Arc::new(RefreshLeases {
+            available: true,
+            requests: Mutex::new(Vec::new()),
+        }),
+        state_port,
+        runtime_policy(),
+    );
+
+    for attempt in 0..=OAUTH_BACKOFF_ATTEMPTS {
+        service.refresh_due().await.expect("backoff attempt");
+        if attempt < OAUTH_BACKOFF_ATTEMPTS {
+            force_due_account(&store).await;
+        }
+    }
+
+    assert_eq!(
+        store
+            .account("acct_refresh")
+            .expect("account after capped retry")
+            .next_refresh_at(),
+        Some(access_token_expires_at)
+    );
+    assert_eq!(
+        credential_state.count("acct_refresh"),
+        OAUTH_BACKOFF_ATTEMPTS + 1
+    );
+}
+
+#[tokio::test]
+async fn recovery_window_exhaustion_marks_account_expired_without_token_exchange() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let mut expired_profile = profile("chatgpt-acct_refresh");
+    expired_profile.access_token_expires_at =
+        Some(chrono::Utc::now() - chrono::Duration::hours(25));
+    store
+        .seed_oauth_credential(ImportCodexOAuthCredential {
+            account_id: "acct_refresh".to_owned(),
+            name: "refresh".to_owned(),
+            secret: secret("old-access"),
+            verified_account: expired_profile,
+            next_refresh_at: Some(chrono::Utc::now() + chrono::Duration::hours(12)),
+            enabled: true,
+        })
+        .await;
+    let refresher = Arc::new(Refresher::new(Ok(success_tokens())));
+    let service = CodexCredentialRefreshService::new(
+        store.repository(),
+        refresher.clone(),
+        Arc::new(VerifiedIdentity),
+        Arc::new(RefreshLeases {
+            available: true,
+            requests: Mutex::new(Vec::new()),
+        }),
+        Arc::new(CountingCredentialState::default()),
+        runtime_policy(),
+    );
+
+    let outcomes = service.refresh_due().await.expect("refresh due");
+
+    assert!(matches!(
+        outcomes.as_slice(),
+        [CodexCredentialRefreshOutcome::Invalidated { account_id }] if account_id == "acct_refresh"
+    ));
+    assert!(refresher.seen.lock().expect("seen tokens lock").is_empty());
+    assert_eq!(
+        store
+            .account("acct_refresh")
+            .expect("expired account")
+            .availability(),
+        AccountAvailability::Expired
+    );
+}
+
 #[tokio::test]
 async fn successful_refresh_clears_backoff_counter() {
     let store = Arc::new(MemoryAccountStore::default());

backend/crates/providers/openai/tests/credential/token_client.rs

@@ -75,6 +75,28 @@ async fn bounded_oauth_response_should_parse_lifetime_and_rotated_token() {
     assert_eq!(tokens.expires_in, Duration::from_secs(3600));
 }
 
+#[tokio::test]
+async fn refresh_response_with_blank_rotated_token_keeps_the_existing_token_eligible() {
+    let server = MockServer::start().await;
+    Mock::given(method("POST"))
+        .and(path("/oauth/token"))
+        .respond_with(ResponseTemplate::new(200).set_body_json(serde_json::json!({
+            "access_token": "access-rotated",
+            "refresh_token": " ",
+            "expires_in": 3600
+        })))
+        .expect(1)
+        .mount(&server)
+        .await;
+
+    let tokens = client(&server)
+        .refresh("refresh-initial")
+        .await
+        .expect("blank rotation field is treated as absent");
+
+    assert_eq!(tokens.refresh_token, None);
+}
+
 #[tokio::test]
 async fn refresh_should_exchange_the_official_form_fields() {
     let server = MockServer::start().await;

backend/crates/providers/openai/tests/provider/contract.rs

@@ -43,8 +43,9 @@ use wiremock::matchers::{header, method, path};
 use wiremock::{Mock, MockServer, ResponseTemplate};
 
 use crate::support::{
-    MemoryAccountStore, MemorySessionAffinity, MemorySessionExclusions, TestLeaseCoordinator,
-    account_policy, agent_identity_service_with_pool, catalog_cache, profile, secret,
+    MemoryAccountStore, MemoryCooldownPort, MemorySessionAffinity, MemorySessionExclusions,
+    TestLeaseCoordinator, account_policy, agent_identity_service_with_pool, catalog_cache, profile,
+    secret,
 };
 use crate::transport::accept_codex_test_websocket;
 
@@ -115,6 +116,21 @@ fn provider_with_affinity_and_base_url_and_leases(
     base_url: String,
     leases: Arc<TestLeaseCoordinator>,
 ) -> CodexProvider {
+    provider_and_quota_with_affinity_and_base_url_and_leases(
+        store,
+        session_affinity,
+        base_url,
+        leases,
+    )
+    .0
+}
+
+fn provider_and_quota_with_affinity_and_base_url_and_leases(
+    store: &Arc<MemoryAccountStore>,
+    session_affinity: Arc<MemorySessionAffinity>,
+    base_url: String,
+    leases: Arc<TestLeaseCoordinator>,
+) -> (CodexProvider, Arc<CodexCredentialQuotaService>) {
     let profile = wire_profile();
     let http = reqwest::Client::builder()
         .no_proxy()
@@ -136,6 +152,7 @@ fn provider_with_affinity_and_base_url_and_leases(
         http.clone(),
         base_url.clone(),
         Arc::clone(&agent_identity),
+        Arc::new(MemoryCooldownPort::new()),
     ));
     let account_feedback = Arc::new(AccountFeedbackStats::default());
     let selector = Arc::new(CodexCredentialSelector::new(
@@ -152,10 +169,10 @@ fn provider_with_affinity_and_base_url_and_leases(
         true,
     ));
 
-    CodexProvider::new(
+    let provider = CodexProvider::new(
         selector,
         catalog,
-        quota,
+        Arc::clone(&quota),
         agent_identity,
         account_feedback,
         http,
@@ -163,7 +180,8 @@ fn provider_with_affinity_and_base_url_and_leases(
         base_url,
         websocket_pool,
     )
-    .expect("official OpenAI provider")
+    .expect("official OpenAI provider");
+    (provider, quota)
 }
 
 async fn create_account(store: &Arc<MemoryAccountStore>, id: &str) {
@@ -196,6 +214,27 @@ fn generate_operation() -> Operation {
     ))
 }
 
+fn generate_with_session_context(
+    session_id: &str,
+    thread_id: Option<&str>,
+    turn_metadata: Option<&str>,
+) -> GenerateRequest {
+    let mut body = Map::from_iter([
+        ("model".to_owned(), json!("gpt-5.4")),
+        ("input".to_owned(), json!("hello")),
+        ("session_id".to_owned(), json!(session_id)),
+    ]);
+    if let Some(thread_id) = thread_id {
+        body.insert("thread_id".to_owned(), json!(thread_id));
+    }
+    if let Some(turn_metadata) = turn_metadata {
+        body.insert("turnMetadata".to_owned(), json!(turn_metadata));
+    }
+    GenerateRequest::from_protocol_payload(
+        ProtocolPayload::json_object("openai", body).expect("OpenAI payload"),
+    )
+}
+
 fn http_generate_operation() -> Operation {
     let payload = ProtocolPayload::json_object(
         "openai",
@@ -1053,6 +1092,66 @@ async fn prompt_cache_key_should_become_an_opaque_session_affinity_lookup_key()
     assert!(keys[0].bytes().all(|byte| byte.is_ascii_hexdigit()));
 }
 
+#[tokio::test]
+async fn subagent_requests_should_use_an_affinity_key_isolated_from_the_root_session() {
+    let store = Arc::new(MemoryAccountStore::default());
+    create_account(&store, "acct_subagent").await;
+    let affinity = Arc::new(MemorySessionAffinity::default());
+    let server = MockServer::start().await;
+    Mock::given(method("POST"))
+        .and(path("/codex/responses"))
+        .respond_with(
+            ResponseTemplate::new(200)
+                .insert_header("content-type", "text/event-stream")
+                .set_body_string(CAPTURE_COMPLETED_SSE),
+        )
+        .mount(&server)
+        .await;
+    let provider = provider_with_affinity_and_base_url(&store, Arc::clone(&affinity), server.uri());
+
+    for (request_id, subagent_kind) in [
+        ("req_root_affinity", None),
+        ("req_subagent_affinity_first", Some("review")),
+        ("req_subagent_affinity_second", Some("review")),
+    ] {
+        let mut body = Map::from_iter([
+            ("model".to_owned(), json!("gpt-5.4")),
+            ("input".to_owned(), json!("new task")),
+            ("prompt_cache_key".to_owned(), json!("root-session-key")),
+        ]);
+        if let Some(subagent_kind) = subagent_kind {
+            body.insert(
+                "client_metadata".to_owned(),
+                json!({"x-openai-subagent": subagent_kind}),
+            );
+        }
+        let generation = GenerateRequest::from_protocol_payload(
+            ProtocolPayload::json_object("openai", body)
+                .expect("OpenAI payload")
+                .with_context(Map::from_iter([("use_websocket".to_owned(), json!(false))])),
+        );
+        let mut stream = provider
+            .execute(
+                planned_request("openai", Operation::Generate(generation)),
+                context(request_id, CancellationToken::new()),
+            )
+            .await
+            .expect("prepare subagent provider stream");
+        while let Some(event) = stream.next().await {
+            event.expect("subagent response");
+        }
+        drop(stream);
+    }
+
+    let keys = affinity.lookup_keys();
+    assert_eq!(keys.len(), 3);
+    assert_ne!(keys[0], keys[1], "subagent must not inherit the root key");
+    assert_eq!(
+        keys[1], keys[2],
+        "the same subagent should retain its own key"
+    );
+}
+
 #[tokio::test]
 async fn explicit_session_id_should_override_turn_specific_prompt_cache_keys_for_affinity() {
     let store = Arc::new(MemoryAccountStore::default());
@@ -1095,6 +1194,58 @@ async fn explicit_session_id_should_override_turn_specific_prompt_cache_keys_for
     assert_eq!(keys[0], keys[1]);
 }
 
+#[tokio::test]
+async fn thread_spawn_children_should_have_independent_session_affinity_keys() {
+    let store = Arc::new(MemoryAccountStore::default());
+    create_account(&store, "acct_thread_spawn_affinity").await;
+    let affinity = Arc::new(MemorySessionAffinity::default());
+    let provider = provider_with_affinity(&store, Arc::clone(&affinity));
+    let thread_spawn = r#"{"subagent_kind":"thread_spawn"}"#;
+
+    for (request_id, thread_id, turn_metadata) in [
+        (
+            "req_thread_spawn_first",
+            Some("child-one"),
+            Some(thread_spawn),
+        ),
+        (
+            "req_thread_spawn_second",
+            Some("child-two"),
+            Some(thread_spawn),
+        ),
+        (
+            "req_thread_spawn_repeat",
+            Some("child-one"),
+            Some(thread_spawn),
+        ),
+        ("req_thread_spawn_fallback", None, Some(thread_spawn)),
+        ("req_thread_spawn_parent", None, None),
+    ] {
+        let stream = provider
+            .execute(
+                planned_request(
+                    "openai",
+                    Operation::Generate(generate_with_session_context(
+                        "parent-session",
+                        thread_id,
+                        turn_metadata,
+                    )),
+                ),
+                context(request_id, CancellationToken::new()),
+            )
+            .await
+            .expect("prepare provider stream");
+        drop(stream);
+    }
+
+    let keys = affinity.lookup_keys();
+    assert_eq!(keys.len(), 5);
+    assert_ne!(keys[0], keys[1]);
+    assert_eq!(keys[0], keys[2]);
+    assert_ne!(keys[3], keys[4]);
+    assert_ne!(keys[0], keys[3]);
+}
+
 #[tokio::test]
 async fn invalid_local_conversation_id_should_fall_back_to_an_opaque_affinity_key() {
     let store = Arc::new(MemoryAccountStore::default());
@@ -1271,9 +1422,10 @@ async fn response_failed_before_semantic_output_is_atomic_and_persists_quota_loc
 }
 
 #[tokio::test]
-async fn usage_limit_failure_returns_promptly_and_refreshes_the_authoritative_quota_snapshot() {
+async fn official_usage_limit_failure_projects_100_percent_and_holds_until_reset_advances() {
     let store = Arc::new(MemoryAccountStore::default());
     let account_id = "acct_usage_limit_request_path";
+    let reset_at = 1_900_000_000;
     create_account(&store, account_id).await;
     let account = store.account(account_id).expect("created account");
     store
@@ -1285,7 +1437,8 @@ async fn usage_limit_failure_returns_promptly_and_refreshes_the_authoritative_qu
                     "rate_limit": {
                         "allowed": true,
                         "limit_reached": false,
-                        "primary_window": {"used_percent": 99, "reset_at": 1_900_000_000}
+                        "primary_window": {"used_percent": 99, "reset_at": reset_at},
+                        "secondary_window": {"used_percent": 0}
                     }
                 })
                 .as_object()
@@ -1301,8 +1454,8 @@ async fn usage_limit_failure_returns_promptly_and_refreshes_the_authoritative_qu
     Mock::given(method("GET"))
         .and(path("/api/codex/usage"))
         .and(header("authorization", format!("Bearer at-{account_id}")))
-        // 生产环境会短暂返回互相矛盾的快照：数字已到 100%，但布尔位仍声称可用。
-        // 后台同步应覆盖展示数据，不能据此撤销真实请求已经确认的额度耗尽。
+        // 生产环境会短暂返回互相矛盾的快照：98/99% 且布尔位仍声称可用。
+        // 后台同步不能覆盖真实请求已确认的额度耗尽。
         .respond_with(
             ResponseTemplate::new(200)
                 // 验证后台 usage 同步不能把原始的额度错误响应拖到查询完成之后。
@@ -1311,7 +1464,8 @@ async fn usage_limit_failure_returns_promptly_and_refreshes_the_authoritative_qu
                     "rate_limit": {
                         "allowed": true,
                         "limit_reached": false,
-                        "primary_window": {"used_percent": 100, "reset_at": 1_900_000_000}
+                        "primary_window": {"used_percent": 99, "reset_at": reset_at},
+                        "secondary_window": {"used_percent": 0}
                     }
                 })),
         )
@@ -1320,16 +1474,13 @@ async fn usage_limit_failure_returns_promptly_and_refreshes_the_authoritative_qu
         .await;
     Mock::given(method("POST"))
         .and(path("/codex/responses"))
-        .respond_with(
-            ResponseTemplate::new(200)
-                .insert_header("content-type", "text/event-stream")
-                .set_body_string(concat!(
-                    "event: response.created\n",
-                    "data: {\"type\":\"response.created\",\"response\":{\"id\":\"resp_usage_limit_confirm\",\"model\":\"gpt-5.4\",\"status\":\"in_progress\"}}\n\n",
-                    "event: response.failed\n",
-                    "data: {\"type\":\"response.failed\",\"status_code\":429,\"retry_after_seconds\":86400,\"response\":{\"id\":\"resp_usage_limit_confirm\",\"status\":\"failed\",\"error\":{\"code\":\"usage_limit_reached\",\"message\":\"usage limit reached\"}}}\n\n"
-                )),
-        )
+        .respond_with(ResponseTemplate::new(429).set_body_json(json!({
+            "error": {
+                "type": "usage_limit_reached",
+                "message": "usage limit reached",
+                "resets_at": reset_at
+            }
+        })))
         .mount(&server)
         .await;
 
@@ -1355,6 +1506,30 @@ async fn usage_limit_failure_returns_promptly_and_refreshes_the_authoritative_qu
     assert_eq!(failure.kind(), ProviderErrorKind::QuotaExhausted);
     let account = store.account(account_id).expect("usage-limit account");
     assert_eq!(account.availability(), AccountAvailability::QuotaExhausted);
+    let projected = store
+        .get_quotas(&[account.id().clone()])
+        .await
+        .expect("read immediate quota projection")
+        .into_iter()
+        .next()
+        .expect("confirmed quota observation");
+    let projected_observed_at = projected.observed_at.expect("projection observed at");
+    let projected = Value::Object(
+        projected
+            .quota
+            .map(OpaqueProviderData::into_inner)
+            .expect("confirmed quota projection"),
+    );
+    assert_eq!(
+        projected
+            .pointer("/rate_limit/primary_window/used_percent")
+            .and_then(Value::as_u64),
+        Some(100)
+    );
+    assert!(
+        projected.pointer("/rate_limit/secondary_window").is_none(),
+        "request failure must not manufacture a secondary quota"
+    );
     tokio::time::sleep(Duration::from_millis(200)).await;
     let requests = server.received_requests().await.expect("received requests");
     assert_eq!(
@@ -1366,35 +1541,49 @@ async fn usage_limit_failure_returns_promptly_and_refreshes_the_authoritative_qu
         "usage refresh must wait for the upstream quota settlement delay"
     );
 
-    let quota = timeout(Duration::from_secs(5), async {
+    timeout(Duration::from_secs(5), async {
         loop {
-            let quota = store
+            let observed_at = store
                 .get_quotas(&[account.id().clone()])
                 .await
-                .expect("read refreshed quota")
+                .expect("read refreshed quota observation")
                 .into_iter()
                 .next()
-                .and_then(|observation| observation.quota)
-                .expect("authoritative quota snapshot");
-            let quota = Value::Object(quota.into_inner());
-            if quota
-                .pointer("/rate_limit/primary_window/used_percent")
-                .and_then(Value::as_u64)
-                == Some(100)
-            {
-                break quota;
+                .and_then(|observation| observation.observed_at);
+            if observed_at.is_some_and(|observed_at| observed_at > projected_observed_at) {
+                break;
             }
             tokio::time::sleep(Duration::from_millis(10)).await;
         }
     })
     .await
-    .expect("background usage refresh must replace the stale quota snapshot");
+    .expect("background usage refresh must run");
+    let quota = store
+        .get_quotas(&[account.id().clone()])
+        .await
+        .expect("read refreshed quota")
+        .into_iter()
+        .next()
+        .and_then(|observation| observation.quota)
+        .map(OpaqueProviderData::into_inner)
+        .map(Value::Object)
+        .expect("authoritative quota projection");
     assert_eq!(
         quota
             .pointer("/rate_limit/primary_window/used_percent")
             .and_then(Value::as_u64),
         Some(100)
     );
+    assert_eq!(
+        quota
+            .pointer("/rate_limit/primary_window/reset_at")
+            .and_then(Value::as_i64),
+        Some(reset_at)
+    );
+    assert!(
+        quota.pointer("/rate_limit/secondary_window").is_none(),
+        "background quota refresh must retain the same window projection"
+    );
     let requests = server.received_requests().await.expect("received requests");
     assert_eq!(
         requests
@@ -1849,6 +2038,85 @@ async fn disabled_account_diagnostic_uses_upstream_without_persisting_account_st
     assert_eq!(affinity.binding_count(), 0);
 }
 
+#[tokio::test]
+async fn quota_limited_account_diagnostic_uses_upstream() {
+    let store = Arc::new(MemoryAccountStore::default());
+    let account_id = "acct_quota_limited_diagnostic";
+    create_account(&store, account_id).await;
+    let account = store.account(account_id).expect("test account");
+    let raw_quota = json!({
+        "rate_limit": {
+            "allowed": false,
+            "limit_reached": true,
+            "primary_window": {"used_percent": 100, "reset_at": 1_900_000_000}
+        }
+    });
+    store
+        .compare_and_swap_quota(QuotaObservation {
+            account_id: account.id().clone(),
+            expected_revision: account.revision(),
+            quota: Some(OpaqueProviderData::new(
+                raw_quota
+                    .as_object()
+                    .expect("quota snapshot object")
+                    .clone(),
+            )),
+            observed_at: Some(SystemTime::now()),
+            limit_reached: Some(true),
+        })
+        .await
+        .expect("seed quota-limited snapshot");
+
+    let server = MockServer::start().await;
+    Mock::given(method("POST"))
+        .and(path("/codex/responses"))
+        .respond_with(
+            ResponseTemplate::new(200)
+                .insert_header("content-type", "text/event-stream")
+                .set_body_string(concat!(
+                    "event: response.created\n",
+                    "data: {\"type\":\"response.created\",\"response\":{\"id\":\"resp_quota_limited_diagnostic\",\"model\":\"gpt-5.4\"}}\n\n",
+                    "event: response.completed\n",
+                    "data: {\"type\":\"response.completed\",\"response\":{\"id\":\"resp_quota_limited_diagnostic\",\"model\":\"gpt-5.4\",\"status\":\"completed\",\"output\":[],\"usage\":{\"input_tokens\":1,\"output_tokens\":1,\"total_tokens\":2}}}\n\n"
+                )),
+        )
+        .expect(1)
+        .mount(&server)
+        .await;
+    let (provider, quota) = provider_and_quota_with_affinity_and_base_url_and_leases(
+        &store,
+        Arc::new(MemorySessionAffinity::default()),
+        server.uri(),
+        Arc::new(TestLeaseCoordinator::default()),
+    );
+    quota
+        .prepare_scheduling(std::slice::from_ref(&account))
+        .await;
+
+    let mut stream = provider
+        .execute(
+            planned_request("openai", http_generate_operation()),
+            diagnostic_context("req_quota_limited_diagnostic", account_id),
+        )
+        .await
+        .expect("quota-limited diagnostic should prepare a fixed-account stream");
+    let mut completed = false;
+    while let Some(event) = stream.next().await {
+        let event = event.expect("quota-limited diagnostic upstream response");
+        completed |= event
+            .canonical_facts()
+            .iter()
+            .any(|event| matches!(event, GatewayEvent::Completed(_)));
+    }
+
+    assert!(completed);
+    let requests = server
+        .received_requests()
+        .await
+        .expect("captured quota-limited diagnostic request");
+    assert_eq!(requests.len(), 1);
+}
+
 #[tokio::test]
 async fn successful_response_preserves_quota_exhaustion_from_rate_limit_headers() {
     let store = Arc::new(MemoryAccountStore::default());

backend/crates/providers/openai/tests/support.rs

@@ -16,9 +16,10 @@ use gateway_core::engine::credential::{
 };
 use gateway_core::error::{StoreError, StoreErrorKind};
 use gateway_core::provider_ports::{
-    ProviderCatalogCacheKey, ProviderCatalogCachePort, ProviderLeaseAcquisition, ProviderLeasePort,
-    ProviderLeaseRequest, ProviderRefreshPolicy, ProviderRuntimePolicyPort,
-    ProviderSchedulingLeaseRequest, ProviderSchedulingState, ProviderSessionAffinityKey,
+    ProviderCatalogCacheKey, ProviderCatalogCachePort, ProviderCooldown, ProviderCooldownPort,
+    ProviderCooldownScope, ProviderLeaseAcquisition, ProviderLeasePort, ProviderLeaseRequest,
+    ProviderRefreshPolicy, ProviderRuntimePolicyPort, ProviderSchedulingLeaseRequest,
+    ProviderSchedulingState, ProviderScopedCooldown, ProviderSessionAffinityKey,
     ProviderSessionAffinityPort, ProviderSessionExclusionPort, ProviderSessionExclusions,
     ProviderStoreError,
 };
@@ -37,6 +38,7 @@ struct StoredAccount {
     credential: gateway_core::engine::credential::PlaintextCredential,
     quota: Option<QuotaObservation>,
     state_observed_at: Option<SystemTime>,
+    last_error_message: Option<String>,
 }
 
 #[derive(Default)]
@@ -69,6 +71,15 @@ impl MemoryAccountStore {
             .map(|stored| stored.account.clone())
     }
 
+    pub(crate) fn last_error_message(&self, id: &str) -> Option<String> {
+        let id = ProviderAccountId::new(id).ok()?;
+        self.accounts
+            .lock()
+            .expect("account store lock")
+            .get(&id)
+            .and_then(|stored| stored.last_error_message.clone())
+    }
+
     pub(crate) async fn advance_credential_revision(&self, id: &str) -> CredentialRevision {
         let account_id = ProviderAccountId::new(id).expect("test account id");
         let current = self
@@ -147,6 +158,7 @@ impl ProviderAccountStore for MemoryAccountStore {
                 credential: input.credential,
                 quota: None,
                 state_observed_at: None,
+                last_error_message: None,
             },
         );
         Ok(())
@@ -320,6 +332,11 @@ impl ProviderAccountStore for MemoryAccountStore {
             },
         );
         stored.state_observed_at = Some(change.observed_at);
+        if change.availability == AccountAvailability::Ready {
+            stored.last_error_message = None;
+        } else if let Some(message) = change.message {
+            stored.last_error_message = Some(message);
+        }
         Ok(())
     }
 
@@ -771,3 +788,90 @@ pub(crate) fn codex_account(id: &str) -> ProviderAccount {
     .with_runtime_state(true, AccountAvailability::Ready)
     .with_refresh_schedule(true, Some(SystemTime::now() + Duration::from_secs(2_700)))
 }
+
+/// 内存 `ProviderCooldownPort`：只实现 `read`/`put_if_later`（openai selector
+/// 429 冷却路径用），其余 scope 变体测试不涉及，返回占位。
+#[derive(Clone, Default)]
+pub(crate) struct MemoryCooldownPort {
+    pub(crate) cooldowns: Arc<Mutex<BTreeMap<ProviderAccountId, ProviderCooldown>>>,
+}
+
+impl MemoryCooldownPort {
+    #[must_use]
+    pub(crate) fn new() -> Self {
+        Self::default()
+    }
+}
+
+impl ProviderCooldownPort for MemoryCooldownPort {
+    fn put_if_later(
+        &self,
+        cooldown: ProviderCooldown,
+    ) -> BoxFuture<'_, Result<bool, ProviderStoreError>> {
+        Box::pin(async move {
+            let mut cooldowns = self.cooldowns.lock().expect("cooldown lock");
+            let should_write = cooldowns.get(cooldown.account_id()).is_none_or(|current| {
+                current.credential_revision() < cooldown.credential_revision()
+                    || (current.credential_revision() == cooldown.credential_revision()
+                        && current.until() < cooldown.until())
+            });
+            if should_write {
+                cooldowns.insert(cooldown.account_id().clone(), cooldown);
+            }
+            Ok(should_write)
+        })
+    }
+
+    fn read<'a>(
+        &'a self,
+        account_id: &'a ProviderAccountId,
+    ) -> BoxFuture<'a, Result<Option<ProviderCooldown>, ProviderStoreError>> {
+        Box::pin(async move {
+            Ok(self
+                .cooldowns
+                .lock()
+                .expect("cooldown lock")
+                .get(account_id)
+                .cloned())
+        })
+    }
+
+    fn clear<'a>(
+        &'a self,
+        _account_id: &'a ProviderAccountId,
+        _through_revision: CredentialRevision,
+    ) -> BoxFuture<'a, Result<bool, ProviderStoreError>> {
+        Box::pin(async { Ok(false) })
+    }
+
+    fn put_scoped_if_later(
+        &self,
+        _cooldown: ProviderScopedCooldown,
+    ) -> BoxFuture<'_, Result<bool, ProviderStoreError>> {
+        Box::pin(async { Ok(false) })
+    }
+
+    fn read_scoped<'a>(
+        &'a self,
+        _account_id: &'a ProviderAccountId,
+        _scope: &'a ProviderCooldownScope,
+    ) -> BoxFuture<'a, Result<Option<ProviderScopedCooldown>, ProviderStoreError>> {
+        Box::pin(async { Ok(None) })
+    }
+
+    fn clear_scoped<'a>(
+        &'a self,
+        _account_id: &'a ProviderAccountId,
+        _scope: &'a ProviderCooldownScope,
+        _through_revision: CredentialRevision,
+    ) -> BoxFuture<'a, Result<bool, ProviderStoreError>> {
+        Box::pin(async { Ok(false) })
+    }
+
+    fn clear_all<'a>(
+        &'a self,
+        _account_id: &'a ProviderAccountId,
+    ) -> BoxFuture<'a, Result<bool, ProviderStoreError>> {
+        Box::pin(async { Ok(false) })
+    }
+}

backend/crates/providers/openai/tests/transport/catalog.rs

@@ -219,6 +219,10 @@ async fn fetch_should_send_official_catalog_headers_and_query() {
         .and(header("authorization", "Bearer oauth-access"))
         .and(header("chatgpt-account-id", "acct_123"))
         .and(header("originator", "codex_cli_rs"))
+        .and(header(
+            "user-agent",
+            "codex_cli_rs/0.144.0 (linux 6.8; x86_64) xterm (codex_cli_rs; 1.0.0)",
+        ))
         .and(header("accept", "application/json"))
         .and(header("x-codex-installation-id", "installation-123"))
         .respond_with(

backend/crates/providers/openai/tests/transport/diagnostics.rs

@@ -146,13 +146,13 @@ fn structured_expiry_code_classifies_without_an_auth_status() {
 }
 
 #[test]
-fn structured_quota_signal_takes_precedence_over_http_429() {
+fn http_429_without_usage_limit_type_remains_a_temporary_rate_limit() {
     assert_eq!(
         classify(
             429,
             r#"{"error":{"code":"insufficient_quota","message":"rate limit reached"}}"#
         ),
-        CodexFailureCategory::QuotaExhausted
+        CodexFailureCategory::RateLimited
     );
 }
 
@@ -168,13 +168,24 @@ fn structured_usage_limit_type_is_resettable_quota_exhaustion() {
 }
 
 #[test]
-fn usage_limit_message_without_a_type_is_resettable_quota_exhaustion() {
+fn http_429_usage_limit_message_without_a_type_remains_a_temporary_rate_limit() {
     assert_eq!(
         classify(
             429,
             r#"{"error":{"message":"The usage limit has been reached"}}"#
         ),
-        CodexFailureCategory::UsageLimitExhausted
+        CodexFailureCategory::RateLimited
+    );
+}
+
+#[test]
+fn http_429_usage_limit_code_without_a_type_remains_a_temporary_rate_limit() {
+    assert_eq!(
+        classify(
+            429,
+            r#"{"error":{"code":"usage_limit_reached","message":"The usage limit has been reached"}}"#
+        ),
+        CodexFailureCategory::RateLimited
     );
 }
 

backend/crates/providers/openai/tests/transport/headers.rs

@@ -789,6 +789,70 @@ async fn backend_http_should_zstd_compress_codex_responses_request_body() {
     assert_eq!(json["instructions"], "compress me");
 }
 
+#[tokio::test]
+async fn backend_http_should_force_upstream_sse_for_non_streaming_client_request() {
+    let listener = TcpListener::bind("127.0.0.1:0")
+        .await
+        .expect("bind non-streaming HTTP server");
+    let address = listener
+        .local_addr()
+        .expect("non-streaming HTTP server address");
+    let server = tokio::spawn(async move {
+        let (mut stream, _) = listener.accept().await.expect("accept HTTP client");
+        let request = read_http_request_with_body(&mut stream).await;
+        write_completed_sse_response(&mut stream).await;
+        request
+    });
+    let mut request = CodexResponsesRequest::from_body(
+        json!({
+            "model": "gpt-test",
+            "instructions": "collect this response",
+            "input": [],
+            "stream": false
+        })
+        .as_object()
+        .expect("request body")
+        .clone(),
+    );
+    request.force_http_sse = true;
+    assert!(
+        !request.stream(),
+        "client-facing preference remains non-streaming"
+    );
+    let client = CodexBackendClient::new(
+        reqwest::Client::builder()
+            .no_proxy()
+            .build()
+            .expect("HTTP client"),
+        format!("http://{address}"),
+        test_wire_profile(),
+    );
+
+    client
+        .create_response(
+            &request,
+            request_context("req_non_streaming_client", Some("acct-non-streaming")),
+        )
+        .await
+        .expect("upstream SSE response");
+
+    let raw = server.await.expect("non-streaming HTTP server task");
+    let separator = raw
+        .windows(4)
+        .position(|window| window == b"\r\n\r\n")
+        .expect("HTTP head/body separator");
+    let compressed_body = &raw[separator + 4..];
+    let decompressed = zstd::stream::decode_all(std::io::Cursor::new(compressed_body))
+        .expect("zstd body should decode");
+    let json: serde_json::Value =
+        serde_json::from_slice(&decompressed).expect("decompressed body should be valid JSON");
+    assert_eq!(json["stream"], true);
+    assert!(
+        !request.stream(),
+        "upstream normalization must not mutate local metadata"
+    );
+}
+
 #[tokio::test]
 async fn websocket_should_keep_an_exact_chain_while_new_connections_adopt_the_latest_codex_core_wire_profile()
  {

backend/crates/providers/openai/tests/transport/profile.rs

@@ -60,13 +60,13 @@ impl CodexCliReleaseTransport for CliReleaseTransport {
 }
 
 #[test]
-fn wire_profile_should_generate_codex_core_user_agent() {
+fn wire_profile_should_generate_bundled_core_app_server_user_agent() {
     let profile = CodexWireProfile {
         originator: "Codex Desktop".to_owned(),
         codex_version: "0.144.2".to_owned(),
         desktop_version: "26.707.72221".to_owned(),
         desktop_build: "72221".to_owned(),
-        os_type: "macOS".to_owned(),
+        os_type: "Mac OS".to_owned(),
         os_version: "15.7.1".to_owned(),
         arch: "arm64".to_owned(),
         terminal: "unknown".to_owned(),
@@ -78,7 +78,7 @@ fn wire_profile_should_generate_codex_core_user_agent() {
 
     assert_eq!(
         profile.user_agent(),
-        "Codex Desktop/0.144.2 (macOS 15.7.1; arm64) unknown"
+        "Codex Desktop/0.144.2 (Mac OS 15.7.1; arm64) unknown (Codex Desktop; 26.707.72221)"
     );
 }
 
@@ -137,6 +137,7 @@ fn desktop_appcast_should_reject_invalid_version_and_build() {
 #[tokio::test]
 async fn wire_profile_release_update_should_change_only_desktop_identity() {
     let original = wire_profile();
+    let original_user_agent = original.user_agent();
     let state = CodexWireProfileState::new(original.clone());
     let service = CodexDesktopReleaseService::new(
         state.clone(),
@@ -155,6 +156,11 @@ async fn wire_profile_release_update_should_change_only_desktop_identity() {
     assert_eq!(updated.arch, original.arch);
     assert_eq!(updated.terminal, original.terminal);
     assert_eq!(updated.verified_at, original.verified_at);
+    assert_ne!(updated.user_agent(), original_user_agent);
+    assert_eq!(
+        updated.user_agent(),
+        "Codex Desktop/0.144.2 (Mac OS 15.7.1; arm64) unknown (Codex Desktop; 26.720.1)"
+    );
     let status = service.status().snapshot();
     assert_eq!(status.latest, Some(release("26.720.1", "72001")));
     assert!(status.checked_at.is_some());
@@ -164,6 +170,7 @@ async fn wire_profile_release_update_should_change_only_desktop_identity() {
 #[tokio::test]
 async fn cli_release_update_should_change_only_codex_core_identity() {
     let original = wire_profile();
+    let original_user_agent = original.user_agent();
     let state = CodexWireProfileState::new(original.clone());
     let service = CodexCliReleaseService::new(
         state.clone(),
@@ -182,6 +189,11 @@ async fn cli_release_update_should_change_only_codex_core_identity() {
     assert_eq!(updated.arch, original.arch);
     assert_eq!(updated.terminal, original.terminal);
     assert_eq!(updated.verified_at, original.verified_at);
+    assert_ne!(updated.user_agent(), original_user_agent);
+    assert_eq!(
+        updated.user_agent(),
+        "Codex Desktop/0.145.0 (Mac OS 15.7.1; arm64) unknown (Codex Desktop; 26.707.72221)"
+    );
 }
 
 #[tokio::test]
@@ -219,7 +231,7 @@ fn wire_profile() -> CodexWireProfile {
         codex_version: "0.144.2".to_owned(),
         desktop_version: "26.707.72221".to_owned(),
         desktop_build: "72221".to_owned(),
-        os_type: "macOS".to_owned(),
+        os_type: "Mac OS".to_owned(),
         os_version: "15.7.1".to_owned(),
         arch: "arm64".to_owned(),
         terminal: "unknown".to_owned(),

backend/crates/providers/openai/tests/transport/request.rs

@@ -401,6 +401,41 @@ fn observability_semantics_should_reuse_codex_turn_metadata() {
     assert!(semantics.compact);
 }
 
+#[test]
+fn encoder_should_extract_subagent_kind_from_wire_or_turn_metadata() {
+    let metadata_request = request(Map::from_iter([
+        ("model".to_owned(), json!("client-model")),
+        ("input".to_owned(), json!("prompt")),
+        (
+            "client_metadata".to_owned(),
+            json!({"x-openai-subagent": "review"}),
+        ),
+    ]));
+    let metadata_encoded =
+        encode_generate_request(&metadata_request, "gpt-test").expect("encode metadata request");
+    assert_eq!(metadata_encoded.subagent_kind().as_deref(), Some("review"));
+
+    let turn_metadata_payload = ProtocolPayload::json_object(
+        "openai",
+        Map::from_iter([
+            ("model".to_owned(), json!("client-model")),
+            ("input".to_owned(), json!("prompt")),
+        ]),
+    )
+    .expect("OpenAI payload")
+    .with_context(Map::from_iter([(
+        "turn_metadata".to_owned(),
+        Value::String(r#"{"subagent_kind":"worker"}"#.to_owned()),
+    )]));
+    let turn_metadata_request = GenerateRequest::from_protocol_payload(turn_metadata_payload);
+    let turn_metadata_encoded = encode_generate_request(&turn_metadata_request, "gpt-test")
+        .expect("encode turn metadata request");
+    assert_eq!(
+        turn_metadata_encoded.subagent_kind().as_deref(),
+        Some("worker")
+    );
+}
+
 #[test]
 fn observability_semantics_should_use_the_transparent_openai_payload() {
     let payload = ProtocolPayload::json_object(

backend/crates/providers/openai/tests/transport/usage.rs

@@ -163,7 +163,12 @@ async fn fetch_should_use_wham_usage_headers_only() {
             "unexpected {name} header"
         );
     }
-    assert!(headers.get("user-agent").is_some());
+    assert_eq!(
+        headers
+            .get("user-agent")
+            .and_then(|value| value.to_str().ok()),
+        Some("codex_cli_rs/0.144.0 (linux 6.8; x86_64) xterm (codex_cli_rs; 1.0.0)")
+    );
     let quota_header_names = headers
         .keys()
         .map(|name| name.as_str())

backend/crates/providers/xai/src/admin.rs

@@ -1048,6 +1048,7 @@ fn project_quota(
             refresh_token_expires_at,
             windows: Vec::new(),
             limit_reached: false,
+            rate_limited_until: None,
             provider_data: None,
         };
     };
@@ -1115,6 +1116,7 @@ fn project_quota(
         limit_reached: billing
             .used_percent()
             .is_some_and(|used| used.is_finite() && used >= 100.0),
+        rate_limited_until: None,
         provider_data: None,
     }
 }

backend/crates/providers/xai/src/credential/client.rs

@@ -221,7 +221,7 @@ impl GrokOAuthClient {
     /// # Errors
     ///
     /// 返回已分类的 OAuth 错误。refresh token 可能已轮换，Ambiguous 类
-    /// transport 失败不得自动重试。
+    /// transport 失败不在本次 exchange 内重试，后续交由 refresh scheduler 退避协调。
     pub async fn refresh(
         &self,
         discovery: &DiscoveryDocument,

backend/crates/providers/xai/src/credential/error.rs

@@ -37,7 +37,7 @@ pub enum OAuthErrorCode {
 pub enum FailureClass {
     /// 瞬时失败，允许后续独立协调的重试。
     Transient,
-    /// 发送状态不确定，一次性凭据不得重放。
+    /// 发送状态不确定，不应在同一 exchange 内重放一次性凭据。
     Ambiguous,
     /// 当前 revision 的凭据被永久拒绝。
     CredentialPermanent,

backend/crates/providers/xai/src/credential/refresh.rs

@@ -12,10 +12,10 @@ use gateway_core::engine::credential::{
     AccountAvailability, CredentialRevision, LoadedCredential, ProviderAccountId,
 };
 use gateway_core::provider_ports::{
-    ProviderCooldown, ProviderCooldownPort, ProviderCredentialStatePort, ProviderLeaseAcquisition,
-    ProviderLeasePort, ProviderLeaseRequest, ProviderRefreshCapacityRequest,
-    ProviderRefreshLeaseRequest, ProviderRefreshPolicy, ProviderRuntimePolicyPort,
-    ProviderStoreError, provider_refresh_backoff_at,
+    ProviderCredentialStatePort, ProviderLeaseAcquisition, ProviderLeasePort, ProviderLeaseRequest,
+    ProviderRefreshCapacityRequest, ProviderRefreshLeaseRequest, ProviderRefreshPolicy,
+    ProviderRuntimePolicyPort, ProviderStoreError, provider_refresh_backoff_at,
+    provider_refresh_retry_at,
 };
 
 use super::catalog::GrokCredentialCatalogService;
@@ -38,12 +38,49 @@ const MAX_SECRET_BYTES: usize = 64 * 1_024;
 const DISCOVERY_NEGATIVE_CACHE_TTL: Duration = Duration::from_secs(5);
 /// 连续失败计数窗口；静默满窗后计数过期归零。
 const REFRESH_BACKOFF_WINDOW: Duration = Duration::from_secs(30 * 60);
-const REFRESH_AMBIGUITY_COOLDOWN: Duration = Duration::from_secs(24 * 60 * 60);
+/// 窗口内先执行五次指数退避，再进入固定恢复周期。
+const REFRESH_BACKOFF_MAX_ATTEMPTS: u32 = 5;
+/// 耗尽指数退避后的 OAuth 恢复周期。
+const REFRESH_RECOVERY_DELAY: Duration = Duration::from_secs(10 * 60);
+/// 过期 AT 仍允许 RT 恢复的最长窗口。
+const REFRESH_RECOVERY_WINDOW: Duration = Duration::from_secs(24 * 60 * 60);
+
+fn refresh_recovery_deadline(access_token_expires_at: Option<SystemTime>) -> Option<SystemTime> {
+    access_token_expires_at.and_then(|expires_at| expires_at.checked_add(REFRESH_RECOVERY_WINDOW))
+}
+
+fn refresh_recovery_window_exhausted(
+    access_token_expires_at: Option<SystemTime>,
+    now: SystemTime,
+) -> bool {
+    refresh_recovery_deadline(access_token_expires_at).is_some_and(|deadline| deadline <= now)
+}
+
+fn bounded_refresh_retry_at(
+    retry_at: SystemTime,
+    access_token_expires_at: Option<SystemTime>,
+    observed_at: SystemTime,
+) -> SystemTime {
+    let mut bounded = retry_at;
+    if let Some(expires_at) = access_token_expires_at
+        && expires_at > observed_at
+        && bounded > expires_at
+    {
+        bounded = expires_at;
+    }
+    if let Some(deadline) = refresh_recovery_deadline(access_token_expires_at)
+        && bounded > deadline
+    {
+        bounded = deadline;
+    }
+    bounded
+}
 
 /// 一个到期且已按 revision 读取明文 RT 的 xAI account。
 pub struct DueGrokCredential {
     account_id: ProviderAccountId,
     credential_revision: CredentialRevision,
+    access_token_expires_at: Option<SystemTime>,
     refresh_token: SecretValue,
     id_token: Option<SecretValue>,
     scope: String,
@@ -77,6 +114,7 @@ impl fmt::Debug for DueGrokCredential {
             .debug_struct("DueGrokCredential")
             .field("account_id", &self.account_id)
             .field("credential_revision", &self.credential_revision)
+            .field("access_token_expires_at", &self.access_token_expires_at)
             .field("refresh_token", &"[REDACTED]")
             .field("id_token", &self.id_token.as_ref().map(|_| "[REDACTED]"))
             .field("scope", &"[REDACTED]")
@@ -277,7 +315,6 @@ pub struct GrokCredentialRefreshService {
     refresher: Arc<dyn GrokCredentialRefresher>,
     catalog: Arc<GrokCredentialCatalogService>,
     leases: Arc<dyn ProviderLeasePort>,
-    cooldowns: Arc<dyn ProviderCooldownPort>,
     credential_state: Arc<dyn ProviderCredentialStatePort>,
     runtime_policy: Arc<dyn ProviderRuntimePolicyPort>,
 }
@@ -288,7 +325,6 @@ impl GrokCredentialRefreshService {
         refresher: Arc<dyn GrokCredentialRefresher>,
         catalog: Arc<GrokCredentialCatalogService>,
         leases: Arc<dyn ProviderLeasePort>,
-        cooldowns: Arc<dyn ProviderCooldownPort>,
         credential_state: Arc<dyn ProviderCredentialStatePort>,
         runtime_policy: Arc<dyn ProviderRuntimePolicyPort>,
     ) -> Self {
@@ -297,7 +333,6 @@ impl GrokCredentialRefreshService {
             refresher,
             catalog,
             leases,
-            cooldowns,
             credential_state,
             runtime_policy,
         }
@@ -331,6 +366,7 @@ impl GrokCredentialRefreshService {
         let credential = DueGrokCredential {
             account_id: account_id.clone(),
             credential_revision,
+            access_token_expires_at: loaded.account.access_token_expires_at(),
             refresh_token: loaded.refresh_token,
             id_token: loaded.id_token,
             scope: loaded.scope,
@@ -447,12 +483,7 @@ impl GrokCredentialRefreshService {
             .into_iter()
             .map(|account_id| GrokCredentialRefreshOutcome::Failed { account_id })
             .collect::<Vec<_>>();
-        let mut credentials = Vec::with_capacity(batch.credentials.len());
-        for credential in batch.credentials {
-            if !self.runtime_cooldown_active(&credential).await {
-                credentials.push(credential);
-            }
-        }
+        let credentials = batch.credentials;
         if credentials.is_empty() {
             return Ok(outcomes);
         }
@@ -474,7 +505,14 @@ impl GrokCredentialRefreshService {
         for (account_id, result) in refreshed {
             match result {
                 Ok(outcome) => outcomes.push(outcome),
-                Err(_) => outcomes.push(GrokCredentialRefreshOutcome::Failed { account_id }),
+                Err(error) => {
+                    tracing::warn!(
+                        account_id = %account_id,
+                        error = %error,
+                        "xAI OAuth refresh attempt failed"
+                    );
+                    outcomes.push(GrokCredentialRefreshOutcome::Failed { account_id });
+                }
             }
         }
         Ok(outcomes)
@@ -486,9 +524,10 @@ impl GrokCredentialRefreshService {
         policy: ProviderRefreshPolicy,
     ) -> Result<GrokCredentialRefreshOutcome, GrokCredentialRefreshError> {
         let account_id = credential.account_id.clone();
+        let observed_at = SystemTime::now();
         if credential
             .refresh_token_expires_at
-            .is_some_and(|expires_at| expires_at <= Utc::now())
+            .is_some_and(|expires_at| expires_at <= DateTime::<Utc>::from(observed_at))
         {
             return self
                 .persist_terminal_failure(
@@ -498,6 +537,15 @@ impl GrokCredentialRefreshService {
                 )
                 .await;
         }
+        if refresh_recovery_window_exhausted(credential.access_token_expires_at, observed_at) {
+            return self
+                .persist_terminal_failure(
+                    credential,
+                    GrokCredentialAvailability::Expired,
+                    "refresh_recovery_window_exhausted",
+                )
+                .await;
+        }
         let capacity = self
             .leases
             .try_acquire(ProviderLeaseRequest::RefreshCapacity(
@@ -545,9 +593,14 @@ impl GrokCredentialRefreshService {
                 .await
             }
             Err(GrokRefreshFailure::Ambiguous) => {
-                self.record_runtime_cooldown(&credential, REFRESH_AMBIGUITY_COOLDOWN)
-                    .await;
-                Ok(GrokCredentialRefreshOutcome::Ambiguous { account_id })
+                if self
+                    .persist_backoff(&credential, "refresh-ambiguous")
+                    .await?
+                {
+                    Ok(GrokCredentialRefreshOutcome::Ambiguous { account_id })
+                } else {
+                    Ok(GrokCredentialRefreshOutcome::Stale { account_id })
+                }
             }
             Err(GrokRefreshFailure::Transient) => {
                 if self
@@ -644,35 +697,19 @@ impl GrokCredentialRefreshService {
             .credential_state
             .clear_refresh_backoff(&account_id)
             .await;
+        tracing::info!(
+            account_id = %account_id,
+            credential_revision = record.credential_revision.get(),
+            access_token_expires_at = %access_expires_at,
+            next_refresh_at = %next_refresh_at,
+            "xAI OAuth refresh succeeded"
+        );
         Ok(GrokCredentialRefreshOutcome::Refreshed {
             account_id,
             credential_revision: record.credential_revision,
         })
     }
 
-    async fn runtime_cooldown_active(&self, credential: &DueGrokCredential) -> bool {
-        // runtime cooldown 是账号/model 级生命周期，与 credential revision 无关：
-        // 轮换不清除、不因 revision 变化判失效（与 selector 的读取语义一致）。
-        let Ok(Some(cooldown)) = self.cooldowns.read(credential.account_id()).await else {
-            return false;
-        };
-        cooldown.until() > SystemTime::now()
-    }
-
-    async fn record_runtime_cooldown(&self, credential: &DueGrokCredential, duration: Duration) {
-        let cooldown_until = SystemTime::now()
-            .checked_add(duration)
-            .unwrap_or_else(SystemTime::now);
-        let _ = self
-            .cooldowns
-            .put_if_later(ProviderCooldown::new(
-                credential.account_id.clone(),
-                credential.credential_revision,
-                cooldown_until,
-            ))
-            .await;
-    }
-
     async fn persist_terminal_failure(
         &self,
         credential: DueGrokCredential,
@@ -702,7 +739,19 @@ impl GrokCredentialRefreshService {
             })
             .await
         {
-            Ok(()) => Ok(GrokCredentialRefreshOutcome::Invalidated { account_id }),
+            Ok(()) => {
+                tracing::warn!(
+                    account_id = %account_id,
+                    ?availability,
+                    reason,
+                    access_token_expires_at = ?credential.access_token_expires_at
+                        .map(DateTime::<Utc>::from),
+                    recovery_deadline = ?refresh_recovery_deadline(credential.access_token_expires_at)
+                        .map(DateTime::<Utc>::from),
+                    "xAI OAuth refresh marked account terminal"
+                );
+                Ok(GrokCredentialRefreshOutcome::Invalidated { account_id })
+            }
             Err(error) if stale_repository_error(&error) => {
                 Ok(GrokCredentialRefreshOutcome::Stale { account_id })
             }
@@ -720,12 +769,19 @@ impl GrokCredentialRefreshService {
             .record_refresh_backoff(&credential.account_id, REFRESH_BACKOFF_WINDOW)
             .await
             .unwrap_or(1);
-        let retry_at = provider_refresh_backoff_at(
-            &credential.account_id,
-            SystemTime::now(),
-            attempt,
-            reason,
-        )?;
+        let observed_at = SystemTime::now();
+        let retry_at = if attempt <= REFRESH_BACKOFF_MAX_ATTEMPTS {
+            provider_refresh_backoff_at(&credential.account_id, observed_at, attempt, reason)?
+        } else {
+            provider_refresh_retry_at(
+                &credential.account_id,
+                observed_at,
+                REFRESH_RECOVERY_DELAY,
+                reason,
+            )?
+        };
+        let retry_at =
+            bounded_refresh_retry_at(retry_at, credential.access_token_expires_at, observed_at);
         match self
             .repository
             .defer_refresh(
@@ -735,7 +791,20 @@ impl GrokCredentialRefreshService {
             )
             .await
         {
-            Ok(()) => Ok(true),
+            Ok(()) => {
+                tracing::warn!(
+                    account_id = %credential.account_id,
+                    attempt,
+                    reason,
+                    retry_at = %DateTime::<Utc>::from(retry_at),
+                    access_token_expires_at = ?credential.access_token_expires_at
+                        .map(DateTime::<Utc>::from),
+                    recovery_deadline = ?refresh_recovery_deadline(credential.access_token_expires_at)
+                        .map(DateTime::<Utc>::from),
+                    "xAI OAuth refresh deferred"
+                );
+                Ok(true)
+            }
             Err(error) if stale_repository_error(&error) => Ok(false),
             Err(error) => Err(error.into()),
         }
@@ -814,6 +883,7 @@ impl GrokCredentialRepository {
             due.push(DueGrokCredential {
                 account_id: account.id().clone(),
                 credential_revision: account.revision(),
+                access_token_expires_at: account.access_token_expires_at(),
                 refresh_token: loaded.refresh_token,
                 id_token: loaded.id_token,
                 scope: loaded.scope,
@@ -844,7 +914,8 @@ fn account_due(
                 | AccountAvailability::Banned
                 | AccountAvailability::Invalid
         )
-        && account.next_refresh_at().is_some_and(|next| next <= now)
+        && (refresh_recovery_window_exhausted(account.access_token_expires_at(), now)
+            || account.next_refresh_at().is_some_and(|next| next <= now))
 }
 
 fn refreshed_tokens(tokens: RefreshedTokenSet) -> Result<GrokRefreshTokens, GrokRefreshFailure> {

backend/crates/providers/xai/src/credential/repository.rs

@@ -6,10 +6,10 @@ use std::time::SystemTime;
 
 use chrono::{DateTime, Utc};
 use gateway_core::engine::credential::{
-    AccountStateChange, CredentialCasOutcome, CredentialCasUpdate, CredentialRevision,
-    LoadedCredential, NewProviderAccount, OpaqueProviderData, PlaintextCredential, ProviderAccount,
-    ProviderAccountId, ProviderAccountStore, ProviderAccountUpdate, QuotaObservation,
-    QuotaWriteOutcome,
+    AccountAvailability, AccountStateChange, CredentialCasOutcome, CredentialCasUpdate,
+    CredentialRevision, LoadedCredential, NewProviderAccount, OpaqueProviderData,
+    PlaintextCredential, ProviderAccount, ProviderAccountId, ProviderAccountStore,
+    ProviderAccountUpdate, QuotaObservation, QuotaWriteOutcome,
 };
 use gateway_core::error::StoreErrorKind;
 use gateway_core::provider_ports::ProviderRefreshPolicy;
@@ -428,14 +428,21 @@ impl GrokCredentialRepository {
         &self,
         input: &UpdateGrokCredentialState,
     ) -> Result<(), GrokCredentialRepositoryError> {
-        validate_reason(input.availability_reason.as_deref())?;
+        let availability = AccountAvailability::from(input.availability);
+        let availability_reason = input
+            .availability_reason
+            .clone()
+            .filter(|value| !value.trim().is_empty())
+            .or_else(|| availability.fallback_error_reason().map(str::to_owned));
+        validate_reason(availability_reason.as_deref())?;
         self.ensure_xai_account(&input.account_id).await?;
         self.store
             .apply_state_change(AccountStateChange {
                 account_id: input.account_id.clone(),
                 expected_revision: input.expected_revision,
-                availability: input.availability.into(),
+                availability,
                 observed_at: to_system_time(input.observed_at),
+                message: availability_reason,
             })
             .await
             .map_err(map_store_error)

backend/crates/providers/xai/src/lib.rs

@@ -171,7 +171,6 @@ pub async fn initialize(
         refresher,
         Arc::clone(&catalog),
         Arc::clone(&leases),
-        cooldowns.clone(),
         ports.credential_state(),
         Arc::clone(&runtime_policy),
     ));

backend/crates/providers/xai/src/provider.rs

@@ -2040,11 +2040,10 @@ impl ScheduledTask for XaiOAuthRefreshTask {
             if context.cancellation().is_cancelled() {
                 return Ok(());
             }
-            let outcomes = self
-                .service
-                .refresh_due()
-                .await
-                .map_err(|_| WorkerTaskError::safe("xAI OAuth refresh failed"))?;
+            let outcomes = self.service.refresh_due().await.map_err(|error| {
+                tracing::error!(error = %error, "xAI OAuth refresh cycle failed");
+                WorkerTaskError::safe("xAI OAuth refresh failed")
+            })?;
             let failures = outcomes
                 .iter()
                 .filter(|outcome| {
@@ -2056,6 +2055,13 @@ impl ScheduledTask for XaiOAuthRefreshTask {
                     )
                 })
                 .count();
+            if !outcomes.is_empty() {
+                tracing::info!(
+                    accounts = outcomes.len(),
+                    failures,
+                    "xAI OAuth refresh cycle completed"
+                );
+            }
             if failures > 0 {
                 tracing::warn!(failures, "xAI OAuth refresh cycle contained failures");
             }

backend/crates/providers/xai/tests/admin.rs

@@ -473,6 +473,7 @@ fn account_record(account: &ProviderAccount) -> AccountRecord {
         enabled: account.enabled(),
         availability: account.availability(),
         availability_observed_at: now,
+        last_error_message: None,
         quota_observed_at: None,
         created_at: now,
         updated_at: now,

backend/crates/providers/xai/tests/credential/catalog.rs

@@ -204,6 +204,7 @@ async fn set_account_state(
 ) {
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: id.clone(),
             expected_revision: CredentialRevision::new(1).expect("revision"),
             availability,
@@ -485,6 +486,7 @@ async fn quota_exhausted_account_remains_eligible_for_catalog_discovery() {
     let id = account_id("quota-exhausted");
     store
         .apply_state_change(AccountStateChange {
+            message: None,
             account_id: id,
             expected_revision: CredentialRevision::new(1).expect("revision"),
             availability: AccountAvailability::QuotaExhausted,
@@ -755,6 +757,7 @@ async fn quota_refresh_preserves_ready_availability_across_concurrent_state_writ
     let transport = Arc::new(MutatingBillingTransport {
         store: Arc::clone(&store),
         mutation: Mutex::new(Some(BillingMutation::State(AccountStateChange {
+            message: None,
             account_id: id.clone(),
             expected_revision: CredentialRevision::new(1).expect("revision"),
             availability: AccountAvailability::Ready,

backend/crates/providers/xai/tests/credential/contract.rs

@@ -233,6 +233,29 @@ async fn state_update_uses_credential_revision_fence() {
     assert_eq!(account.availability(), AccountAvailability::Ready);
 }
 
+#[tokio::test]
+async fn terminal_state_without_upstream_message_uses_stable_error_reason() {
+    let (store, repository) = repository();
+    let input = create_input("state-fallback", "subject-state-fallback");
+    seed_input(&store, &input).await.expect("create");
+
+    repository
+        .update_state(&UpdateGrokCredentialState {
+            account_id: input.account_id.clone(),
+            expected_revision: CredentialRevision::new(1).expect("revision"),
+            availability: GrokCredentialAvailability::Banned,
+            availability_reason: None,
+            observed_at: Utc::now(),
+        })
+        .await
+        .expect("state update");
+
+    assert_eq!(
+        store.last_error_message(&input.account_id).as_deref(),
+        Some("account_banned")
+    );
+}
+
 #[tokio::test]
 async fn admin_prepare_does_not_mutate_provider_account_store() {
     let (store, _) = repository();

backend/crates/providers/xai/tests/credential/refresh.rs

@@ -10,8 +10,8 @@ use gateway_core::engine::credential::{
     ProviderAccountId, ProviderAccountStore, ProviderAccountUpdate,
 };
 use gateway_core::provider_ports::{
-    ProviderCooldown, ProviderCooldownPort, ProviderCredentialState, ProviderCredentialStatePort,
-    ProviderLeaseAcquisition, ProviderLeasePort, ProviderLeaseRequest, ProviderStoreError,
+    ProviderCredentialState, ProviderCredentialStatePort, ProviderLeaseAcquisition,
+    ProviderLeasePort, ProviderLeaseRequest, ProviderStoreError,
 };
 use provider_xai::{
     GrokCredentialCatalogCache, GrokCredentialRecovery, GrokCredentialRecoveryOutcome,
@@ -22,12 +22,13 @@ use provider_xai::{
 };
 
 use crate::support::{
-    MemoryCooldownPort, MemoryGrokCatalogCache, MemoryProviderAccountStore, create_input,
-    credential_object, runtime_policy, seed_input,
+    MemoryGrokCatalogCache, MemoryProviderAccountStore, create_input, credential_object,
+    runtime_policy, seed_input,
 };
 
 const OFFICIAL_FIXTURE: &[u8] =
     include_bytes!("../transport/catalog/fixtures/official_grok_models_snapshot.json");
+const OAUTH_BACKOFF_ATTEMPTS: u32 = 5;
 
 struct StaticCatalogTransport;
 
@@ -143,48 +144,6 @@ async fn fixture(
     fixture_many([input], responses, lease_available).await
 }
 
-/// 带外部注入 cooldown 端口的 fixture（AUD-05 组合测试用）。
-#[allow(clippy::type_complexity)]
-async fn fixture_with_cooldowns(
-    input: provider_xai::CreateGrokCredential,
-    responses: impl IntoIterator<Item = Result<GrokRefreshTokens, GrokRefreshFailure>>,
-    lease_available: bool,
-    cooldowns: Arc<MemoryCooldownPort>,
-) -> (
-    Arc<MemoryProviderAccountStore>,
-    GrokCredentialRepository,
-    Arc<QueueRefresher>,
-    GrokCredentialRefreshService,
-    Arc<MemoryCooldownPort>,
-) {
-    let store = MemoryProviderAccountStore::shared();
-    let account_store: Arc<dyn ProviderAccountStore> = store.clone();
-    let repository = GrokCredentialRepository::new(account_store);
-    seed_input(&store, &input).await.expect("create account");
-    let refresher = QueueRefresher::new(responses);
-    let refresher_port: Arc<dyn GrokCredentialRefresher> = refresher.clone();
-    let cache: Arc<dyn GrokCredentialCatalogCache> = MemoryGrokCatalogCache::shared();
-    let catalog = Arc::new(crate::support::grok_catalog_service(
-        repository.clone(),
-        Arc::new(StaticCatalogTransport),
-        cache,
-    ));
-    let leases = Arc::new(TestRefreshLeases {
-        available: lease_available,
-        calls: AtomicUsize::new(0),
-    });
-    let service = GrokCredentialRefreshService::new(
-        repository.clone(),
-        refresher_port,
-        catalog,
-        leases,
-        cooldowns.clone(),
-        Arc::new(CountingCredentialState::default()),
-        runtime_policy(),
-    );
-    (store, repository, refresher, service, cooldowns)
-}
-
 /// 真实累加连续失败计数的测试 double，用于断言退避的指数增长与成功清零。
 #[derive(Default)]
 struct CountingCredentialState {
@@ -313,13 +272,11 @@ async fn fixture_many_with_state(
         available: lease_available,
         calls: AtomicUsize::new(0),
     });
-    let cooldowns = Arc::new(MemoryCooldownPort::default());
     let service = GrokCredentialRefreshService::new(
         repository.clone(),
         refresher_port,
         catalog,
         leases,
-        cooldowns,
         credential_state,
         runtime_policy(),
     );
@@ -354,6 +311,38 @@ async fn force_due(store: &MemoryProviderAccountStore, id: &ProviderAccountId) {
     ));
 }
 
+async fn set_token_deadlines(
+    store: &MemoryProviderAccountStore,
+    id: &ProviderAccountId,
+    access_token_expires_at: SystemTime,
+    next_refresh_at: SystemTime,
+) {
+    let account = store.account(id).expect("account to update");
+    let credential = store.credential(id).expect("credential to update");
+    let update = CredentialCasUpdate::new(
+        id.clone(),
+        account.revision(),
+        ProviderAccountUpdate {
+            account_id: id.clone(),
+            name: account.name().to_owned(),
+            email: account.email().map(str::to_owned),
+            plan_type: account.plan_type().map(str::to_owned),
+        },
+        credential,
+        true,
+        Some(access_token_expires_at),
+        Some(next_refresh_at),
+    )
+    .expect("valid token deadline update");
+    assert!(matches!(
+        store
+            .compare_and_swap_credential(update)
+            .await
+            .expect("update token deadlines"),
+        CredentialCasOutcome::Updated(_)
+    ));
+}
+
 #[tokio::test]
 async fn successful_refresh_rotates_plaintext_tokens_once() {
     let input = due_input("success");
@@ -415,6 +404,10 @@ async fn unauthorized_recovery_marks_a_permanently_rejected_refresh_expired() {
         store.account(&id).expect("account").availability(),
         AccountAvailability::Expired
     );
+    assert_eq!(
+        store.last_error_message(&id).as_deref(),
+        Some("refresh_invalid_grant")
+    );
 }
 
 #[tokio::test]
@@ -573,14 +566,25 @@ async fn banned_failure_marks_account_banned() {
         store.account(&id).expect("account").availability(),
         AccountAvailability::Banned
     );
+    assert_eq!(
+        store.last_error_message(&id).as_deref(),
+        Some("account_banned")
+    );
 }
 
 #[tokio::test]
-async fn ambiguous_refresh_uses_runtime_cooldown_without_persisting_account_state() {
+async fn ambiguous_refresh_applies_bounded_oauth_backoff() {
     let input = due_input("ambiguous");
     let id = input.account_id.clone();
-    let (store, _, refresher, service) =
-        fixture(input, [Err(GrokRefreshFailure::Ambiguous)], true).await;
+    let counting = Arc::new(CountingCredentialState::default());
+    let (store, _, refresher, service) = fixture_many_with_state(
+        [input],
+        [Err(GrokRefreshFailure::Ambiguous)],
+        true,
+        counting.clone(),
+    )
+    .await;
+    let started_at = SystemTime::now();
     let outcomes = service.refresh_due().await.expect("refresh");
     assert!(matches!(
         outcomes.as_slice(),
@@ -590,7 +594,18 @@ async fn ambiguous_refresh_uses_runtime_cooldown_without_persisting_account_stat
         store.account(&id).expect("account").availability(),
         AccountAvailability::Unknown
     );
-    assert_eq!(store.account(&id).expect("account").revision().get(), 1);
+    let account = store.account(&id).expect("account");
+    assert_eq!(account.revision().get(), 2);
+    let retry_delay = account
+        .next_refresh_at()
+        .expect("OAuth retry scheduled")
+        .duration_since(started_at)
+        .expect("OAuth retry is in the future");
+    assert!(
+        retry_delay <= Duration::from_secs(7),
+        "first retry must use bounded OAuth backoff, got {retry_delay:?}"
+    );
+    assert_eq!(counting.count(&id), 1);
     assert!(
         service
             .refresh_due()
@@ -602,7 +617,7 @@ async fn ambiguous_refresh_uses_runtime_cooldown_without_persisting_account_stat
 }
 
 #[tokio::test]
-async fn pre_send_transient_failure_applies_bounded_cooldown() {
+async fn pre_send_transient_failure_applies_bounded_oauth_backoff() {
     let input = due_input("transient");
     let id = input.account_id.clone();
     let (store, _, _, service) = fixture(input, [Err(GrokRefreshFailure::Transient)], true).await;
@@ -672,6 +687,121 @@ async fn refresh_backoff_grows_exponentially_across_attempts() {
     );
 }
 
+#[tokio::test]
+async fn exhausted_oauth_backoff_continues_after_access_token_expiry() {
+    let input = due_input("backoff-recovery");
+    let id = input.account_id.clone();
+    let counting = Arc::new(CountingCredentialState::default());
+    let (store, _, refresher, service) = fixture_many_with_state(
+        [input],
+        [
+            Err(GrokRefreshFailure::Transient),
+            Err(GrokRefreshFailure::Transient),
+            Err(GrokRefreshFailure::Transient),
+            Err(GrokRefreshFailure::Transient),
+            Err(GrokRefreshFailure::Transient),
+            Err(GrokRefreshFailure::Transient),
+        ],
+        true,
+        counting.clone(),
+    )
+    .await;
+    set_token_deadlines(
+        &store,
+        &id,
+        (Utc::now() - chrono::Duration::minutes(1)).into(),
+        SystemTime::now() - Duration::from_secs(1),
+    )
+    .await;
+
+    for _ in 0..OAUTH_BACKOFF_ATTEMPTS {
+        service.refresh_due().await.expect("backoff attempt");
+        force_due(&store, &id).await;
+    }
+
+    let observed_at = SystemTime::now();
+    service.refresh_due().await.expect("recovery attempt");
+    let recovery_delay = store
+        .account(&id)
+        .expect("account after recovery defer")
+        .next_refresh_at()
+        .expect("recovery scheduled")
+        .duration_since(observed_at)
+        .expect("recovery is in the future");
+    assert!(
+        (Duration::from_secs(8 * 60)..=Duration::from_secs(12 * 60 + 1)).contains(&recovery_delay),
+        "recovery delay should be about ten minutes, got {recovery_delay:?}"
+    );
+    assert_eq!(counting.count(&id), OAUTH_BACKOFF_ATTEMPTS + 1);
+    assert_eq!(refresher.prepare_calls.load(Ordering::SeqCst), 6);
+}
+
+#[tokio::test]
+async fn oauth_backoff_never_passes_an_unexpired_access_token_expiry() {
+    let mut input = due_input("backoff-expiry-cap");
+    input.account.access_token_expires_at = Utc::now() + chrono::Duration::seconds(30);
+    let access_expires_at: SystemTime = input.account.access_token_expires_at.into();
+    let id = input.account_id.clone();
+    let counting = Arc::new(CountingCredentialState::default());
+    let (store, _, _, service) = fixture_many_with_state(
+        [input],
+        [
+            Err(GrokRefreshFailure::Transient),
+            Err(GrokRefreshFailure::Transient),
+            Err(GrokRefreshFailure::Transient),
+            Err(GrokRefreshFailure::Transient),
+            Err(GrokRefreshFailure::Transient),
+            Err(GrokRefreshFailure::Transient),
+        ],
+        true,
+        counting.clone(),
+    )
+    .await;
+
+    for attempt in 0..=OAUTH_BACKOFF_ATTEMPTS {
+        service.refresh_due().await.expect("backoff attempt");
+        if attempt < OAUTH_BACKOFF_ATTEMPTS {
+            force_due(&store, &id).await;
+        }
+    }
+
+    assert_eq!(
+        store
+            .account(&id)
+            .expect("account after capped retry")
+            .next_refresh_at(),
+        Some(access_expires_at)
+    );
+    assert_eq!(counting.count(&id), OAUTH_BACKOFF_ATTEMPTS + 1);
+}
+
+#[tokio::test]
+async fn recovery_window_exhaustion_marks_account_expired_without_token_exchange() {
+    let input = due_input("recovery-window-exhausted");
+    let id = input.account_id.clone();
+    let (store, _, refresher, service) = fixture(input, [Ok(success_tokens(None))], true).await;
+    set_token_deadlines(
+        &store,
+        &id,
+        (Utc::now() - chrono::Duration::hours(25)).into(),
+        (Utc::now() + chrono::Duration::hours(12)).into(),
+    )
+    .await;
+
+    let outcomes = service.refresh_due().await.expect("refresh due");
+
+    assert!(matches!(
+        outcomes.as_slice(),
+        [GrokCredentialRefreshOutcome::Invalidated { account_id }] if account_id == &id
+    ));
+    assert_eq!(
+        store.account(&id).expect("expired account").availability(),
+        AccountAvailability::Expired
+    );
+    assert_eq!(refresher.prepare_calls.load(Ordering::SeqCst), 1);
+    assert_eq!(refresher.responses.lock().expect("queue").len(), 1);
+}
+
 #[tokio::test]
 async fn successful_refresh_clears_backoff_counter() {
     let input = due_input("backoff-clear");
@@ -776,52 +906,33 @@ async fn malformed_account_refresh_does_not_stop_later_accounts() {
 }
 
 #[tokio::test]
-async fn runtime_cooldown_survives_credential_rotation_and_blocks_refresh() {
-    // cooldown 写于 revision N，轮换到 N+1 后 refresh worker 不得清除
-    // 或失效它；cooldown 活跃期间账号被跳过刷新。
-    let input = create_input(
-        "cooldown-survives-rotation",
-        "subject-cooldown-survives-rotation",
-    );
+async fn scheduled_oauth_refresh_runs_after_credential_rotation() {
+    let input = due_input("refresh-after-rotation");
     let id = input.account_id.clone();
-    let cooldowns = Arc::new(MemoryCooldownPort::default());
-    let (store, _, _, service, cooldowns) = fixture_with_cooldowns(
+    let (store, _, refresher, service) = fixture(
         input,
-        [Ok(success_tokens(Some("rotated-refresh")))],
+        [
+            Ok(success_tokens(Some("recovered-refresh"))),
+            Ok(success_tokens(Some("scheduled-refresh"))),
+        ],
         true,
-        cooldowns,
     )
     .await;
-    // 先写 account cooldown（revision 1，未来到期）。
-    let until = SystemTime::now() + Duration::from_secs(3600);
-    cooldowns
-        .put_if_later(ProviderCooldown::new(
-            id.clone(),
-            CredentialRevision::new(1).expect("revision"),
-            until,
-        ))
-        .await
-        .expect("write cooldown");
 
-    // 轮换到 revision 2。
     let outcome = service
         .recover_unauthorized(&id, CredentialRevision::new(1).expect("revision"))
         .await;
     assert_eq!(outcome, GrokCredentialRecoveryOutcome::Recovered);
     assert_eq!(store.account(&id).expect("account").revision().get(), 2);
 
-    // cooldown 仍有效（轮换不清除）。
-    assert!(
-        cooldowns
-            .cooldown(&id)
-            .is_some_and(|cooldown| cooldown.until() > SystemTime::now()),
-        "rotation must not clear account runtime cooldown"
-    );
-    // cooldown 活跃期间 refresh worker 跳过该账号（不清除、不刷新）。
+    force_due(&store, &id).await;
     let outcomes = service.refresh_due().await.expect("refresh due");
-    assert!(
-        outcomes.is_empty(),
-        "cooldown-active account must be skipped by refresh worker"
-    );
-    assert_eq!(store.account(&id).expect("account").revision().get(), 2);
+    assert!(matches!(
+        outcomes.as_slice(),
+        [GrokCredentialRefreshOutcome::Refreshed {
+            account_id,
+            credential_revision,
+        }] if account_id == &id && credential_revision.get() == 4
+    ));
+    assert_eq!(refresher.prepare_calls.load(Ordering::SeqCst), 2);
 }

backend/crates/providers/xai/tests/support.rs

@@ -194,6 +194,7 @@ struct StoredAccount {
     account: ProviderAccount,
     credential: PlaintextCredential,
     quota: Option<QuotaObservation>,
+    last_error_message: Option<String>,
 }
 
 #[derive(Default)]
@@ -214,6 +215,12 @@ impl MemoryProviderAccountStore {
             .map(|stored| stored.account.clone())
     }
 
+    pub fn last_error_message(&self, id: &ProviderAccountId) -> Option<String> {
+        lock(&self.accounts)
+            .get(id)
+            .and_then(|stored| stored.last_error_message.clone())
+    }
+
     pub fn credential(&self, id: &ProviderAccountId) -> Option<PlaintextCredential> {
         lock(&self.accounts)
             .get(id)
@@ -246,6 +253,7 @@ impl ProviderAccountStore for MemoryProviderAccountStore {
                 account: account.account,
                 credential: account.credential,
                 quota: None,
+                last_error_message: None,
             },
         );
         Ok(())
@@ -388,6 +396,11 @@ impl ProviderAccountStore for MemoryProviderAccountStore {
                 plan_type: stored.account.plan_type().map(str::to_owned),
             },
         );
+        if change.availability == AccountAvailability::Ready {
+            stored.last_error_message = None;
+        } else if let Some(message) = change.message {
+            stored.last_error_message = Some(message);
+        }
         Ok(())
     }
 

backend/migrations/.frozen-sha256

@@ -1,4 +1 @@
-54be07512951a561391b6e611355a1bfc119d82363253fb9bbdb225c9e3e39bb  0001_initial.sql
-f480ac4cb70f6963eee71c0b8a67777ebb492f85501c90ebe7544bee0b2c0525  0002_remove_account_cooldown.sql
-5cf5b5387c66b4d437d79c002cec46cb85196789060fa87c39e97a382487cac9  0004_fix_quota_limit_reached_null.sql
-a3f236d558e7598f69277520458c801ee43130f7704ca25455df468cf8899ff2  0003_materialize_quota_limit_reached.sql
+5f79e74c9a7a7366c39cd1afe43175eff6c44afe9217c0f26f33f173bd40c781  0001_initial.sql

backend/migrations/0001_initial.sql

@@ -131,9 +131,9 @@ create table provider_accounts (
   next_refresh_at timestamptz,
   enabled boolean not null default true,
   availability text not null default 'unknown',
-  availability_reason text,
-  cooldown_until timestamptz,
   provider_quota_json jsonb,
+  quota_limit_reached boolean not null default false,
+  last_error_message text,
   availability_observed_at timestamptz not null,
   quota_observed_at timestamptz,
   created_at timestamptz not null,
@@ -163,23 +163,12 @@ create table provider_accounts (
     availability in (
       'unknown',
       'ready',
-      'cooldown',
       'quota_exhausted',
       'expired',
       'banned',
       'invalid'
     )
   ),
-  constraint provider_accounts_cooldown_ck check (
-    (
-      availability != 'cooldown'
-      or cooldown_until is not null
-    )
-    and (
-      cooldown_until is null
-      or availability in ('cooldown', 'quota_exhausted')
-    )
-  ),
   constraint provider_accounts_time_ck check (
     created_at <= updated_at
     and availability_observed_at <= updated_at
@@ -196,7 +185,7 @@ create unique index provider_accounts_upstream_identity_uq
 create index provider_accounts_runtime_idx
   on provider_accounts (provider_kind, enabled, id);
 create index provider_accounts_availability_idx
-  on provider_accounts (availability, cooldown_until, id);
+  on provider_accounts (availability, id);
 create index provider_accounts_access_expiry_idx
   on provider_accounts (access_token_expires_at, id);
 create index provider_accounts_refresh_due_idx
@@ -235,8 +224,6 @@ create table model_requests (
   outcome text not null default 'running',
   client_status_code integer,
   upstream_status_code integer,
-  -- Provider response IDs are opaque JSON strings. Bytes preserve embedded NUL
-  -- and avoid imposing PostgreSQL text-index limits on protocol handles.
   client_response_id bytea,
   upstream_request_id text,
   upstream_response_id bytea,
@@ -438,6 +425,7 @@ create index model_requests_running_deadline_idx
 create index model_requests_retention_idx
   on model_requests (completed_at)
   where outcome <> 'running';
+
 -- 运行期警告与错误事件。
 create table ops_events (
   id text primary key,
@@ -691,11 +679,9 @@ create unique index backup_records_active_uq
 create unique index backup_records_object_key_uq
   on backup_records (object_key)
   where object_key is not null;
--- 同时服务 retentionDays 与 retentionCount。
 create index backup_records_scheduled_completed_idx
   on backup_records (completed_at desc, id desc)
   where trigger_kind = 'scheduled' and status = 'completed';
--- 可清理记录（expires_at 已到期且非活跃）的扫描索引。
 create index backup_records_retention_idx
   on backup_records (expires_at)
   where expires_at is not null and status in ('completed', 'failed');

backend/migrations/0002_remove_account_cooldown.sql

@@ -1,43 +0,0 @@
--- 移除账号级 cooldown 状态模型（照 v2 语义：429 限流不进 availability，由 quota 数据驱动）。
---
--- 存量迁移：
---   - availability='cooldown' 的账号：429/Cloudflare 等临时冷却 → 'ready'
---     （限流事实留在 provider_quota_json，由投影滚动兜底）；
---     402 语义的 usage_limit_exhausted 且未到期 → 'quota_exhausted'。
-
--- 1. 迁移存量 cooldown 账号。
-update provider_accounts
-set availability = case
-      when availability_reason = 'usage_limit_exhausted'
-           and cooldown_until is not null
-           and cooldown_until > now()
-        then 'quota_exhausted'
-      else 'ready'
-    end
-where availability = 'cooldown';
-
--- 2. 删除依赖 cooldown_until 的约束与索引。
-alter table provider_accounts drop constraint provider_accounts_cooldown_ck;
-drop index provider_accounts_availability_idx;
-
--- 3. availability CHECK 去掉 'cooldown'。
-alter table provider_accounts drop constraint provider_accounts_availability_ck;
-alter table provider_accounts
-  add constraint provider_accounts_availability_ck check (
-    availability in (
-      'unknown',
-      'ready',
-      'quota_exhausted',
-      'expired',
-      'banned',
-      'invalid'
-    )
-  );
-
--- 4. 删除列。
-alter table provider_accounts drop column availability_reason;
-alter table provider_accounts drop column cooldown_until;
-
--- 5. 重建可用性索引（不含 cooldown_until）。
-create index provider_accounts_availability_idx
-  on provider_accounts (availability, id);

backend/migrations/0003_materialize_quota_limit_reached.sql

@@ -1,20 +0,0 @@
--- 物化快照级限流事实：Dashboard 复用 Provider 解析结果，不再 SQL 解析 raw quota JSON。
---
--- `quota_limit_reached` 由 Provider 在写 quota 时计算（快照级：顶层或任一窗口触顶，
--- 已过 reset 的窗口不计数），Admin/Dashboard 直接读该列。
-
-alter table provider_accounts
-  add column quota_limit_reached boolean not null default false;
-
--- 存量回填：从现有 raw JSON 推导（仅 primary_window 的 limit_reached + reset 未过期）。
-update provider_accounts
-set quota_limit_reached = coalesce(
-  provider_quota_json is not null
-  and provider_quota_json->'rate_limit'->>'limit_reached' = 'true'
-  and (
-    provider_quota_json->'rate_limit'->'primary_window'->>'reset_at' is null
-    or (provider_quota_json->'rate_limit'->'primary_window'->>'reset_at')::bigint > extract(epoch from now())::bigint
-  ),
-  false
-)
-where provider_quota_json is not null;

backend/migrations/0004_fix_quota_limit_reached_null.sql

@@ -1,8 +0,0 @@
--- 修正 0003 存量回填：raw JSON 缺 rate_limit 键时表达式为 NULL，
--- 赋 NULL 违反 not-null 约束。0004 把仍为 NULL 的行兜底为 false
--- （0003 已应用的环境，add column 用 default false 填充，通常无 NULL；
--- 本迁移保证任何环境下该列都非 NULL）。
-
-update provider_accounts
-set quota_limit_reached = coalesce(quota_limit_reached, false)
-where quota_limit_reached is null;

deploy/README.md

@@ -111,9 +111,10 @@ PostgreSQL 是账号、Client Key、运行设置、请求记录与审计的权
 Provider schema 以明文 JSON 保存于 PostgreSQL。Redis 只保存可重建、可过期的协调状态，例如
 会话亲和、lease、cooldown、OAuth pending flow 与套餐模型目录 cache。
 
-备份必须包含 `.runtime/postgres`。若希望保留短期 Redis 状态、OpenAI 会话亲和和更新状态，
-同时备份 `.runtime/redis` 与 `.runtime/data`；后两者丢失不会造成账号 credential 丢失，但会使
-缓存和会话锚点重新建立。
+备份必须包含 `.runtime/postgres`。要保留 OpenAI OAuth 的 AT/RT 恢复能力，必须保持
+`host.logging.file.enabled: true` 并备份 `.runtime/logs`；该记录受普通日志的 `retention_days` 和
+`max_files` 约束。若希望保留短期 Redis 状态和会话锚点，也同时备份 `.runtime/redis` 与
+`.runtime/data`。
 
 完整运行时、Provider、revision 与恢复边界见 [架构文档](../docs/architecture.md)。
 
@@ -202,6 +203,11 @@ Release 必须提供当前 OS/架构的 `codex-proxy-rs_<version>_<os>_<arch>.ta
   Debian 官方源补 `libpq5` 依赖，不引入第三方 APT 源，也不依赖运行时动态安装。
 - 备份暂存目录为 `/app/.runtime/data/backup-staging`，由 `.runtime/data` 卷持久化，权限
   `0700`（仅 `cpr` 用户可读写）。部署卷至少预留一个最大数据库归档的空间。
+- OpenAI OAuth 每次成功取得 AT/RT 后，都会在账号资料补全、过期时间计算和数据库写入前，以
+  `openai_oauth_recovery` 结构化日志事件写入普通应用日志。事件含原始 AT/RT，完全沿用
+  `.runtime/logs` 的按日、按大小分割和保留清理策略；文件日志开启时不会被普通日志级别筛掉。
+  它遵循普通日志现有的非阻塞写入机制，不会阻断导入、授权或刷新。
+  `.runtime/logs` 因此属于敏感数据，必须按现有普通日志的访问控制和加密备份策略处理。
 - S3/R2 存储、Cron 计划、保留策略与备份记录都保存在 PostgreSQL（`backup_settings` /
   `backup_records`），备份记录行在删除成功后硬删除，操作历史进入 `admin_audit_events`。
 

deploy/config.example.yaml

@@ -57,16 +57,17 @@ api:
 openai:
   # 上游地址、WebSocket 池、额度刷新和 OAuth 使用 Provider 内置默认值；模板只保留画像启动基线。
   wire_profile:
-    # Provider 每 24 小时读取官方 npm metadata 与 Desktop appcast，仅在进程内更新版本字段。
+    # codex_version 是 npm CLI 的启动基线，并用于模型目录与请求 UA；Provider 每 24 小时更新它。
+    # Desktop appcast 同期只更新 Desktop 版本与构建号。
     originator: 'Codex Desktop'
     codex_version: '0.146.0'
     desktop_version: '26.727.51351'
     desktop_build: '6119'
-    os_type: 'macOS'
+    os_type: 'Mac OS'
     os_version: '15.7.1'
     arch: 'arm64'
     terminal: 'unknown'
-    verified_at: '2026-07-25T00:00:00+08:00'
+    verified_at: '2026-08-04T00:00:00+08:00'
 
 xai:
   # OAuth 端点、额度与模型目录策略由 xAI Provider 固定管理；这里只声明请求画像启动基线。

docs/api.md

@@ -122,7 +122,7 @@ OpenAI 路径保留客户端 Responses wire 语义：请求 body 的未知字段
 - xAI 从单账号 object 或 `accounts` 数组中提取 OAuth token；包装中的代理、并发、优先级等字段不参与认证；
 - xAI 批量导入逐条独立校验：失败条目跳过并记录日志，不中断其余条目，仅当没有任何条目成功时整个导入才报错；
 - xAI API Key 不是受支持的账号 credential；
-- 导入仍会通过 Provider 的身份或刷新链路校验，不能只凭文件外形写入账号。
+- 导入不会只凭文件外形写入账号；目标 Provider 使用认证材料完成必要的 token exchange 或已认证账号资料补全。
 
 OpenAI 的 CPR 导出保持 OAuth 账号的既有字段；Agent Identity 账号则输出
 `authMode: "agentIdentity"`、`agentRuntimeId`、`agentPrivateKey` 和可选的
@@ -171,20 +171,22 @@ OAuth start 使用：
 
 ### OpenAI 身份、额度与状态
 
-- 导入会校验签名 access token 和稳定账号身份；`/wham/usage` 用于补全账号/套餐事实和额度，
-  不是判断 token 是否有效的唯一依据。
+- OAuth 文件导入使用可用的 access token 调用 OpenAI 已认证账号接口，补全实际使用的账号 ID、用户 ID、
+  邮箱和套餐；仅有 refresh token 时，先换取 access token 后执行同一补全。导入不要求 JWT payload 含有
+  特定字段，不执行 JWKS 验签，也不将文件中的身份字段与上游资料比对。
+- 首次 OAuth 保留回调 `state`、PKCE 与官方 token exchange；换得 access token 后调用同一已认证账号接口
+  补全账号资料。重新授权也保留这些回调保护，但只轮换目标账号的 token。回调地址只承载 `code`/`state`，
+  不以 host/path 形式作为拒绝条件。
 - 账号文件导入和首次 OAuth 创建在 credential 提交后立即尝试一次额度观测。观测失败只记录告警，
-  不回滚已提交的账号；重新授权只轮换目标账号 credential，不隐式等同于手工额度刷新。
-- 重新授权在 start 时绑定目标账号，complete 时重新读取目标的当前 credential revision，并验证上游
-  account/user 身份没有换绑。Free token 缺少 `chatgpt_account_id` 时，后端使用目标账号 header 调用
-  `/wham/usage` 交叉确认，而不是把该账号推断成 K12。
+  不回滚已提交的账号；重新授权和手工或后台 RT 刷新只更新 token，不隐式等同于手工额度刷新，也不更新
+  既有账号资料或 OAuth principal。
 - OAuth pending flow 先取得带过期时间的独占 claim，只有账号事务提交成功后才消费。失败会释放 claim，
   但上游 authorization code 本身通常只能交换一次；已完成过 token exchange 时应重新创建 OAuth flow。
 - `GET /accounts/quota` 只读取最后一次落库快照；`POST /accounts/quota/refresh` 才访问上游。access token
   已过期时，额度刷新要求先走 credential 刷新或重新授权，不会拿过期 token 探测额度。
 - 成功额度观测会 revision-fenced 写入 quota，并在 `active` 与 `quota_exhausted` 等额度所属状态间同步；
   不会清除 `expired`、`banned` 等 credential/封禁终态。额度接口的 401/403 也不足以判定 refresh token
-  永久失效，credential 终态只由 OAuth refresh 或身份校验链路写入。
+  永久失效，credential 终态只由 OAuth refresh 的明确永久错误写入。
 - 正常 Responses 请求会解析上游响应的 rate-limit headers，合并进同一 quota 快照并同步状态。Free、
   K12 等套餐共用该状态机；套餐只参与账号展示和按套餐隔离的模型目录 cache，不存在 K12 专属额度路径。
 - 账号页没有定时静默轮询。手工额度刷新完成后页面会重新读取该账号；请求驱动或后台任务产生的状态