From 069f71bd58862eb8b549d52d8631811c48df34db Mon Sep 17 00:00:00 2001
From: Neil <4138956+nwparker@users.noreply.github.com>
Date: Sat, 1 Aug 2026 07:57:43 -0700
Subject: [PATCH] test(dictation): pin the churn clock and the window's lower
 bound

The two 250-refresh churn tests measured real elapsed time against a 5s
window, so a loaded runner that took longer than one window to run the
loop would roll the window and go red. Pin Date.now in both.

Pinning alone leaves CHURN_WINDOW_MS unpinned below: shrinking it 5_000 ->
50 kept all 13 tests green, yet a 50ms window can never accumulate 250
refreshes and the detector would be dead. Add a storm spread across most
of one window so the constant has to be wide enough to hold a sustained
storm, not just an instantaneous burst.

Co-authored-by: Orca <help@stably.ai>
---
 ...oice-model-download-progress-storm.test.tsx |  4 ++++
 ...dictation-model-state-stabilisation.test.ts | 18 ++++++++++++++++++
 2 files changed, 22 insertions(+)

diff --git a/src/renderer/src/components/settings/voice-model-download-progress-storm.test.tsx b/src/renderer/src/components/settings/voice-model-download-progress-storm.test.tsx
index 4744c57988d..a482ca54e2b 100644
--- a/src/renderer/src/components/settings/voice-model-download-progress-storm.test.tsx
+++ b/src/renderer/src/components/settings/voice-model-download-progress-storm.test.tsx
@@ -112,6 +112,7 @@ afterEach(() => {
   act(() => root?.unmount())
   document.body.replaceChildren()
   vi.unstubAllGlobals()
+  vi.restoreAllMocks()
 })
 
 /**
@@ -191,6 +192,9 @@ describe('speech model download progress storm (render pressure)', () => {
   })
 
   it('records the churn breadcrumb when refreshes outrun coalescing', async () => {
+    // Why pin Date.now: the threshold is per 5s window, so a loaded machine that
+    // takes longer than one window to run the loop would measure itself, not the rule.
+    vi.spyOn(Date, 'now').mockReturnValue(1_000_000)
     // Own store instance: the churn window is per-slice state the other tests advance.
     const churnStore = create<DictationTestStore>(dictationSlice)
     for (let refresh = 0; refresh < 250; refresh += 1) {
diff --git a/src/renderer/src/store/slices/dictation-model-state-stabilisation.test.ts b/src/renderer/src/store/slices/dictation-model-state-stabilisation.test.ts
index 3f0bd908e1e..c0a8d07bd52 100644
--- a/src/renderer/src/store/slices/dictation-model-state-stabilisation.test.ts
+++ b/src/renderer/src/store/slices/dictation-model-state-stabilisation.test.ts
@@ -124,6 +124,7 @@ describe('dictation model-state stabilisation', () => {
   })
 
   it('reports churn once the renderer is driven far past whole-percent pacing', async () => {
+    pinClock()
     const store = newStore()
     reply = [{ id: 'whisper-tiny', status: 'downloading', progress: 0.42 }]
     for (let refresh = 0; refresh < 250; refresh += 1) {
@@ -135,6 +136,23 @@ describe('dictation model-state stabilisation', () => {
     expect(churn[0].data).toMatchObject({ refreshes: 250, noOpRefreshes: 249 })
   })
 
+  // Why spread across the window: a burst-only test passes for any window length,
+  // including one too short for a real storm to ever accumulate the threshold. The
+  // window has to be wide enough to hold a sustained storm, not just an instant one.
+  it('still reports a storm that is spread across most of one window', async () => {
+    const clock = pinClock()
+    const store = newStore()
+    reply = [{ id: 'whisper-tiny', status: 'downloading', progress: 0.42 }]
+    for (let refresh = 0; refresh < 250; refresh += 1) {
+      await store.getState().refreshModelStates()
+      clock.advance(19)
+    }
+
+    const churn = churnBreadcrumbs()
+    expect(churn).toHaveLength(1)
+    expect(churn[0].data).toMatchObject({ refreshes: 250, windowMs: 249 * 19 })
+  })
+
   // Stopping at exactly the threshold cannot tell "fires once" from "fires on every
   // refresh after", and the second floods the 30-entry ring this breadcrumb is
   // coalesced into to protect.