diff --git a/src/table/src/table/scan.rs b/src/table/src/table/scan.rs
index 1bfaea0d1df..fc7aaa24dc6 100644
--- a/src/table/src/table/scan.rs
+++ b/src/table/src/table/scan.rs
@@ -522,13 +522,21 @@ impl ExecutionPlan for RegionScanExec {
         let statistics = if scanner.properties().total_rows_is_exact()
             && !scanner.has_predicate_without_region()
         {
+            // Only the row count is known. DataFusion answers `count(col)` as
+            // `num_rows - null_count` and `count(DISTINCT col)` as `distinct_count`
+            // without scanning, so these must stay absent unless the schema
+            // guarantees them. Non-nullable columns (including the time index that
+            // `count(*)` is rewritten to) keep an exact zero null count.
             let column_statistics = self
                 .arrow_schema
                 .fields
                 .iter()
-                .map(|_| ColumnStatistics {
-                    distinct_count: Precision::Exact(self.total_rows),
-                    null_count: Precision::Exact(0), // all null rows are counted for append-only table
+                .map(|field| ColumnStatistics {
+                    null_count: if field.is_nullable() {
+                        Precision::Absent
+                    } else {
+                        Precision::Exact(0)
+                    },
                     ..Default::default()
                 })
                 .collect();
@@ -1218,7 +1226,8 @@ mod test {
             (true, true, Precision::Exact(3)),
             (false, true, Precision::Absent),
         ] {
-            let (schema, batch, metadata) = dynamic_filter_test_data(5685, false);
+            // Column `a` is nullable and `ts` is not.
+            let (schema, batch, metadata) = dynamic_filter_test_data(5685, true);
             let scanner = Box::new(RepeatableScanner {
                 batches: RecordBatches::try_new(schema, vec![batch]).unwrap(),
                 properties: ScannerProperties::default()
@@ -1229,13 +1238,19 @@ mod test {
                 dynamic_filters: Arc::new(Mutex::new(Vec::new())),
             });
             let plan = RegionScanExec::new(scanner, ScanRequest::default(), None).unwrap();
-            assert_eq!(
-                StatisticsContext::new()
-                    .compute(&plan, &StatisticsArgs::new())
-                    .unwrap()
-                    .num_rows,
-                expected,
-            );
+            let stats = StatisticsContext::new()
+                .compute(&plan, &StatisticsArgs::new())
+                .unwrap();
+            assert_eq!(stats.num_rows, expected);
+            if expected != Precision::Absent {
+                let [a, ts] = stats.column_statistics.as_slice() else {
+                    panic!("unexpected column statistics: {stats:?}");
+                };
+                assert_eq!(a.null_count, Precision::Absent);
+                assert_eq!(ts.null_count, Precision::Exact(0));
+                assert_eq!(a.distinct_count, Precision::Absent);
+                assert_eq!(ts.distinct_count, Precision::Absent);
+            }
             assert_eq!(
                 StatisticsContext::new()
                     .compute(&plan, &StatisticsArgs::new().with_partition(Some(0)))
diff --git a/tests/cases/standalone/common/aggregate/append_count_nulls.result b/tests/cases/standalone/common/aggregate/append_count_nulls.result
new file mode 100644
index 00000000000..acb4714224b
--- /dev/null
+++ b/tests/cases/standalone/common/aggregate/append_count_nulls.result
@@ -0,0 +1,165 @@
+-- count(col) and count(DISTINCT col) on append-only tables must not be answered
+-- from row-count statistics, while count(*) still is.
+CREATE TABLE n1 (ts TIMESTAMP TIME INDEX, v BIGINT, s STRING) WITH (append_mode='true');
+
+Affected Rows: 0
+
+INSERT INTO n1 VALUES (1000, 1, 'a'), (2000, NULL, 'a'), (3000, NULL, NULL), (4000, 4, 'b');
+
+Affected Rows: 4
+
+SELECT count(*), count(v), count(s), count(DISTINCT s) FROM n1;
+
++----------+-------------+-------------+----------------------+
+| count(*) | count(n1.v) | count(n1.s) | count(DISTINCT n1.s) |
++----------+-------------+-------------+----------------------+
+| 4        | 2           | 3           | 2                    |
++----------+-------------+-------------+----------------------+
+
+SELECT count(v) FROM n1;
+
++-------------+
+| count(n1.v) |
++-------------+
+| 2           |
++-------------+
+
+-- SQLNESS REPLACE (metrics.*) REDACTED
+-- SQLNESS REPLACE (RoundRobinBatch.*) REDACTED
+-- SQLNESS REPLACE (-+) -
+-- SQLNESS REPLACE (\s\s+) _
+-- SQLNESS REPLACE (peers.*) REDACTED
+-- SQLNESS REPLACE region=\d+\(\d+,\s+\d+\) region=REDACTED
+EXPLAIN ANALYZE SELECT count(*) FROM n1;
+
++-+-+-+
+| stage | node | plan_|
++-+-+-+
+| 0_| 0_|_CooperativeExec REDACTED
+|_|_|_MergeScanExec: REDACTED
+|_|_|_|
+| 1_| 0_|_ProjectionExec: expr=[4 as count(*)] REDACTED
+|_|_|_PlaceholderRowExec REDACTED
+|_|_|_|
+|_|_| Total rows: 1_|
++-+-+-+
+
+ADMIN FLUSH_TABLE('n1');
+
++-------------------------+
+| ADMIN FLUSH_TABLE('n1') |
++-------------------------+
+| 0                       |
++-------------------------+
+
+SELECT count(*), count(v), count(s), count(DISTINCT s) FROM n1;
+
++----------+-------------+-------------+----------------------+
+| count(*) | count(n1.v) | count(n1.s) | count(DISTINCT n1.s) |
++----------+-------------+-------------+----------------------+
+| 4        | 2           | 3           | 2                    |
++----------+-------------+-------------+----------------------+
+
+-- SQLNESS REPLACE (metrics.*) REDACTED
+-- SQLNESS REPLACE (RoundRobinBatch.*) REDACTED
+-- SQLNESS REPLACE (-+) -
+-- SQLNESS REPLACE (\s\s+) _
+-- SQLNESS REPLACE (peers.*) REDACTED
+-- SQLNESS REPLACE region=\d+\(\d+,\s+\d+\) region=REDACTED
+EXPLAIN ANALYZE SELECT count(*) FROM n1;
+
++-+-+-+
+| stage | node | plan_|
++-+-+-+
+| 0_| 0_|_CooperativeExec REDACTED
+|_|_|_MergeScanExec: REDACTED
+|_|_|_|
+| 1_| 0_|_ProjectionExec: expr=[4 as count(*)] REDACTED
+|_|_|_PlaceholderRowExec REDACTED
+|_|_|_|
+|_|_| Total rows: 1_|
++-+-+-+
+
+DROP TABLE n1;
+
+Affected Rows: 0
+
+CREATE TABLE n2 (
+  host STRING,
+  ts TIMESTAMP TIME INDEX,
+  v BIGINT,
+  s STRING,
+  PRIMARY KEY (host)
+)
+PARTITION ON COLUMNS (host) (
+  host < 'm',
+  host >= 'm'
+)
+WITH (append_mode='true');
+
+Affected Rows: 0
+
+INSERT INTO n2 VALUES
+  ('a', 1000, 1, 'a'),
+  ('b', 2000, NULL, 'a'),
+  ('x', 3000, NULL, NULL),
+  ('y', 4000, 4, 'b');
+
+Affected Rows: 4
+
+-- count is pushed down to each region as a state aggregate.
+SELECT count(v), count(s) FROM n2;
+
++-------------+-------------+
+| count(n2.v) | count(n2.s) |
++-------------+-------------+
+| 2           | 3           |
++-------------+-------------+
+
+ADMIN FLUSH_TABLE('n2');
+
++-------------------------+
+| ADMIN FLUSH_TABLE('n2') |
++-------------------------+
+| 0                       |
++-------------------------+
+
+SELECT count(v), count(s) FROM n2;
+
++-------------+-------------+
+| count(n2.v) | count(n2.s) |
++-------------+-------------+
+| 2           | 3           |
++-------------+-------------+
+
+-- SQLNESS REPLACE (metrics.*) REDACTED
+-- SQLNESS REPLACE (RoundRobinBatch.*) REDACTED
+-- SQLNESS REPLACE (-+) -
+-- SQLNESS REPLACE (\s\s+) _
+-- SQLNESS REPLACE (peers.*) REDACTED
+-- SQLNESS REPLACE region=\d+\(\d+,\s+\d+\) region=REDACTED
+EXPLAIN ANALYZE SELECT count(*) FROM n2;
+
++-+-+-+
+| stage | node | plan_|
++-+-+-+
+| 0_| 0_|_ProjectionExec: expr=[count(Int64(1))@0 as count(*)] REDACTED
+|_|_|_AggregateExec: mode=Final, gby=[], aggr=[__count_merge(__count_state(n2.ts)) as count(Int64(1))] REDACTED
+|_|_|_CoalescePartitionsExec REDACTED
+|_|_|_AggregateExec: mode=Partial, gby=[], aggr=[__count_merge(__count_state(n2.ts)) as count(Int64(1))] REDACTED
+|_|_|_RepartitionExec: partitioning=REDACTED
+|_|_|_MergeScanExec: REDACTED
+|_|_|_|
+| 1_| 0_|_ProjectionExec: expr=[{count[count]:2} as __count_state(n2.ts)] REDACTED
+|_|_|_PlaceholderRowExec REDACTED
+|_|_|_|
+| 1_| 1_|_ProjectionExec: expr=[{count[count]:2} as __count_state(n2.ts)] REDACTED
+|_|_|_PlaceholderRowExec REDACTED
+|_|_|_|
+|_|_| Total rows: 1_|
++-+-+-+
+
+DROP TABLE n2;
+
+Affected Rows: 0
+
diff --git a/tests/cases/standalone/common/aggregate/append_count_nulls.sql b/tests/cases/standalone/common/aggregate/append_count_nulls.sql
new file mode 100644
index 00000000000..8f2e1f5c2f7
--- /dev/null
+++ b/tests/cases/standalone/common/aggregate/append_count_nulls.sql
@@ -0,0 +1,67 @@
+-- count(col) and count(DISTINCT col) on append-only tables must not be answered
+-- from row-count statistics, while count(*) still is.
+CREATE TABLE n1 (ts TIMESTAMP TIME INDEX, v BIGINT, s STRING) WITH (append_mode='true');
+
+INSERT INTO n1 VALUES (1000, 1, 'a'), (2000, NULL, 'a'), (3000, NULL, NULL), (4000, 4, 'b');
+
+SELECT count(*), count(v), count(s), count(DISTINCT s) FROM n1;
+
+SELECT count(v) FROM n1;
+
+-- SQLNESS REPLACE (metrics.*) REDACTED
+-- SQLNESS REPLACE (RoundRobinBatch.*) REDACTED
+-- SQLNESS REPLACE (-+) -
+-- SQLNESS REPLACE (\s\s+) _
+-- SQLNESS REPLACE (peers.*) REDACTED
+-- SQLNESS REPLACE region=\d+\(\d+,\s+\d+\) region=REDACTED
+EXPLAIN ANALYZE SELECT count(*) FROM n1;
+
+ADMIN FLUSH_TABLE('n1');
+
+SELECT count(*), count(v), count(s), count(DISTINCT s) FROM n1;
+
+-- SQLNESS REPLACE (metrics.*) REDACTED
+-- SQLNESS REPLACE (RoundRobinBatch.*) REDACTED
+-- SQLNESS REPLACE (-+) -
+-- SQLNESS REPLACE (\s\s+) _
+-- SQLNESS REPLACE (peers.*) REDACTED
+-- SQLNESS REPLACE region=\d+\(\d+,\s+\d+\) region=REDACTED
+EXPLAIN ANALYZE SELECT count(*) FROM n1;
+
+DROP TABLE n1;
+
+CREATE TABLE n2 (
+  host STRING,
+  ts TIMESTAMP TIME INDEX,
+  v BIGINT,
+  s STRING,
+  PRIMARY KEY (host)
+)
+PARTITION ON COLUMNS (host) (
+  host < 'm',
+  host >= 'm'
+)
+WITH (append_mode='true');
+
+INSERT INTO n2 VALUES
+  ('a', 1000, 1, 'a'),
+  ('b', 2000, NULL, 'a'),
+  ('x', 3000, NULL, NULL),
+  ('y', 4000, 4, 'b');
+
+-- count is pushed down to each region as a state aggregate.
+SELECT count(v), count(s) FROM n2;
+
+ADMIN FLUSH_TABLE('n2');
+
+SELECT count(v), count(s) FROM n2;
+
+-- SQLNESS REPLACE (metrics.*) REDACTED
+-- SQLNESS REPLACE (RoundRobinBatch.*) REDACTED
+-- SQLNESS REPLACE (-+) -
+-- SQLNESS REPLACE (\s\s+) _
+-- SQLNESS REPLACE (peers.*) REDACTED
+-- SQLNESS REPLACE region=\d+\(\d+,\s+\d+\) region=REDACTED
+EXPLAIN ANALYZE SELECT count(*) FROM n2;
+
+DROP TABLE n2;