diff --git a/AGENTS.md b/AGENTS.md
index 747a9dc5..60819108 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -53,7 +53,10 @@ were name-only across ALL libraries. Fixed; do not regress.
   broke on 2026-10-14 within hours of session capture shipping).
   Tests that need deterministic data use `setup.RegularToken` +
   `getRegularUserID` (the fresh per-test user), or assert ranges
-  instead of equals.
+  instead of equals. Related: **no table-wide DELETEs in test
+  cleanups** — sync_integration's `DELETE FROM reading_history WHERE
+  true` destroyed the live reader's stats data on every suite run
+  until scoped to the test-sync users (cascade via user delete).
 
 ## Timezones
 
diff --git a/cmd/server/tests/sync_integration_test.go b/cmd/server/tests/sync_integration_test.go
index 9b49c8a0..6b90f661 100644
--- a/cmd/server/tests/sync_integration_test.go
+++ b/cmd/server/tests/sync_integration_test.go
@@ -32,12 +32,16 @@ func setupSyncTestDB(t *testing.T) *database.Queries {
 
 	db := database.New(dbPool)
 	t.Cleanup(func() {
+		// Scoped wipes ONLY: reading_history now carries real captured
+		// sessions for the shared dev accounts, so a table-wide DELETE
+		// here silently destroyed a live reader's stats data on every
+		// suite run (2026-10-14). Sync-test sessions belong to the
+		// test-sync users; deleting those users cascades their rows.
 		_, _ = dbPool.Exec(ctx, "DELETE FROM sync_queue WHERE true")
-		_, _ = dbPool.Exec(ctx, "DELETE FROM reading_history WHERE true")
+		_, _ = dbPool.Exec(ctx, "DELETE FROM users WHERE email LIKE 'test-sync%'")
 		_, _ = dbPool.Exec(ctx, "DELETE FROM media_items WHERE title LIKE 'Test %'")
 		_, _ = dbPool.Exec(ctx, "DELETE FROM libraries WHERE name LIKE 'Test %'")
 		_, _ = dbPool.Exec(ctx, "DELETE FROM devices WHERE device_name LIKE 'Test %'")
-		_, _ = dbPool.Exec(ctx, "DELETE FROM users WHERE email LIKE 'test-sync%'")
 		dbPool.Close()
 	})