diff --git a/CHANGELOG.md b/CHANGELOG.md
index af841d4..a3bbf20 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,7 +8,16 @@ so this log can be regenerated from history (e.g. with `git-cliff`).
 
 ## [Unreleased]
 
-_Changes on `main` that have not yet been tagged._
+### Changed
+- `excalidraw.spec` resets per test. Its two long-standing "flaky" specs were
+  deterministic failures: all three tests bootstrap through `/setup`, but the
+  file reset once per file with `beforeAll`, and `POST /auth/setup` returns 410
+  once a user exists — `SetupPage` renders that error without navigating, so
+  tests 2 and 3 sat on `/setup` for the full 30s timeout. They passed on retry
+  only because a retry restarts the worker and re-runs `beforeAll`. That is why
+  it was always exactly two flaky specs: the number of tests after the first.
+  The suite now reports `42 passed` with no flakes, and the playwright job
+  dropped from ~10m to ~5m.
 
 ## [0.2.1] - 2026-08-17
 
diff --git a/e2e/flows/excalidraw.spec.ts b/e2e/flows/excalidraw.spec.ts
index 956a5b4..6f69f71 100644
--- a/e2e/flows/excalidraw.spec.ts
+++ b/e2e/flows/excalidraw.spec.ts
@@ -16,7 +16,14 @@ function reset() {
   );
 }
 
-test.beforeAll(reset);
+// beforeEACH, not beforeAll. All three tests below bootstrap through
+// /setup, and POST /auth/setup returns 410 (auth.setup_closed) once any user
+// exists — SetupPage then renders an error and does NOT navigate, so every
+// test after the first hung until the 30s timeout and only passed on retry
+// (a retry restarts the worker, which re-runs a beforeAll). That read as
+// flakiness for a long time; it was deterministic. Same reason
+// create-placement, new-doc-edit-mode and tree-reorder reset per test.
+test.beforeEach(reset);
 
 /**
  * Drive a rectangle into the open Excalidraw modal via the dev-only