From 18df606127e9ccc4fd338f18d31714d63a17255e Mon Sep 17 00:00:00 2001
From: James <james.russo@heygen.com>
Date: Mon, 18 May 2026 21:25:46 +0000
Subject: [PATCH 1/2] chore(engine): structured logging for sub-composition
 timeline poll
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Captures host IDs, timeline IDs before/after the poll, the diff set, and
whether __hfForceTimelineRebind fired — emitted as a single JSON log line
per render. Lets us correlate flaky regression runs (style-7-prod,
gsap-letters-render-compat, style-3-prod) with whether the count-based
rebind heuristic fired.

Behavior is unchanged: the rebind condition `addedDuringPoll.length > 0`
is equivalent to the prior `timelinesAfterPoll > timelinesBeforePoll`
under the invariant that timeline IDs are never removed mid-poll.

Intended to be reverted once the race condition is confirmed and patched.
---
 packages/engine/src/services/frameCapture.ts | 40 ++++++++++++++++----
 1 file changed, 33 insertions(+), 7 deletions(-)

diff --git a/packages/engine/src/services/frameCapture.ts b/packages/engine/src/services/frameCapture.ts
index 62b3729e7..8109489a0 100644
--- a/packages/engine/src/services/frameCapture.ts
+++ b/packages/engine/src/services/frameCapture.ts
@@ -365,20 +365,46 @@ async function pollSubCompositionTimelines(
     }
     return true;
   })()`;
-  const timelinesBeforePoll = Number(
-    await page.evaluate(`Object.keys(window.__timelines || {}).length`),
-  );
+  // Observability snapshot: capture host IDs + timeline IDs before and after
+  // the poll so flaky CI runs can be correlated with whether the count-based
+  // rebind heuristic fired. Temporary — drop once the race condition behind
+  // the regression flakes is confirmed and patched.
+  const beforeSnapshot = (await page.evaluate(`(function() {
+    var hosts = document.querySelectorAll("[data-composition-id]");
+    var hostIds = [];
+    for (var i = 0; i < hosts.length; i++) {
+      var id = hosts[i].getAttribute("data-composition-id");
+      if (id) hostIds.push(id);
+    }
+    return { hostIds: hostIds, timelineIds: Object.keys(window.__timelines || {}) };
+  })()`)) as { hostIds: string[]; timelineIds: string[] };
+  const pollStart = Date.now();
   const ready = await pollPageExpression(page, expression, timeoutMs, intervalMs);
-  const timelinesAfterPoll = Number(
-    await page.evaluate(`Object.keys(window.__timelines || {}).length`),
-  );
-  if (ready && timelinesAfterPoll > timelinesBeforePoll) {
+  const pollMs = Date.now() - pollStart;
+  const afterTimelineIds = (await page.evaluate(
+    `Object.keys(window.__timelines || {})`,
+  )) as string[];
+  const beforeSet = new Set(beforeSnapshot.timelineIds);
+  const addedDuringPoll = afterTimelineIds.filter((id) => !beforeSet.has(id));
+  const rebindFired = ready && addedDuringPoll.length > 0;
+  if (rebindFired) {
     await page.evaluate(`(function() {
       if (typeof window.__hfForceTimelineRebind === "function") {
         window.__hfForceTimelineRebind();
       }
     })()`);
   }
+  console.log(
+    `[FrameCapture] pollSubCompositionTimelines ${JSON.stringify({
+      hostIds: beforeSnapshot.hostIds,
+      timelineIdsBefore: beforeSnapshot.timelineIds,
+      timelineIdsAfter: afterTimelineIds,
+      addedDuringPoll,
+      pollMs,
+      ready,
+      rebindFired,
+    })}`,
+  );
   if (!ready) {
     const missing = await page.evaluate(`(function() {
       var hosts = document.querySelectorAll("[data-composition-id]");

From ddc6c2e16ee88b3180beec01f59db4d24a1a7437 Mon Sep 17 00:00:00 2001
From: James <james.russo@heygen.com>
Date: Mon, 18 May 2026 21:27:04 +0000
Subject: [PATCH 2/2] ci: disable fail-fast in regression matrix for this
 investigation

Lets every shard run to completion and emit its pollSubCompositionTimelines
JSON log line, so we can correlate rebindFired across all shards on a single
run instead of only seeing data from whichever shard happened to fail first.

Restore fail-fast: true before merging or reverting this PR.
---
 .github/workflows/regression.yml | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/regression.yml b/.github/workflows/regression.yml
index 62e2bc503..b33f4c477 100644
--- a/.github/workflows/regression.yml
+++ b/.github/workflows/regression.yml
@@ -50,7 +50,11 @@ jobs:
     runs-on: ubuntu-latest
     timeout-minutes: 60
     strategy:
-      fail-fast: true
+      # Temporarily disabled for the regression-flake investigation on
+      # chore/regression-poll-observability — we need every shard to emit
+      # its pollSubCompositionTimelines log even when an early shard fails.
+      # Restore to `true` before merging this PR (or before reverting it).
+      fail-fast: false
       matrix:
         # Shards are bin-packed by measured per-test duration (LPT heuristic on
         # CI run 25893372795) so each row carries ~15-16 min of work. When a