Observed behavior
Native macos-15-intel promotion smoke timed out twice at a 90-second deadline after successful Guardian takeover. The second attempt also spawned a real Shell, but did not reach the PTY connect sentinel.
A diagnostic-only run passed all checks and localized the delay:
- Main returned the Workspace creation response (HTTP 201) at 2026-09-04T16:17:23.386Z.
- Renderer observed response headers at 16:18:24.943Z (~61.56 seconds later).
- Shell spawn, PTY attachment and CLI socket round trip then completed normally; total takeover smoke was 88 seconds.
- Packaged Workspace/Pi and previous-desktop upgrade also passed.
Evidence: https://github.com/TraderAlice/OpenAlice/actions/runs/33893555127 (Intel job 101091105278). Initial failures: https://github.com/TraderAlice/OpenAlice/actions/runs/33891119539 (attempts 1 and 2). Local Apple Silicon takeover and packaged acceptance passed promptly.
Suspected boundary and next investigation
apps/desktop/src/main.ts app protocol handler / renderer fetch, apps/desktop/src/ipc.ts and Chromium renderer scheduling. The backend had already produced a successful response. Evidence does not yet distinguish renderer CPU/scheduling, native protocol delivery, or hosted-machine effects; no root cause is claimed.
Capture renderer long tasks, app-protocol timing and native process/CPU evidence on Intel. Compare source and packaged modes, including a simple protocol response, before adding runtime workarounds.
Why deferred
The diagnostic run accepted every functional assertion and final packaged/upgrade path. A bounded smoke deadline of 180 seconds avoids treating the observed successful 88-second path as a guaranteed failure at 90; this does not fix or dismiss the delivery delay. Keep this performance finding separate from release artifact correctness. Related promotion #1349.
Observed behavior
Native macos-15-intel promotion smoke timed out twice at a 90-second deadline after successful Guardian takeover. The second attempt also spawned a real Shell, but did not reach the PTY connect sentinel.
A diagnostic-only run passed all checks and localized the delay:
Evidence: https://github.com/TraderAlice/OpenAlice/actions/runs/33893555127 (Intel job 101091105278). Initial failures: https://github.com/TraderAlice/OpenAlice/actions/runs/33891119539 (attempts 1 and 2). Local Apple Silicon takeover and packaged acceptance passed promptly.
Suspected boundary and next investigation
apps/desktop/src/main.ts app protocol handler / renderer fetch, apps/desktop/src/ipc.ts and Chromium renderer scheduling. The backend had already produced a successful response. Evidence does not yet distinguish renderer CPU/scheduling, native protocol delivery, or hosted-machine effects; no root cause is claimed.
Capture renderer long tasks, app-protocol timing and native process/CPU evidence on Intel. Compare source and packaged modes, including a simple protocol response, before adding runtime workarounds.
Why deferred
The diagnostic run accepted every functional assertion and final packaged/upgrade path. A bounded smoke deadline of 180 seconds avoids treating the observed successful 88-second path as a guaranteed failure at 90; this does not fix or dismiss the delivery delay. Keep this performance finding separate from release artifact correctness. Related promotion #1349.