perf(desktop): a starting local backend no longer delays connecting to the primary by 10 s

[tusharbhardwaj-bk]

Problem. Reopening the managed desktop app ("StageBKT3Code") sometimes took about 10 s before any message synced. The expbkt3 traces for this app show the same gap on 2 of 3 window opens (08:34:51 → 08:35:02 and 14:13:53 → 14:14:03 on 2026-09-27): the renderer fetched /api/auth/session, /api/orchestration/users and /.well-known/t3/environment straight away, then waited ~10.5 s before it requested a WebSocket ticket.

Cause. buildPlatformRegistrations builds every registration before returning any. After the primary it walks the desktop-local backends one by one. A managed build registers its bundled backend and starts it after the window opens. While that backend is starting, its readiness middleware holds every request, so the descriptor fetch waited out its default 10 s timeout. The primary was installed, and the supervisor allowed to connect, only after that. The loopback requests never reach the remote server, which is why its traces show an empty gap.

Fix. Pass a 1.5 s timeoutMs to the desktop-local descriptor fetch (desktopLocalDiscovery.expbkt3.ts, one marked seam in platform.ts). A ready loopback backend answers in milliseconds. One still starting is skipped for this poll and retried on the next, 3 s later. The bearer bootstrap is untouched: it runs only once the descriptor has answered, and by then the backend is ready.

Verification.

• Web typecheck and lint are clean.
• No unit test: the change is a timeout value, and the platform source has no test harness.
• The proof is the next window open on a staging desktop build: the gap between /.well-known/t3/environment and /api/auth/websocket-ticket in the expbkt3 traces should be under 2 s.

🤖 Generated with Claude Code

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	13.8 KiB	—	15.1 KiB	✅
Codex	Thread snapshot wire	—	7.3 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket wire	—	6.5 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket decoded	—	57.1 KiB	—	66.4 KiB	✅
Codex	Live turn messages	—	9	—	21	✅
Claude	Total thread wire	—	13.8 KiB	—	15.1 KiB	✅
Claude	Thread snapshot wire	—	7.3 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket wire	—	6.5 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket decoded	—	57.9 KiB	—	66.4 KiB	✅
Claude	Live turn messages	—	9	—	21	✅

Baseline: unavailable · PR result: a722a0e · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 114.9 KiB
• Claude decoded thread snapshot: 115.6 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.