#840 fix(chat): stop the transcript settle effect from looping

Problem

• Every open chat thread burns CPU re-rendering the whole message list, and a thread left open long enough crashes the panel with React error #185.
• The settle effect in useWorkspaceAiChat.ts calls setMessages(transcriptData) with no equality guard. The AI SDK copies the array on every set, so messages identity changes, the effect (which lists messages) re-fires, and the two feed each other forever.
• experimental_throttle: 100 paces the loop at roughly ten passive-effect updates per second, so React only throws once the nested passive-update count crosses fifty — about five seconds in.
• The guard passes for any settled thread that has a transcript, so this is the normal idle state of the core chat surface, not an edge case. The recorded exception count is a floor: most sessions leave the thread before the limit trips and only pay the wasted renders.

Changes

• Deep-compare the server snapshot against the local messages before setting. After the first set the two match, so the next effect run returns early and the cycle breaks.
• Added transcriptsEqual next to serverTranscriptAdvanced, reusing the same JSON.stringify comparison the recovery path already relies on.
• The compare runs only when the thread is settled and ready. During streaming the effect still returns early before the compare, so there is no per-chunk cost.
• Added two unit tests: a fresh copy of the same transcript reads equal (the effect stops), and durable metadata missing from the local row reads as a difference (the snapshot still wins).

Test plan

• pnpm test for the affected file (6 passed)
• pnpm check (format, lint, types — clean)

Why

Users with a chat thread open sit in a passive-effect render loop that wastes CPU on the product's core chat surface and eventually crashes the panel.

---

Created with PostHog Desktop from this inbox report.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}