refactor(chat): stop post-turn reconciliation; trust the anchor invariant

[vellum-apollo-bot]

PR 2b.3 of the chat-state-slimming workstream. The endgame piece.

What

With the anchor invariant landed in 2b.1 (#32012) — assistant emits messageId = state.lastAssistantMessageId, client preserves the anchor via first-id-wins lock in appendTextDelta and role-based finalize in finalizeMessageComplete — the post-turn reconciliation loop that fired on every message_complete and activity_state idle is now redundant. The live client state is already correct; the refetch existed only to paper over id drift.

Removes startReconciliationLoop calls from:

• handleAssistantActivityState (idle phase)
• handleMessageComplete

Drops the now-unused epoch parameter from both handler signatures.

What stays

Reconciliation still fires legitimately on:

• Load / switch — chat-page.tsx top-level wiring
• SSE reopen / post-watchdog reconnect — use-event-stream.ts:361,408
• POST resolve confirmation — use-send-message.ts:429

The dispatcher still uses epoch for stale-epoch guards on the dispatch side, and cancelReconciliation calls remain (in handleAssistantTextDelta, handleGenerationHandoff, tool-call-handlers) since those still-spawned loops can be cancelled by text deltas and generation handoff.

Diff

 apps/web/src/domains/chat/hooks/use-stream-event-handler.ts            |  4 ++--
 apps/web/src/domains/chat/utils/stream-handlers/message-handlers.test.ts | 14 ++++----------
 apps/web/src/domains/chat/utils/stream-handlers/message-handlers.ts    |  4 ----
 3 files changed, 6 insertions(+), 16 deletions(-)

Tests

Updated 6 tests in message-handlers.test.ts to drop the epoch arg and flip the startReconciliationLoop assertion from toHaveBeenCalledWith(1) to not.toHaveBeenCalled() on the idle + message_complete paths. Renamed the affected tests to make the no-reconcile expectation explicit.

• Lint clean
• tsc clean (same pre-existing baseline drift in unrelated use-assistant-lifecycle.ts / ai-page.tsx / billing files as main)
• bun test src/domains/chat matches main exactly: 1331 pass / 29 fail / 3 errors — zero new failures introduced. (The 29/3 baseline is cross-suite mock isolation flakiness; affected tests pass in isolation.)

Closes the chat-state-slimming endgame for the streaming path. Next up: PR 2c (client rewrites reconcile to pure id-match + deletes stableId).

[chatgpt-codex-connector]

💡 Codex Review

vellum-assistant/apps/web/src/domains/chat/utils/stream-handlers/message-handlers.ts

Lines 72 to 75 in 1acf802

	ctx.setMessages(finalizeOnIdle);
	const turnPhaseBefore = ctx.getTurnState().phase;
	ctx.turnActions.completeTurn();
	if (convId) {

Re-start reconciliation on terminal stream events

Removing reconciliation starts from both handleAssistantActivityState(...idle) and handleMessageComplete leaves normal SSE-driven turns with no post-turn reconcile trigger. In useSendMessage, startReconciliationLoop runs only on the poll fallback path (when there is no matching active stream), so the common "active stream" path now never enters the loop after terminal events. That regresses the silent-stall rescue path (onPollReconciled) and can leave stale streaming/tool-call state or missed final server-side message fields whenever terminal SSE delivery is imperfect but the connection stays open.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".