perf(server): drop duplicate provider session status events

[tusharbhardwaj-bk]

Problem

Claude reports system/status: requesting on every model request inside a turn. The Claude adapter maps each one to session.state.changed, and ProviderRuntimeIngestion turned every one into a thread.session.set whose only new value was updatedAt.

In the last 20k prod events, 2,040 of 2,666 thread.session-set (77%) were exact repeats of the previous one apart from updatedAt. One thread had 213 in 3.6 h. Each one costs:

• an event row and a projection write;
• a push to every shell and thread subscriber;
• a replay item on reconnect;
• for non-running statuses, a ThreadPullRequestReactor sync.

Fix

• New fork module orchestration/sessionSetDedupe.expbkt3.ts. It skips the dispatch when the next session equals the projected thread.session in status, activeTurnId, lastError, providerThreadId, runtimeMode, providerName and providerInstanceId.
  • The last two are an addition to the audit's list, so a provider or instance switch with the same status is never swallowed.
• One marked seam in ProviderRuntimeIngestion.ts: the upstream orchestrationEngine.dispatch({ call becomes dispatchSessionSetUnlessNoOp(...)({. The command literal is untouched.
• Only session.state.changed is filtered. session.started, thread.started, turn.*, session.exited and runtime.error always dispatch.

Why this is safe:

• Projection writes run inside the dispatch transaction (OrchestrationEngine.ts withTransaction → projectEventDeferred), so the projected session is current when ingestion compares.

• The readers of session.updatedAt all key off a real transition, which is never dropped:

  • the reaper idle window (ProviderSessionReaper.ts:340);
  • the settled-turn completion time;
  • snooze wake-on-error (ThreadSettlementPolicy.ts:129).

• Recorded edge case: the fork branch of ProviderSessionReaper keeps a session alive when the adapter still holds a turn but the projection's activeTurnId is null. It does this through latestThreadEventAt within the 30 min threshold. Before this PR, status pings produced session-set events that kept that timestamp fresh; now only real events do. This only matters if the projection has already lost activeTurnId and the agent emits nothing else for 30 min, so it is accepted.

Evidence

Fork test ProviderRuntimeIngestion.expbkt3.test.ts, 10 identical running pings in one burst:

	thread.session-set events persisted
before (dedupe disabled)	10
after	1

Real transitions (running → ready → running) and a changed error message still land.

Tests run:

• NODE_ENV=test vp test run on sessionSetDedupe.expbkt3.test.ts, ProviderRuntimeIngestion.expbkt3.test.ts, and the existing ProviderRuntimeIngestion.test.ts + .approval.test.ts: 89 + 77 pass.
• vp run typecheck in apps/server is clean.
• The fork-marker check passes.

After the expbkt3 deploy I will re-count duplicate session-sets on the expbkt3 DB.

Candidate to upstream: the same no-op comparison would help every T3 Code user.

Model/harness: Claude Opus 5.5 via Claude Code in T3 Code.

🤖 Generated with Claude Code

---

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

[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	—	10	—	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: 7ac7f45 · 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.