PR #21314 — server orchestration behavior proof
Current local head: 74912d7dd34b3bb7f32ba5d7aa0d53ef58767ee9
Pinned main:1451e7cb23034d29774338608368a15d8845df36
Harness: Codex in T3 Code, headless Linux, disposable test fixtures.
These tests run persisted orchestration and fake-provider effects, not live provider CLIs or live GitHub PR polling.

BEFORE: focused PR regressions on main
     × keeps queued user selections and pauses only the limited provider 94ms
     × delivers a queued PR update on the current provider through automatic 1290ms
     × delivers a queued PR update on the current provider through queue.resume 358ms
 FAIL  apps/server/src/orchestration-v2/runtimeLayer.test.ts > OrchestrationV2LayerLive lifecycle > keeps queued user selections and pauses only the limited provider
AssertionError: expected 'starting' to equal 'queued'
 FAIL  apps/server/src/orchestration-v2/testkit/ProviderSwitch.integration.test.ts > orchestration v2 provider switching > delivers a queued PR update on the current provider through automatic
AssertionError: expected 'codex' to equal 'claudeAgent'
 FAIL  apps/server/src/orchestration-v2/testkit/ProviderSwitch.integration.test.ts > orchestration v2 provider switching > delivers a queued PR update on the current provider through queue.resume
 Test Files  2 failed (2)
      Tests  3 failed | 154 skipped (157)
   Duration  10.35s (transform 59%, import 30%, tests 10%, setup 1%)

BEFORE REVIEW FIXES: new regressions on416c14c
     × delivers a queued monitor update from provider after conversation selection {"instanceId":"codex","model":"gpt-5.4"} 74ms
     × delivers a queued monitor update from provider after conversation selection {"instanceId":"codex_alternate","model":"gpt-5.4"} 56ms
     × delivers a queued PR update on the current provider through automatic (compatible Codex) 810ms
     × delivers a queued PR update on the current provider through queue.resume (compatible Codex) 377ms
 FAIL  apps/server/src/orchestration-v2/runtimeLayer.test.ts > OrchestrationV2LayerLive lifecycle > delivers a queued monitor update from provider after conversation selection {"instanceId":"codex","model":"gpt-5.4"}
AssertionError: expected 'runtime-layer-queue-current-conversat…' to equal 'provider-thread:provider:codex:native…'
 FAIL  apps/server/src/orchestration-v2/runtimeLayer.test.ts > OrchestrationV2LayerLive lifecycle > delivers a queued monitor update from provider after conversation selection {"instanceId":"codex_alternate","model":"gpt-5.4"}
AssertionError: expected { instanceId: 'codex_alternate', …(1) } to deeply equal { instanceId: 'codex', …(1) }
 FAIL  apps/server/src/orchestration-v2/testkit/ProviderSwitch.integration.test.ts > orchestration v2 provider switching > delivers a queued PR update on the current provider through automatic (compatible Codex)
AssertionError: expected 'codex' to equal 'codex-alternate'
 FAIL  apps/server/src/orchestration-v2/testkit/ProviderSwitch.integration.test.ts > orchestration v2 provider switching > delivers a queued PR update on the current provider through queue.resume (compatible Codex)
AssertionError: expected [ { driver: 'codex', …(8) } ] to have a length of 2 but got 1
 Test Files  2 failed (2)
      Tests  4 failed | 3 passed | 155 skipped (162)
   Duration  11.97s (transform 55%, import 28%, tests 16%, setup 1%)

AFTER: primary agent focused verification at74912d7
 ✓ apps/server/src/orchestration-v2/runtimeLayer.test.ts > OrchestrationV2LayerLive lifecycle > keeps queued user selections and pauses only the limited provider 127ms
 ✓ apps/server/src/orchestration-v2/runtimeLayer.test.ts > OrchestrationV2LayerLive lifecycle > delivers a queued monitor update from server after conversation selection {"instanceId":"codex","model":"gpt-5.4"} 29ms
 ✓ apps/server/src/orchestration-v2/runtimeLayer.test.ts > OrchestrationV2LayerLive lifecycle > delivers a queued monitor update from provider after conversation selection {"instanceId":"codex","model":"gpt-5.4"} 36ms
 ✓ apps/server/src/orchestration-v2/runtimeLayer.test.ts > OrchestrationV2LayerLive lifecycle > delivers a queued monitor update from provider after conversation selection {"instanceId":"codex_alternate","model":"gpt-5.4"} 39ms
 ✓ apps/server/src/orchestration-v2/testkit/ProviderSwitch.integration.test.ts > orchestration v2 provider switching > delivers a queued PR update on the current provider through automatic (Claude) 1231ms
 ✓ apps/server/src/orchestration-v2/testkit/ProviderSwitch.integration.test.ts > orchestration v2 provider switching > delivers a queued PR update on the current provider through queue.resume (Claude) 732ms
 ✓ apps/server/src/orchestration-v2/testkit/ProviderSwitch.integration.test.ts > orchestration v2 provider switching > delivers a queued PR update on the current provider through automatic (compatible Codex) 841ms
 ✓ apps/server/src/orchestration-v2/testkit/ProviderSwitch.integration.test.ts > orchestration v2 provider switching > delivers a queued PR update on the current provider through queue.resume (compatible Codex) 665ms
 Test Files  2 passed (2)
      Tests  8 passed | 154 skipped (162)
   Duration  12.86s (transform 53%, import 28%, tests 17%, setup 1%)

AFTER: complete two affected suites at74912d7
 Test Files  2 passed (2)
      Tests  162 passed (162)
   Duration  71.10s (tests 79%, transform 14%, import 7%)

Other scoped checks: server typecheck0errors/0warnings (154 existing Effect suggestions); changed-file lint passed with2existing warnings; format and git diff --check passed.
Limitations: no live CLI takeover, real PR-watch polling, Claude/Grok native wake-buffer ingestion, macOS/Windows/iOS/Android/Electron, remote/relay/tunnel runtime verification. No UI code changes; screenshots not applicable.