PR #22425 runtime verification, 2026-10-30 UTC
main: 1da41f949950bda57f45dc73effb7543037c844f
pr: 67d41c66adf2ffce5de9d775e9c4328dce560d50
vp i completed in both checkouts. One dev server at a time.
main: t3 project add <real repository>; thread start --provider codex --model gpt-6.1-sol --wait --json
{"threadId":"mcp:704eaf38-a696-41bf-a2b5-0c8faec19e7a","runId":"run:thread:mcp%3A704eaf38-a696-41bf-a2b5-0c8faec19e7a:ordinal:1","status":"completed","reply":"MAIN_PROVIDER_OK","error":null}

pr: same real provider through isolated dev server
{"threadId":"mcp:c3bfa255-4f99-443b-b3a0-5c26db0d3eee","runId":"run:thread:mcp%3Ac3bfa255-4f99-443b-b3a0-5c26db0d3eee:ordinal:1","status":"completed","reply":"PR_PROVIDER_OK","error":null}

browser composer: run sleep 15, then reply ACTIVE_OK
while active: t3 thread send --json <thread> <prompt>
{"status":"queued","delivery":"queued"}
browser: ACTIVE_OK then QUEUED_TURN_OK rendered automatically
settings: added Codex second; composer selected its gpt-6.1-sol model; SECOND_INSTANCE_OK rendered
read-only persisted run projection:
[
  {
    "ordinal": 1,
    "status": "completed",
    "providerInstanceId": "codex",
    "modelSelection": {
      "instanceId": "codex",
      "model": "gpt-6.1-sol"
    }
  },
  {
    "ordinal": 2,
    "status": "completed",
    "providerInstanceId": "codex",
    "modelSelection": {
      "instanceId": "codex",
      "model": "gpt-6.1-sol"
    }
  },
  {
    "ordinal": 3,
    "status": "completed",
    "providerInstanceId": "codex",
    "modelSelection": {
      "instanceId": "codex",
      "model": "gpt-6.1-sol"
    }
  },
  {
    "ordinal": 4,
    "status": "completed",
    "providerInstanceId": "codex_codex_second",
    "modelSelection": {
      "instanceId": "codex_codex_second",
      "model": "gpt-6.1-sol"
    }
  }
]
blacksmith-check focused existing runtimeLayer regression: 2 passed, 102 filtered out.
github checks: 21 pass, 8 skipped, no pending/failures.
LIMITATION: no genuinely exhausted account available. No native usage_limit error observed. Original main bug and PR hold/resume/reset behavior are unverified. No mocks or injected errors.