PR 33000 real dev verification, Linux, 2026-11-10
main: bcee0f5d2872ba193c66ed9ba979bbe424f89ff9
PR:   79ed278fd131f12b1a42ca90db1df412b8bfe60c
Browser: repo Playwright + /usr/bin/google-chrome. Provider: Codex gpt-6.1-sol.
Project and conversation were created with the web UI in a real git repository.
The stopped main server and PR server used separate state. PR began with a read-only backup of the app-created pre-recall baseline.

main post-compaction send: Success (server trace)
ProjectionStore.readCanonicalProjection: Success
  SELECT payload_json FROM orchestration_v2_projection_runs WHERE thread_id = ? ORDER BY ordinal ASC
  SELECT payload_json FROM orchestration_v2_projection_run_attempts WHERE thread_id = ? ORDER BY run_id ASC, attempt_ordinal ASC
  SELECT payload_json FROM orchestration_v2_projection_provider_turns WHERE thread_id = ? ORDER BY provider_thread_id ASC…[truncated]
  SELECT payload_json FROM orchestration_v2_projection_context_transfers WHERE (source_thread_id = ? OR target_thread_id = ?) ORDER BY row…[truncated]

PR post-compaction send: Success (server trace)
ProjectionStore.runsSinceLastCompaction: Success
  SELECT COALESCE( (SELECT previous.ordinal FROM orchestration_v2_projection_runs AS previous WHERE previous.thread_id = ? AND previous.ordinal < latest.ordinal …[truncated]
ProjectionStore.readCanonicalProjection: Success
  SELECT payload_json FROM (SELECT run_id AS recent_run_id FROM ( SELECT run_id FROM orchestration_v2_projection_runs WHERE thread_id = ? AND ordinal…[truncated]
  SELECT payload_json FROM orchestration_v2_projection_run_attempts WHERE thread_id = ? AND run_id IN ( SELECT run_id FROM orchestration_…[truncated]
  SELECT payload_json FROM orchestration_v2_projection_provider_turns INDEXED BY orchestration_v2_projection_provider_turns_attempt_idx WHERE thread_id =…[truncated]
  SELECT payload_json FROM orchestration_v2_projection_context_transfers WHERE target_thread_id = ? AND status IN ('pending', 'running', 'waiting') …[truncated]

Persisted real run results:
[
  {
    "ordinal": 1,
    "status": "completed",
    "freshContext": null,
    "modelSelection": "{\"instanceId\":\"codex\",\"model\":\"gpt-6.1-sol\"}"
  },
  {
    "ordinal": 2,
    "status": "completed",
    "freshContext": null,
    "modelSelection": "{\"instanceId\":\"codex\",\"model\":\"gpt-6.1-sol\"}"
  },
  {
    "ordinal": 3,
    "status": "completed",
    "freshContext": null,
    "modelSelection": "{\"instanceId\":\"codex\",\"model\":\"gpt-6.1-sol\"}"
  },
  {
    "ordinal": 4,
    "status": "completed",
    "freshContext": null,
    "modelSelection": "{\"instanceId\":\"codex\",\"model\":\"gpt-6.1-sol\"}"
  },
  {
    "ordinal": 5,
    "status": "completed",
    "freshContext": 1,
    "modelSelection": "{\"instanceId\":\"codex\",\"model\":\"gpt-6.1-sol\"}"
  },
  {
    "ordinal": 6,
    "status": "completed",
    "freshContext": null,
    "modelSelection": "{\"instanceId\":\"codex\",\"model\":\"gpt-6.1-sol\"}"
  }
]

Compaction:
[{"type":"compaction","status":"completed"}]

Native provider thread IDs (fresh session created a second ID):
01a1c4f4-1963-7dc1-b9ab-bc82f687d06e
01a1c4fa-8cca-73a0-86c1-6a5b094943b7

Observed in UI: MARKER SAVED; TURN TWO; Context compacted; cobalt-maple-33000; New session; FRESH SESSION OK; cat README.md; compaction verification.
Thread title remained Save Verification Marker. Reload retained history and results.

Remote checks: 176 focused tests passed.
blacksmith-check run --id <task-owned-box> "vp run check:changed ade473018d32003658b7d38693e083ee2e314bd8"
check:changed summary
  pass  format  0.2s
  pass  lint  0.7s
  pass  knip 1/2  5.3s
  pass  knip 2/2  11.6s
  pass  graphql (apps/server)  0.4s
  pass  typecheck (3 packages)  5.9s
  pass  test (13 files)  30.9s
  CI also runs: full test suites, build
  unused patch check skipped: no dependency files changed
  pass not cached: files changed during the run



Limit: small app-created history. This reproduces main's full-history query path, not its reported 9-14 second large-thread delay. No large-history speedup claimed.
Not exercised: native desktop/mobile, other providers, cross-provider switch, live subagent completion, explicit steering to an older finished run, large-database migration cost.