Advance supplied promotion readiness inputs out of stale not-ready state

[ozand]

Summary

After #456 and #455 repairs, live eeepc no longer has source_commit_missing and subagent execution no longer blocks on bare_python_executor_command, but the freshest promotion candidate is still surfaced as not_ready_for_policy_review even though its readiness-input packet says the inputs are complete and the recommended next action is ready_for_policy_review.

Live evidence

Observed at 2026-05-02T14:30Z from the local dashboard after /collect.

Latest /api/promotions.items[0]:

{
  "identity_key": "promotion-635b34e7041d",
  "status": "not_ready_for_policy_review",
  "readiness_checks": {
    "schema_version": "promotion-readiness-inputs-v1",
    "artifact_present": true,
    "evidence_refs_present": true,
    "provenance_complete": true,
    "missing_inputs": []
  },
  "readiness_blocker": {
    "state": "ready_for_policy_review",
    "reason": "promotion_readiness_inputs_supplied",
    "missing_inputs": [],
    "recommended_next_action": "ready_for_policy_review"
  },
  "recommended_next_action": "ready_for_policy_review",
  "governance_packet.review_packet_status": "blocked_not_ready",
  "provenance.source_commit": "f9a71ba0e5a0d1aba4c5c4f84c1861d1eb965ef4"
}

Mission Control headline:

{
  "headline": "Blocked: promotion_candidate_not_ready_for_policy_review"
}

Expected behavior

When supply_missing_promotion_readiness_inputs finds no missing inputs (provenance_complete=true, missing_inputs=[]) the durable promotion candidate/governance packet should advance consistently to ready_for_policy_review (or be accepted if the runtime policy allows), instead of leaving the top-level status and governance packet in not_ready_for_policy_review / blocked_not_ready.

Acceptance criteria

• Diagnose the writer path that leaves top-level review_status / decision and governance_packet.review_packet_status stale after readiness inputs are supplied.
• Add a regression reproducing the live shape: complete readiness inputs but stale not-ready top-level/governance state.
• Fix the writer/normalizer so latest promotion surfaces a coherent ready_for_policy_review state when readiness inputs are supplied.
• Run focused and relevant full tests.
• Deploy to eeepc and verify /api/promotions.items[0] and /api/mission-control no longer report promotion_candidate_not_ready_for_policy_review for a candidate whose readiness inputs are complete.

[ozand]

#461 complete: supplied readiness no longer clobbered to not-ready

Root cause fixed in PR #462 / commit 2998ffa5047db482c39fae91b84d6799735a3761.

Root cause

nanobot/runtime/coordinator.py called supply_missing_promotion_readiness_inputs() and that helper correctly returned:

• state=ready_for_policy_review
• readiness_checks.missing_inputs=[]
• readiness_checks.provenance_complete=true
• recommended_next_action=ready_for_policy_review

But the surrounding not-ready branch then rewrote the durable promotion record back to stale blocked values:

• review_status=not_ready_for_policy_review
• decision=not_ready_for_policy_review
• decision_record=blocked_not_ready
• accepted_record=not_created_not_ready
• governance_packet.review_packet_status=blocked_not_ready

nanobot/runtime/state.py also lacked an explicit ready_for_policy_review replay branch and classified it as blocked/not_accepted.

Fix

• Preserve ready_for_policy_review from supplied readiness inputs in top-level candidate, report, latest.json, and outbox surfaces.
• Use decision_record=pending_operator_review_packet and accepted_record=null for pending review rather than blocked sentinels.
• Set governance_packet.review_packet_status=pending_operator_review.
• Add runtime-state replay classification for ready_for_policy_review.

Verification

RED before fix:

• python3 -m pytest tests/test_runtime_coordinator.py::test_cycle_promotes_supplied_readiness_inputs_to_ready_for_policy_review -q
  • failed because report["review_status"] stayed not_ready_for_policy_review
• python3 -m pytest tests/test_runtime_coordinator.py::test_load_runtime_state_classifies_ready_for_policy_review_as_pending_review -q
  • failed because replay state was blocked

GREEN after fix:

• python3 -m pytest tests/test_runtime_coordinator.py::test_cycle_promotes_supplied_readiness_inputs_to_ready_for_policy_review -q
  • 1 passed
• python3 -m pytest tests/test_runtime_coordinator.py::test_load_runtime_state_classifies_ready_for_policy_review_as_pending_review tests/test_runtime_coordinator.py::test_cycle_promotes_supplied_readiness_inputs_to_ready_for_policy_review -q
  • 2 passed
• python3 -m pytest tests/test_runtime_coordinator.py tests/test_promotion_workflow.py -q
  • 70 passed
• python3 -m pytest tests -q
  • 699 passed, 5 skipped
• PYTHONPATH=ops/dashboard/src:ops/dashboard python3 -m pytest ops/dashboard/tests -q
  • 165 passed

GitHub CI for PR #462:

• test (3.11): pass, 1m23s
• test (3.12): pass, 1m31s
• test (3.13): pass, 2m40s

Live deploy/proof

Deployed to eeepc /opt health/emitter runtime:

• release: /opt/eeepc-agent/runtimes/self-evolving-agent/releases/20260502-issue461-2998ffa5
• SOURCE_COMMIT=2998ffa5047db482c39fae91b84d6799735a3761
• service: eeepc-self-evolving-agent-health.service
• result: status=0/SUCCESS
• journal: Self-evolving cycle PASS — goal=goal-bootstrap — evidence=evidence-8ce06aa6fe95

After /collect, latest /api/promotions.items[0] is coherent:

{
  "identity_key": "promotion-26cbaba1d30b",
  "status": "ready_for_policy_review",
  "decision_record": "pending_operator_review_packet",
  "accepted_record": null,
  "readiness_checks": {
    "artifact_present": true,
    "evidence_refs_present": true,
    "provenance_complete": true,
    "missing_inputs": []
  },
  "readiness_blocker": {
    "state": "ready_for_policy_review",
    "recommended_next_action": "ready_for_policy_review"
  },
  "recommended_next_action": "ready_for_policy_review",
  "governance_packet.review_packet_status": "pending_operator_review",
  "governance_packet.decision": "ready_for_policy_review",
  "source_commit": "2998ffa5047db482c39fae91b84d6799735a3761"
}

The #461 blocker is resolved: the latest complete-readiness candidate is no longer surfaced as not_ready_for_policy_review / blocked_not_ready.

A new downstream blocker is now honestly exposed by Mission Control: Blocked: accepted_record. I am tracking that as the next bounded issue rather than mixing it into #461.