Skip to content

Re-ask signed-source role mix-ups and resume classification-only rounds (#2035) - #2044

Merged
wwind123 merged 3 commits into
mainfrom
agent-loop/managed-2035
Nov 20, 2026
Merged

wwind123 merged 3 commits into
mainfrom
agent-loop/managed-2035

Conversation

@wwind123

Copy link
Copy Markdown
Owner

Summary

Fixes #2035

Two defects in signed-source classification, implemented per the approved plan.

1. A role mix-up gets re-asked instead of stopping the run

  • requirement_evidence.parse_associations now raises MissingJudgementFieldError(field_path="requirement_evidence") for role/field consistency violations, following the issue's split between judgement and shape errors:
    • human-only must name its request, and other roles must use "";
    • mixed and evidence-only need answered items, and code and human-only must use [];
    • only mixed carries leftover instructions.
  • The message names requirement_evidence[i], the requirement_id and the violated rule, for example "mixed requires at least one answered_item_id; if no prior evidence item of yours is answered, use code or human-only."
  • Shape errors still raise a plain AgentLoopError: not an array, wrong keys, unknown role, wrong types, empty or duplicate IDs, and a repeated source.
  • The classification invoke settings now live in one constant, CLASSIFICATION_INVOKE_OPTIONS, so tests drive the exact production settings. It opts into the typed judgement-exhaustion branch through a new reviewer_judgement_turn flag.
  • I did not set repair_expected_kind="pr_review" on that invoke, because doing so would also have switched Antigravity classification turns to structured output. The plan allows this fallback.
  • When the re-ask budget runs out, the failure reads "repair skipped because the reviewer did not supply requirement_evidence". It now carries a reviewer-judgement category instead of schema-validation, and the same label applies to the existing PR/plan judgement-exhaustion path.

2. Rerunning a round where only classification failed

  • response_validation.classification_resume_kind admits a same-round, same-head complete code review that the ordinary resume predicate rejects. Admission requires an open round, an unchanged signed requirement set (exact digest-identity equality) and matching plan and acquisition identities:
    • pending: a publication record whose associations do not cover every current source;
    • complete: a record of any phase whose validated associations cover every current source, so a settlement written by an earlier successful classification stays resumable.
  • admit_classification_resumes merges admitted records into resumed_by_name. Reconciled rounds and seat-binding mismatches are never admitted.
  • classify_review gains two arguments:
    • force_continuation (pending records only) runs the classification continuation even when no existing trigger fires;
    • pinned_review (both kinds) applies to every continuation that runs. A changed state or blocking findings raises HumanDecisionRequiredError before coder dispatch. Otherwise only requirement_evidence and the evidence requests are taken.
  • When a posted review still cannot be admitted, the existing partial-round refusal now names the failing check: changed signed requirement set (added/withdrawn), reconciled round, plan, acquisition or seat-binding change, or legacy identities.
  • A test confirms the Run B diagnosis: the unacknowledged publication fails only the acknowledgement predicate, and with admission disabled the rerun hits the deletion refusal.

Behaviour change to note

A single-reviewer round whose unacknowledged posted review already classifies every source now resumes as classification-complete instead of re-running the reviewer. tests/test_pr_loop_agent_turns.py was updated to pin this; the resumed carrier still keeps its metadata assessment and degradation record.

Docs

  • docs/architecture/round-state-and-resume.md, plus a decision-log entry.
  • The partial-round recovery section of docs/local_agent_loop.md.
  • tests/fixtures/plan_execution_patch_inventory.json gains two rows for the new tests' repair patches.
  • tests/.test_durations is untouched.

Tests

  • New tests in tests/test_requirement_evidence.py, tests/test_validated_agent_recovery.py, tests/test_pr_loop_resume.py and tests/test_partial_round_recovery.py.
  • The full suite ran once through the managed wrapper: 4 failures. Two were inventory/layering guards and two pinned the old re-run behaviour. All four are fixed, and the affected files plus the focused set pass at the current tree.

🤖 Generated with Claude Code

AGENT_MANAGED_CI_UNPROTECTED_OVERRIDE_V1 nonce=HQTdXMYIrSdzpN3Ssa8WNmAW

…ds (#2035)

Role/field consistency violations in requirement_evidence associations are
reviewer judgements: parse_associations now raises MissingJudgementFieldError
naming the entry index, requirement_id and violated rule, so the
classification continuation gets the bounded judgement re-ask and, when it is
exhausted, a deterministic reviewer-judgement failure instead of a
schema-validation stop. Shape errors stay plain AgentLoopError.

A published review whose only gap is the signed-source classification step is
admitted on rerun as classification-pending (publication record, sources not
covered) or classification-complete (validated associations cover every
source) in an unreconciled round with an unchanged signed requirement set.
Only the classification continuation reruns, pinned to the published verdict
and findings; a change stops for a human decision before coder dispatch.
Other rejections keep the partial-round refusal, which now names the failing
check.

Fixes #2035

Agent-Issue-Provenance: v1 repo=wwind123/coding-review-agent-loop issue=2035 flow=approved plan=8523c814220e10db
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@wwind123 wwind123 added the agent-loop-managed agent-loop managed CI label Nov 20, 2026
@wwind123

Copy link
Copy Markdown
Owner Author

Agent-loop managed-CI authorization record (machine-readable). Binds issue #2035 to pull request #2044 at head 2f088b3. Not an agent response; keep this comment.

@wwind123

Copy link
Copy Markdown
Owner Author

Expected closing issues for PR #2044: #2035.

Origin flow: approved-plan-implementation
Contract hash: 9dad0464d4e9a245382c654dbf0e2b96bab0210512da274b6b1e53296860f37e

-- coding-review-agent-loop

@wwind123

Copy link
Copy Markdown
Owner Author

Issue implementation

Implemented the approved plan for #2035 and opened draft PR #2044. (1) parse_associations raises MissingJudgementFieldError naming the entry index, requirement_id and rule for role/field consistency violations; shape errors stay plain AgentLoopError. The classification invoke (shared CLASSIFICATION_INVOKE_OPTIONS) re-asks once, and on exhaustion takes the typed judgement path, now labelled reviewer-judgement instead of schema-validation. It uses a reviewer_judgement_turn flag instead of repair_expected_kind so Antigravity classification turns do not switch to structured output. (2) A published review whose only gap is classification is admitted on rerun as classification-pending or classification-complete in an unreconciled round with an unchanged signed requirement set. Only the pinned classification continuation reruns, and a changed verdict or finding stops for a human decision before coder dispatch. Other rejections keep the partial-round refusal with a named cause. The plan asked for the CI suite, which justified the one full-suite run (tests/ -q): 4 failures out of 25857. Two were layering/patch-inventory guards and two pinned the old behaviour, where an unacknowledged but fully classified single-reviewer record was re-run; it now resumes as classification-complete. All four were fixed, and the affected files plus the focused set (requirement_evidence, validated_agent_recovery, pr_loop_resume, partial_round_recovery, judgement re-ask, failure classification, shape audit, file size, docs and architecture guards, review_parallel/rounds, closed round, human requirements, response_validation, checkpoint resume) pass at the final tree. Docs updated: round-state-and-resume.md, its decision log and the partial-round recovery section of local_agent_loop.md.

Result

Pull request reported: #2044.

Tests run

  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_partial_round_recovery.py tests/test_pr_loop_resume.py tests/test_validated_agent_recovery.py tests/test_shape_check_audit.py tests/test_source_file_size.py (agent-loop instrumented; whole-command timeout 1800s)
  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest tests/ -q (agent-loop instrumented; whole-command timeout 1800s)
  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider -n 0 tests/test_pr_loop_agent_turns.py tests/test_module_split_imports.py tests/test_orchestrator_split.py (agent-loop instrumented; whole-command timeout 1800s)
  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py (agent-loop instrumented; whole-command timeout 1800s)

Test observation receipts

Risk-based mode and transition test matrix evidence

  • Matrix identity: 4804dbb9c0b3de536467bbc8fbfca8fc6c641f4769f239bb667212a483568e17
Full matrix evidence (11 rows)
  • closed-round-gate — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_closed_round_admits_no_classification_resume
    • Locations: tests/test_pr_loop_resume.py
    • Workflow path: Real run_pr_loop: a round reconciled with an unacknowledged complete-association publication, truncated right after reconciliation, then rerun.
    • Expected outcome assertions: Every live admission call for round 1 sees reconciled=True and admits no kind.; The same record would be complete outside the closed round.
    • Forbidden-effect assertions: No classification continuation prompt runs.; No new round-1 reviewer record is posted.; No gh DELETE.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py [fe903a2e13cd43ee942abb885f66be7e; current-result]
  • consistency-rules-typed — verified
    • Tests: tests/test_requirement_evidence.py::test_consistency_rules_raise_typed_judgement_error, tests/test_requirement_evidence.py::test_consistency_rule_order_reports_first_violation
    • Locations: tests/test_requirement_evidence.py
    • Workflow path: parse_associations direct and through parse_structured_pr_review.
    • Expected outcome assertions: Each of the six consistency violations raises MissingJudgementFieldError with field_path requirement_evidence.; The message names requirement_evidence[1], the requirement_id and the rule; allowed_values carries the rule.
    • Forbidden-effect assertions: Rule order is unchanged: the human-only request rule is reported before the answers rule.; A valid preceding entry is still accepted.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py [fe903a2e13cd43ee942abb885f66be7e; current-result]
  • reask-exhausted — verified
    • Tests: tests/test_validated_agent_recovery.py::test_classification_reask_exhausted_takes_judgement_failure_path
    • Locations: tests/test_validated_agent_recovery.py
    • Workflow path: Signed-source classification continuation with exactly the kwargs pr_loop passes (CLASSIFICATION_INVOKE_OPTIONS), no extra test kwargs.
    • Expected outcome assertions: The failure text contains 'repair skipped because the reviewer did not supply requirement_evidence'.; failure_category is deterministic.
    • Forbidden-effect assertions: The message does not contain schema-validation.; Only two codex invocations happen (no third).; Repair is never called.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py [fe903a2e13cd43ee942abb885f66be7e; current-result]
    • Caveats: The invoke opts in via reviewer_judgement_turn instead of repair_expected_kind pr_review, the plan's allowed fallback, to avoid enabling Antigravity structured output on classification turns.
  • reask-mixed-empty — verified
    • Tests: tests/test_validated_agent_recovery.py::test_classification_mixed_without_answers_is_reasked_then_accepted
    • Locations: tests/test_validated_agent_recovery.py
    • Workflow path: Signed-source classification continuation through _run_validated_agent with the production CLASSIFICATION_INVOKE_OPTIONS and the real PR review validator.
    • Expected outcome assertions: A mixed association with empty answered_item_ids gets exactly one judgement re-ask whose prompt names requirement_evidence[0], the requirement_id and the mixed rule.; The corrected code classification is accepted on the second turn.
    • Forbidden-effect assertions: The structured repair pass is never called.; No schema-validation abort occurs.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py [fe903a2e13cd43ee942abb885f66be7e; current-result]
  • resume-classification-only — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_rerun_redoes_only_the_missing_classification, tests/test_pr_loop_resume.py::test_2035_diagnosis_unacknowledged_publication_is_the_rejecting_predicate
    • Locations: tests/test_pr_loop_resume.py
    • Workflow path: Real run_pr_loop with FakeRunner: Run A publishes both reviews and Codex's classification exhausts its re-ask; the rerun resumes the round.
    • Expected outcome assertions: The rerun invokes Codex once with the pinned classification continuation prompt and never invokes Gemini.; The run completes with outcome 0 and Codex's settlement carries the code classification with the published verdict.; With admission disabled the same rerun raises the deletion-list PartialReviewRoundError, pinning the rejecting predicate.
    • Forbidden-effect assertions: No gh DELETE command is issued.; Codex's publication record is not duplicated.; No fresh review turn runs for either reviewer.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py [fe903a2e13cd43ee942abb885f66be7e; current-result]
    • Caveats: Exercised on a review_parallel spool round; the sequential-path refusal shares the same resumed_by_name admission.
  • resume-complete-conflict-pinned — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_complete_reviewer_peer_conflict_continuation_is_pinned, tests/test_requirement_evidence.py::test_pinned_conflict_continuation, tests/test_requirement_evidence.py::test_pinned_complete_associations_without_conflict_get_no_continuation
    • Locations: tests/test_pr_loop_resume.py, tests/test_requirement_evidence.py
    • Workflow path: PR loop rerun: Codex is pending and classifies code; Gemini's unacknowledged blocking publication with complete associations is admitted complete and meets the role conflict.
    • Expected outcome assertions: A conflict continuation that changes the verdict raises HumanDecisionRequiredError naming Gemini.; A continuation keeping the verdict and findings is accepted and the settlement records role code with state blocking.; The continuation prompt asks to reconcile the role.
    • Forbidden-effect assertions: No coder invocation on the refused path.; A pinned complete record without a conflict gets no continuation.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py [fe903a2e13cd43ee942abb885f66be7e; current-result]
    • Caveats: The complete record uses the human-only role with its emitted request instead of evidence-only, which avoids fabricating native evidence items; the conflict path is the same.
  • resume-other-mismatch — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_other_mismatch_is_never_admitted_and_names_its_cause, tests/test_pr_loop_resume.py::test_2035_refusal_cause_names_the_rejecting_reason
    • Locations: tests/test_pr_loop_resume.py
    • Workflow path: classification_resume_kind and admit_classification_resumes on a real Run A publication record with altered identities.
    • Expected outcome assertions: Plan, acquisition, seat, legacy-identity, round and head mismatches are never admitted.; The rejection reason names the plan, acquisition, seat or legacy cause, and the refusal cause text includes it.
    • Forbidden-effect assertions: resumed_by_name is not changed for a mismatched record.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py [fe903a2e13cd43ee942abb885f66be7e; current-result]
  • resume-repeated-classification-failure — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_repeated_classification_failure_keeps_classified_peer_resumable
    • Locations: tests/test_pr_loop_resume.py
    • Workflow path: Real run_pr_loop: Run A fails Codex classification; rerun 1 classifies Codex without acknowledgement and fails Gemini; rerun 2 resumes.
    • Expected outcome assertions: Codex's settlement record is classification-complete.; Rerun 2 never invokes Codex and invokes Gemini once with the classification continuation.
    • Forbidden-effect assertions: No partial-round refusal and no gh DELETE.; No Codex record ever gains an acknowledgement it did not give.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py [fe903a2e13cd43ee942abb885f66be7e; current-result]
  • resume-requirements-changed — verified
    • Tests: tests/test_partial_round_recovery.py::test_2035_changed_requirement_set_refuses_with_named_cause
    • Locations: tests/test_partial_round_recovery.py
    • Workflow path: Real run_pr_loop rerun after a signed requirement was added, edited or withdrawn following publication.
    • Expected outcome assertions: PartialReviewRoundError names Codex's rejected own review and 'the signed requirement set changed since publication' with the added or withdrawn IDs.
    • Forbidden-effect assertions: No codex or gemini turn runs in the rerun.; No gh DELETE is issued by the loop.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py [fe903a2e13cd43ee942abb885f66be7e; current-result]
  • resume-verdict-pinned — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_classification_redo_cannot_change_published_verdict, tests/test_requirement_evidence.py::test_pinned_continuation_cannot_change_verdict_or_findings, tests/test_requirement_evidence.py::test_force_continuation_with_pin_keeps_published_verdict
    • Locations: tests/test_pr_loop_resume.py, tests/test_requirement_evidence.py
    • Workflow path: PR loop rerun of a classification-pending blocking publication, plus classify_review unit cases.
    • Expected outcome assertions: A continuation returning approved or different blocking items raises HumanDecisionRequiredError naming Codex.; A pinned continuation that keeps the verdict keeps the published state, findings and summary and takes only requirement_evidence.
    • Forbidden-effect assertions: No claude (coder) invocation happens.; No gh DELETE.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py [fe903a2e13cd43ee942abb885f66be7e; current-result]
  • shape-stays-fatal — verified
    • Tests: tests/test_requirement_evidence.py::test_shape_errors_stay_plain_agent_loop_error, tests/test_requirement_evidence.py::test_decode_associations_maps_consistency_errors_to_value_error
    • Locations: tests/test_requirement_evidence.py
    • Workflow path: parse_associations direct and decode_associations transport decode.
    • Expected outcome assertions: Non-array, wrong keys, unknown role, wrong types, duplicate answers, empty id and repeated source raise plain AgentLoopError.; decode_associations maps shape and consistency failures to ValueError.
    • Forbidden-effect assertions: Shape errors are not MissingJudgementFieldError, so they gain no re-ask.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_judgement_field_reask.py tests/test_agent_failure_classification.py tests/test_shape_check_audit.py tests/test_source_file_size.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py tests/test_review_parallel.py tests/test_review_rounds.py tests/test_pr_loop_closed_round.py tests/test_pr_loop_human_requirements.py tests/test_response_validation.py tests/test_checkpoint_round_resume.py [fe903a2e13cd43ee942abb885f66be7e; current-result]

Risk-matrix coverage map

Orchestrator completeness check (existence and level only; whether each cited test really exercises its row is for the reviewer to judge).

  • reask-mixed-empty: tests tests/test_validated_agent_recovery.py::test_classification_mixed_without_answers_is_reasked_then_accepted, tests/test_validated_agent_recovery.py; declared level unit vs required unit
  • reask-exhausted: tests tests/test_validated_agent_recovery.py::test_classification_reask_exhausted_takes_judgement_failure_path, tests/test_validated_agent_recovery.py; declared level unit vs required unit
  • consistency-rules-typed: tests tests/test_requirement_evidence.py::test_consistency_rules_raise_typed_judgement_error, tests/test_requirement_evidence.py::test_consistency_rule_order_reports_first_violation, tests/test_requirement_evidence.py; declared level unit vs required unit
  • shape-stays-fatal: tests tests/test_requirement_evidence.py::test_shape_errors_stay_plain_agent_loop_error, tests/test_requirement_evidence.py::test_decode_associations_maps_consistency_errors_to_value_error, tests/test_requirement_evidence.py; declared level unit vs required unit
  • resume-classification-only: tests tests/test_pr_loop_resume.py::test_2035_rerun_redoes_only_the_missing_classification, tests/test_pr_loop_resume.py::test_2035_diagnosis_unacknowledged_publication_is_the_rejecting_predicate, tests/test_pr_loop_resume.py; declared level integration vs required integration
  • resume-repeated-classification-failure: tests tests/test_pr_loop_resume.py::test_2035_repeated_classification_failure_keeps_classified_peer_resumable, tests/test_pr_loop_resume.py; declared level integration vs required integration
  • resume-verdict-pinned: tests tests/test_pr_loop_resume.py::test_2035_classification_redo_cannot_change_published_verdict, tests/test_requirement_evidence.py::test_pinned_continuation_cannot_change_verdict_or_findings, tests/test_requirement_evidence.py::test_force_continuation_with_pin_keeps_published_verdict, tests/test_pr_loop_resume.py, tests/test_requirement_evidence.py; declared level integration vs required integration
  • resume-complete-conflict-pinned: tests tests/test_pr_loop_resume.py::test_2035_complete_reviewer_peer_conflict_continuation_is_pinned, tests/test_requirement_evidence.py::test_pinned_conflict_continuation, tests/test_requirement_evidence.py::test_pinned_complete_associations_without_conflict_get_no_continuation, tests/test_pr_loop_resume.py, tests/test_requirement_evidence.py; declared level integration vs required integration
  • resume-requirements-changed: tests tests/test_partial_round_recovery.py::test_2035_changed_requirement_set_refuses_with_named_cause, tests/test_partial_round_recovery.py; declared level integration vs required integration
  • resume-other-mismatch: tests tests/test_pr_loop_resume.py::test_2035_other_mismatch_is_never_admitted_and_names_its_cause, tests/test_pr_loop_resume.py::test_2035_refusal_cause_names_the_rejecting_reason, tests/test_pr_loop_resume.py; declared level unit vs required unit
  • closed-round-gate: tests tests/test_pr_loop_resume.py::test_2035_closed_round_admits_no_classification_resume, tests/test_pr_loop_resume.py; declared level integration vs required integration
  • Status: complete at 2f088b3 (deterministic completeness check passed)

-- Anthropic Claude: claude-opus-5-5 (medium)

@wwind123

Copy link
Copy Markdown
Owner Author

Agent-loop managed-CI resume provenance audit record (machine-readable). Records resume provenance for managed CI on pull request #2044 at head 2f088b3. Not an agent response; keep this comment.

AGENT_MANAGED_CI_UNPROTECTED_OVERRIDE_V1 nonce=n9JOLlLrZ92J1PefLpu0zRsn_2_3zLvs repo=wwind123/coding-review-agent-loop base=main head=2f088b38917f3efd9fe73e0c8fabb31e677eaf80 protection=voluntary active_label_event_id=36779423960 resume_from=7042389412 provenance_head=2f088b38917f3efd9fe73e0c8fabb31e677eaf80 generation=b7ffZ-6XXDoxwFKB

Resume provenance only: the prior issue-created audit is not an authorization token.

@wwind123

Copy link
Copy Markdown
Owner Author

PR review scheduling audit: selected Codex; paused Claude, Antigravity; reason: strict pre-panel fallback: architecture identity changed; scheduler metadata recovery: latest scheduler metadata is missing or invalid; scheduler metadata recovery: current-head records lack scheduler metadata; scheduler metadata is missing or invalid; primary re-invoked with full context; phase: primary; head: 2f088b3; primary: Codex; active owners: (none); force-full: False (source: none); scheduler-policy calls avoided cumulatively: 2.

@wwind123

Copy link
Copy Markdown
Owner Author

Review verdict: Blocking

Reviewed the complete local base-to-head diff at 2f088b3. The judgement re-ask change is sound, but classification recovery loses acquisition identity and can fail to persist reconciled associations. Required workflow coverage also remains incomplete. GitHub checks are passing; no tests were executed during this read-only review.

Blocking issues

  • [P1] Preserve acquisition identity when posting a resumed classification settlement. The new continuation at src/coding_review_agent_loop/pr_loop.py:5782 causes a resumed publication to receive a settlement, but both posting calls at lines 6000 and 6032 pass identity=None for resumed records. _post_pr_reviewer_comment then omits provider, configured_model and configured_effort, whose round-metadata defaults are None. On another rerun, _resumed_pr_reviewer_identity_rejection requires those fields to match whenever scheduling or reviewer seats are enabled; provider must match even when no model was explicitly configured. Thus, if A successfully classifies and B subsequently fails, A's new settlement is rejected for an acquisition mismatch and the deletion-only refusal returns. Preserve the original independent review's acquisition identity in these settlements and add the repeated-failure regression under a scheduler-enabled policy and a seat configuration.
  • [P2] Persist successful conflict continuations on classification-complete settlement records. classification_resume_kind now admits complete associations from any phase, and pr_loop.py:5782 permits a pinned conflict continuation on those records. However, both settlement-posting branches still require resumed_record.metadata.phase == 'publication' (lines 5932 and 6013). For an authoritative settlement admitted as complete, a successful role reconciliation therefore updates only classified_review_carriers; discarding early_published_pr_reviewers does not reach either posting call. A later peer classification failure followed by rerun restores the old conflicting associations, and historical_projection also sees the stale role. Post the changed classification for admitted complete settlements while preserving their existing finding IDs and acquisition identity. Cover a complete authoritative settlement that reconciles a role, followed by a peer failure and another rerun.
  • [P2] Complete the workflow assertions claimed by the risk-matrix coverage map. The cited tests do not establish all required outcomes for resume-classification-only and resume-repeated-classification-failure.

-- OpenAI Codex: gpt-6.1-sol (medium)

@wwind123

Copy link
Copy Markdown
Owner Author

PR review round 1 reconciliation: settled reviewers: Codex. Finalization continues after reconciliation. Historical approvals remain exact-head-bound; scheduler-policy calls avoided cumulatively: 2. Phase: primary; force-full: False (source: none).

@wwind123

Copy link
Copy Markdown
Owner Author

PR #2044 coder dispatch record: round 1, attempt 1, head 2f088b38917f3efd9fe73e0c8fabb31e677eaf80. Written before the coder is invoked so an interrupted or rejected turn can be resumed.

…ion settlements (#2035)

A resumed review's settlement now carries the original review's acquisition
identity instead of none, so scheduler/seat acquisition checks still admit
it on a later rerun. An admitted authoritative settlement whose pinned
conflict continuation reconciles its role posts the new classification with
its existing finding IDs, so a later rerun restores the reconciled role.

Tests cover the round-2 #2310 shape in sequential and parallel reruns
(with and without a scheduler policy), pin the repeated-failure rerun's
final settlement, reconciliation and acknowledgement boundary, and the
persisted reconciled role across a further rerun.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@wwind123

Copy link
Copy Markdown
Owner Author

Coder follow-up

Fixed all three round-1 items. Resumed settlements now carry the original review's acquisition identity, and an admitted authoritative settlement whose role is reconciled posts its new classification with its existing finding IDs. The workflow tests now cover the round-2 #2310 shape in sequential and parallel reruns, with and without a scheduler, and the repeated-failure rerun's final outcome is pinned explicitly. Generalization: this generalizes the fix for [item-1] and [item-2]: every settlement posted for a review restored through classification-only resume carries the restored record's acquisition identity and existing finding IDs, whatever its phase. Test rationale: test_pr_loop_resume/test_partial_round_recovery/test_requirement_evidence/test_validated_agent_recovery cover the changed admission and settlement paths. pr_loop_agent_turns, human_requirements, review_parallel, closed_round, review_scheduling, checkpoint_round_resume, reviewer_boards, board_change, evidence_conversion and carried_review_context cover pr_loop's resumed-settlement posting. The size, module split and docs guards cover the edited pr_loop and docs.

Addressed items

  • item-1: Blocking issue from OpenAI Codex: gpt-6.1-sol (medium), round 1: [P1] Preserve acquisition identity when posting a resumed classification settlement. The new cont...
    • Resolution: Both settlement posts for resumed records pass the restored record's metadata as identity (pr_loop.py), so provider, configured_model, configured_effort and observation fields carry over. test_2035_repeated_classification_failure_keeps_classified_peer_resumable is parametrized over the default and selective-intermediate scheduler. It asserts the settlement's identity fields equal the publication's and that classification_resume_kind admits the settlement under an acquisition contract. That is the same check seat configurations use (reviewer_acquisition_contract applies when selective_policy or reviewer_seats), so seats are covered by that predicate rather than by a separate seat-board run.
  • item-2: Blocking issue from OpenAI Codex: gpt-6.1-sol (medium), round 1: [P2] Persist successful conflict continuations on classification-complete settlement records. cla...
    • Resolution: After a pinned continuation changes the classification of an admitted non-publication record, pr_loop posts its settlement with new_items=metadata.new_items (existing IDs) and the restored identity. test_2035_complete_authoritative_settlement_persists_reconciled_role reconciles a complete authoritative settlement from code to human-only, then simulates a peer failure and reruns again. The new settlement keeps the item IDs and identity, and the further rerun does not invoke Gemini. The test fails with the post disabled (verified locally).
  • item-3: Blocking issue from OpenAI Codex: gpt-6.1-sol (medium), round 1: [P2] Complete the workflow assertions claimed by the risk-matrix coverage-map (quoted). The cited tests do...
    • Resolution: s1: new test_2035_round_two_rerun_preserves_records_and_redoes_only_classification covers round 1 block, coder, then round 2 all-published with a blocking unacknowledged Codex review. Reruns are parametrized sequential/parallel x default/selective, and a spy confirms the sequential guard path. Assertions: only the classification continuation and then the coder run; existing comments are kept verbatim, including coder-dispatch and scheduler checkpoints; no round-2 coder-dispatch or publication is re-posted (only the ordinary per-invocation scheduler audit may be); published verdicts survive; the run completes without a managed-CI flag. s2: the repeated-failure test now requires the exact 'scripted agent output exhausted' coder stop, the gemini-then-claude invocation order, Gemini's code settlement, the round-1 reconciliation, the acknowledgement-repair-failed boundary and the round-2 coder dispatch.

Remaining items

  • None.

Sub-item progress

  • item-3: 0/2 sub-items resolved (reviewer-confirmed); coder claims 2 more addressed this round (unverified); coder reports the whole item addressed, which implies every open sub-item (unverified)
    • item-3.s1: open; coder claims addressed (unverified)
    • item-3.s2: open; coder claims addressed (unverified)

Tests run

  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py (agent-loop instrumented; whole-command timeout 1800s)
  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_review_scheduling.py tests/test_checkpoint_round_resume.py tests/test_pr_loop_reviewer_boards.py tests/test_board_change.py tests/test_pr_loop_evidence_conversion.py tests/test_carried_review_context.py (agent-loop instrumented; whole-command timeout 1800s)

Test observation receipts

Risk-based mode and transition test matrix evidence

  • Matrix identity: 4804dbb9c0b3de536467bbc8fbfca8fc6c641f4769f239bb667212a483568e17
11 changed rows; 0 unchanged since round 1
  • closed-round-gate — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_closed_round_admits_no_classification_resume
    • Locations: tests/test_pr_loop_resume.py
    • Workflow path: Real run_pr_loop rerun of a reconciled round holding an unacknowledged complete-association publication.
    • Expected outcome assertions: Every live admission call for round 1 sees reconciled=True and admits nothing.
    • Forbidden-effect assertions: No classification continuation.; No new round 1 reviewer record.; No gh DELETE.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [8eaedadf4e7b41259d3d431b805e4642; current-result]
  • consistency-rules-typed — verified
    • Tests: tests/test_requirement_evidence.py::test_consistency_rules_raise_typed_judgement_error, tests/test_requirement_evidence.py::test_consistency_rule_order_reports_first_violation
    • Locations: tests/test_requirement_evidence.py
    • Workflow path: parse_associations direct and through parse_structured_pr_review.
    • Expected outcome assertions: All six consistency violations raise MissingJudgementFieldError with field_path requirement_evidence naming index, requirement_id and rule.
    • Forbidden-effect assertions: Rule order and acceptance of valid entries are unchanged.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [8eaedadf4e7b41259d3d431b805e4642; current-result]
  • reask-exhausted — verified
    • Tests: tests/test_validated_agent_recovery.py::test_classification_reask_exhausted_takes_judgement_failure_path
    • Locations: tests/test_validated_agent_recovery.py
    • Workflow path: Classification continuation with exactly the production CLASSIFICATION_INVOKE_OPTIONS kwargs.
    • Expected outcome assertions: Failure text contains 'repair skipped because the reviewer did not supply requirement_evidence'.; failure_category is deterministic.
    • Forbidden-effect assertions: Message does not contain schema-validation.; Exactly two invocations.; Repair never called.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [8eaedadf4e7b41259d3d431b805e4642; current-result]
    • Caveats: Opts in through reviewer_judgement_turn rather than repair_expected_kind pr_review (the plan's allowed fallback) so Antigravity classification turns keep their output channel.
  • reask-mixed-empty — verified
    • Tests: tests/test_validated_agent_recovery.py::test_classification_mixed_without_answers_is_reasked_then_accepted
    • Locations: tests/test_validated_agent_recovery.py
    • Workflow path: Signed-source classification continuation through _run_validated_agent with the production CLASSIFICATION_INVOKE_OPTIONS and the real PR review validator.
    • Expected outcome assertions: Mixed with empty answered_item_ids gets exactly one judgement re-ask naming requirement_evidence[0], the requirement_id and the mixed rule.; The corrected classification is accepted.
    • Forbidden-effect assertions: Structured repair is never called.; No schema-validation abort.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [8eaedadf4e7b41259d3d431b805e4642; current-result]
  • resume-classification-only — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_round_two_rerun_preserves_records_and_redoes_only_classification, tests/test_pr_loop_resume.py::test_2035_rerun_redoes_only_the_missing_classification, tests/test_pr_loop_resume.py::test_2035_diagnosis_unacknowledged_publication_is_the_rejecting_predicate
    • Locations: tests/test_pr_loop_resume.py
    • Workflow path: Real run_pr_loop: round 1 blocks, the coder runs, round 2 publishes a blocking unacknowledged Codex review and Gemini's approval, Codex's classification exhausts its re-ask; rerun in sequential and parallel modes, with and without the selective-intermediate scheduler.
    • Expected outcome assertions: The rerun invokes only Codex's pinned classification continuation, then the coder for round 3.; The sequential partial-round guard runs in the sequential variant and not in the parallel one.; Codex's round 2 settlement keeps blocking state and its finding, with role code; Gemini's published approval is its round 2 verdict.; With admission disabled the same rerun raises the deletion-list refusal.
    • Forbidden-effect assertions: Every pre-existing comment, including the coder-dispatch record and scheduler checkpoints, is kept verbatim (prefix-equal).; No round 2 coder-dispatch or publication record is re-posted; only the ordinary per-invocation scheduler audit may be.; No gh DELETE and no fresh review turn.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [8eaedadf4e7b41259d3d431b805e4642; current-result]
    • Caveats: The FakeRunner PR has no managed-CI qualification checkpoint; checkpoint preservation is shown on the scheduler checkpoints and coder-dispatch record.
  • resume-complete-conflict-pinned — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_complete_reviewer_peer_conflict_continuation_is_pinned, tests/test_pr_loop_resume.py::test_2035_complete_authoritative_settlement_persists_reconciled_role, tests/test_requirement_evidence.py::test_pinned_conflict_continuation, tests/test_requirement_evidence.py::test_pinned_complete_associations_without_conflict_get_no_continuation
    • Locations: tests/test_pr_loop_resume.py, tests/test_requirement_evidence.py
    • Workflow path: PR loop reruns where a classification-complete publication or authoritative settlement meets an earlier reviewer's different role.
    • Expected outcome assertions: A conflict continuation changing the verdict raises HumanDecisionRequiredError naming Gemini.; A kept-verdict reconciliation is accepted and posted, including for an authoritative settlement, with its existing finding IDs and acquisition identity.; A further rerun restores the reconciled role, so Gemini is not invoked again.
    • Forbidden-effect assertions: No coder invocation on the refused path.; No continuation for a complete record without conflict.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [8eaedadf4e7b41259d3d431b805e4642; current-result]
    • Caveats: The complete record uses human-only with its emitted request instead of evidence-only; the authoritative case simulates a lost peer settlement by truncating history.
  • resume-other-mismatch — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_other_mismatch_is_never_admitted_and_names_its_cause, tests/test_pr_loop_resume.py::test_2035_refusal_cause_names_the_rejecting_reason
    • Locations: tests/test_pr_loop_resume.py
    • Workflow path: classification_resume_kind and admit_classification_resumes on a real publication with altered identities.
    • Expected outcome assertions: Plan, acquisition, seat, legacy, round and head mismatches are never admitted.; The reason names the cause and the refusal includes it.
    • Forbidden-effect assertions: resumed_by_name is unchanged for mismatched records.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [8eaedadf4e7b41259d3d431b805e4642; current-result]
  • resume-repeated-classification-failure — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_repeated_classification_failure_keeps_classified_peer_resumable
    • Locations: tests/test_pr_loop_resume.py
    • Workflow path: Real run_pr_loop with and without the selective-intermediate scheduler: Run A fails Codex classification; rerun 1 settles Codex without acknowledgement and fails Gemini; rerun 2 resumes.
    • Expected outcome assertions: Codex's settlement keeps provider/model/effort/acquisition identity and is complete under an acquisition contract.; Rerun 2 invokes only Gemini's classification continuation, then the coder.; Gemini's settlement records role code; round 1 reconciliation, the acknowledgement-repair-failed boundary and the round 2 coder dispatch are posted.
    • Forbidden-effect assertions: No Codex record is posted in rerun 2 and none gains an acknowledgement.; No partial-round refusal and no gh DELETE.; The final stop is only the empty coder script.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [8eaedadf4e7b41259d3d431b805e4642; current-result]
  • resume-requirements-changed — verified
    • Tests: tests/test_partial_round_recovery.py::test_2035_changed_requirement_set_refuses_with_named_cause
    • Locations: tests/test_partial_round_recovery.py
    • Workflow path: Real run_pr_loop rerun after a signed requirement was added, edited or withdrawn.
    • Expected outcome assertions: The refusal names Codex's rejected review and the changed signed requirement set with added/withdrawn IDs.
    • Forbidden-effect assertions: No reviewer turn in the rerun.; No gh DELETE.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [8eaedadf4e7b41259d3d431b805e4642; current-result]
  • resume-verdict-pinned — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_classification_redo_cannot_change_published_verdict, tests/test_requirement_evidence.py::test_pinned_continuation_cannot_change_verdict_or_findings, tests/test_requirement_evidence.py::test_force_continuation_with_pin_keeps_published_verdict
    • Locations: tests/test_pr_loop_resume.py, tests/test_requirement_evidence.py
    • Workflow path: PR loop rerun of a classification-pending blocking publication, plus classify_review unit cases.
    • Expected outcome assertions: A changed verdict or findings raises HumanDecisionRequiredError naming Codex.; A kept verdict keeps state, findings and summary.
    • Forbidden-effect assertions: No coder invocation.; No gh DELETE.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [8eaedadf4e7b41259d3d431b805e4642; current-result]
  • shape-stays-fatal — verified
    • Tests: tests/test_requirement_evidence.py::test_shape_errors_stay_plain_agent_loop_error, tests/test_requirement_evidence.py::test_decode_associations_maps_consistency_errors_to_value_error
    • Locations: tests/test_requirement_evidence.py
    • Workflow path: parse_associations direct and decode_associations transport decode.
    • Expected outcome assertions: Shape violations raise plain AgentLoopError.; decode_associations maps failures to ValueError.
    • Forbidden-effect assertions: Shape errors are never MissingJudgementFieldError.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_agent_turns.py tests/test_pr_loop_human_requirements.py tests/test_review_parallel.py tests/test_pr_loop_closed_round.py tests/test_source_file_size.py tests/test_module_split_imports.py tests/test_orchestrator_split.py tests/test_docs_guidance.py tests/test_architecture_set_validation.py [8eaedadf4e7b41259d3d431b805e4642; current-result]

0 rows unchanged since round 1; full matrix in round 1.

Risk-matrix coverage map

Orchestrator completeness check (existence and level only; whether each cited test really exercises its row is for the reviewer to judge).

  • reask-mixed-empty: tests tests/test_validated_agent_recovery.py::test_classification_mixed_without_answers_is_reasked_then_accepted, tests/test_validated_agent_recovery.py; declared level unit vs required unit
  • reask-exhausted: tests tests/test_validated_agent_recovery.py::test_classification_reask_exhausted_takes_judgement_failure_path, tests/test_validated_agent_recovery.py; declared level unit vs required unit
  • consistency-rules-typed: tests tests/test_requirement_evidence.py::test_consistency_rules_raise_typed_judgement_error, tests/test_requirement_evidence.py::test_consistency_rule_order_reports_first_violation, tests/test_requirement_evidence.py; declared level unit vs required unit
  • shape-stays-fatal: tests tests/test_requirement_evidence.py::test_shape_errors_stay_plain_agent_loop_error, tests/test_requirement_evidence.py::test_decode_associations_maps_consistency_errors_to_value_error, tests/test_requirement_evidence.py; declared level unit vs required unit
  • resume-classification-only: tests tests/test_pr_loop_resume.py::test_2035_round_two_rerun_preserves_records_and_redoes_only_classification, tests/test_pr_loop_resume.py::test_2035_rerun_redoes_only_the_missing_classification, tests/test_pr_loop_resume.py::test_2035_diagnosis_unacknowledged_publication_is_the_rejecting_predicate, tests/test_pr_loop_resume.py; declared level integration vs required integration
  • resume-repeated-classification-failure: tests tests/test_pr_loop_resume.py::test_2035_repeated_classification_failure_keeps_classified_peer_resumable, tests/test_pr_loop_resume.py; declared level integration vs required integration
  • resume-verdict-pinned: tests tests/test_pr_loop_resume.py::test_2035_classification_redo_cannot_change_published_verdict, tests/test_requirement_evidence.py::test_pinned_continuation_cannot_change_verdict_or_findings, tests/test_requirement_evidence.py::test_force_continuation_with_pin_keeps_published_verdict, tests/test_pr_loop_resume.py, tests/test_requirement_evidence.py; declared level integration vs required integration
  • resume-complete-conflict-pinned: tests tests/test_pr_loop_resume.py::test_2035_complete_reviewer_peer_conflict_continuation_is_pinned, tests/test_pr_loop_resume.py::test_2035_complete_authoritative_settlement_persists_reconciled_role, tests/test_requirement_evidence.py::test_pinned_conflict_continuation, tests/test_requirement_evidence.py::test_pinned_complete_associations_without_conflict_get_no_continuation, tests/test_pr_loop_resume.py, tests/test_requirement_evidence.py; declared level integration vs required integration
  • resume-requirements-changed: tests tests/test_partial_round_recovery.py::test_2035_changed_requirement_set_refuses_with_named_cause, tests/test_partial_round_recovery.py; declared level integration vs required integration
  • resume-other-mismatch: tests tests/test_pr_loop_resume.py::test_2035_other_mismatch_is_never_admitted_and_names_its_cause, tests/test_pr_loop_resume.py::test_2035_refusal_cause_names_the_rejecting_reason, tests/test_pr_loop_resume.py; declared level unit vs required unit
  • closed-round-gate: tests tests/test_pr_loop_resume.py::test_2035_closed_round_admits_no_classification_resume, tests/test_pr_loop_resume.py; declared level integration vs required integration
  • Status: complete at 505628d (deterministic completeness check passed)

-- Anthropic Claude: claude-opus-5-5 (medium)

@wwind123

Copy link
Copy Markdown
Owner Author

Agent-loop managed-CI authorization continuity record (machine-readable). Binds issue #2035 to pull request #2044 at head 505628d. Not an agent response; keep this comment.

@wwind123

Copy link
Copy Markdown
Owner Author

PR review scheduling audit: selected Codex; paused Claude, Antigravity; reason: strict pre-panel fallback: diff path 'docs/architecture/round-state-and-resume.md' is outside obligation scopes; primary re-invoked with full context; phase: primary; head: 505628d; primary: Codex; active owners: Codex; force-full: False (source: none); scheduler-policy calls avoided cumulatively: 4.

@wwind123

Copy link
Copy Markdown
Owner Author

Review verdict: Blocking

Reviewed the complete base-to-head diff at 505628d. The acquisition-identity and classification-persistence defects are fixed. The risk-matrix coverage map still overstates managed-CI checkpoint coverage for resume-classification-only; the remaining carried test obligation needs completion. Other matrix rows have suitable source and test coverage. GitHub checks are passing; no tests were run during this read-only review.

Prior unresolved item dispositions

  • [item-1] RESOLVED
    • Original finding: Blocking issue from OpenAI Codex: gpt-6.1-sol (medium), round 1: [P1] Preserve acquisition identity when posting a resumed classification settlement. The new cont...
  • [item-2] RESOLVED
    • Original finding: Blocking issue from OpenAI Codex: gpt-6.1-sol (medium), round 1: [P2] Persist successful conflict continuations on classification-complete settlement records. cla...
  • [item-3] BLOCKING: The qualification-checkpoint portion of item-3.s1 remains untested. tests/test_pr_loop_resume.py:2018-2049 creates the recovery fixture without a managed-CI qualification checkpoint, and lines 2085-2096 only preserve/count records that happen to exist. Scheduler-checkpoint and coder-dispatch assertions therefore do not establish the approved resume-classification-only row's qualification-checkpoint preservation or absence of a fresh managed-CI authorization requirement. The coverage map explicitly acknowledges this caveat. Add a classification-failure/rerun case containing a real matching qualification checkpoint and managed-CI authorization state; assert the existing checkpoint remains valid and unduplicated and recovery proceeds without --managed-ci-fresh. Sequential blocking-review coverage is now present, and item-3.s2 now explicitly asserts settlement, reconciliation and a constrained terminal error.
    • Original finding: Blocking issue from OpenAI Codex: gpt-6.1-sol (medium), round 1: [P2] Complete the workflow assertions claimed by the risk-matrix coverage map. The cited tests do...
    • Sub-items (per this review only, not the round outcome; the confirmed outcome is posted after reconciliation): resolves s2, reopens s1 → 1/2

-- OpenAI Codex: gpt-6.1-sol (medium)

@wwind123

Copy link
Copy Markdown
Owner Author

PR review round 2 reconciliation: settled reviewers: Codex. Finalization continues after reconciliation. Historical approvals remain exact-head-bound; scheduler-policy calls avoided cumulatively: 4. Phase: primary; force-full: False (source: none).

@wwind123

Copy link
Copy Markdown
Owner Author

PR #2044 coder dispatch record: round 2, attempt 1, head 505628d435d739f1b59a3508cb87895ec12023c5. Written before the coder is invoked so an interrupted or rejected turn can be resumed.

…on checkpoint (#2035)

A qualifying managed-CI checkpoint whose requirement identity a new signed
source invalidates forces the round's reviews; Codex's classification then
fails. The rerun redoes only that classification, keeps the original
checkpoint verbatim, reattaches the same qualification attempt without a
fresh dispatch or managed-CI authorization, and records exactly the
checkpoint sequence an uninterrupted control round records.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@wwind123

Copy link
Copy Markdown
Owner Author

Coder follow-up

Completed item-3.s1's qualification-checkpoint obligation with a test against a real managed-CI qualification checkpoint, which is now in tests/test_pr_loop_resume.py. Production code is unchanged this round. Test rationale: test_pr_loop_resume holds the new and existing resume tests; test_partial_round_recovery, test_requirement_evidence and test_validated_agent_recovery cover the other matrix rows cited earlier; test_pr_loop_managed_ci covers the managed-CI paths the new test drives; test_module_split_imports covers the patch inventory guard.

Addressed items

  • item-3: Blocking issue from OpenAI Codex: gpt-6.1-sol (medium), round 1: [P2] Complete the workflow assertions claimed by the risk-matrix coverage-map (quoted). The cited tests do...
    • Resolution: s1: new test_2035_managed_qualification_checkpoint_survives_classification_resume seeds a real managed-CI qualifying checkpoint (attempt 123/1, matching acquisition digest) plus a signed source that invalidates its requirements digest and forces the round's reviews. Codex's classification then fails. On the rerun, only the classification continuation runs and the original checkpoint comment is kept verbatim. Qualification reattaches attempt 123/1 with no fresh dispatch (so no fresh managed-CI authorization) and merges the exact head. The qualification-checkpoint sequence equals that of an uninterrupted control run of the same scenario, so nothing is duplicated, and the last checkpoint binds the current requirement digest. Parallel publication order is completion order, so that assertion is order-independent. The test passes under xdist (receipt above). s2 was already resolved.

Remaining items

  • None.

Sub-item progress

  • item-3: 1/2 sub-items resolved (reviewer-confirmed); coder claims 1 more addressed this round (unverified); coder reports the whole item addressed, which implies every open sub-item (unverified)
    • item-3.s1: open; coder claims addressed (unverified)
    • item-3.s2: resolved (round 2)

Tests run

  • /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider -rf tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_managed_ci.py tests/test_module_split_imports.py (agent-loop instrumented; whole-command timeout 1800s)

Test observation receipts

Risk-based mode and transition test matrix evidence

  • Matrix identity: 4804dbb9c0b3de536467bbc8fbfca8fc6c641f4769f239bb667212a483568e17
11 changed rows; 0 unchanged since round 2
  • closed-round-gate — missing
    • Tests: none
    • Locations: none
    • Workflow path: PR loop resume of a reconciled round
    • Expected outcome assertions: Neither classification resume kind is admitted; the closed-round gate behaves as today; no classification continuation runs.
    • Forbidden-effect assertions: A classification continuation or settlement posted into a closed round.; Bypassing the closed-round turn check by marking the reviewer resumed.
    • Caveats: No semantic coverage claim was supplied for this enforceable row.
  • consistency-rules-typed — missing
    • Tests: none
    • Locations: none
    • Workflow path: parse_associations direct
    • Expected outcome assertions: MissingJudgementFieldError with field_path requirement_evidence and message naming the index, requirement_id and rule.
    • Forbidden-effect assertions: Rule order or accepted valid inputs change.
    • Caveats: No semantic coverage claim was supplied for this enforceable row.
  • reask-exhausted — missing
    • Tests: none
    • Locations: none
    • Workflow path: PR review / signed-source classification continuation, with the invoke kwargs pr_loop passes
    • Expected outcome assertions: Deterministic judgement-failure message ('repair skipped because the reviewer did not supply requirement_evidence').
    • Forbidden-effect assertions: Failure category is not reported as schema-validation.; No third invocation.; The test does not supply kwargs that production omits.
    • Caveats: No semantic coverage claim was supplied for this enforceable row.
  • reask-mixed-empty — missing
    • Tests: none
    • Locations: none
    • Workflow path: PR review / signed-source classification continuation
    • Expected outcome assertions: One judgement re-ask naming requirement_evidence[i], the requirement_id and the mixed rule; a corrected response is accepted and the round settles.
    • Forbidden-effect assertions: No schema-validation abort.; No format repair pass that edits requirement_evidence.
    • Caveats: No semantic coverage claim was supplied for this enforceable row.
  • resume-complete-conflict-pinned — missing
    • Tests: none
    • Locations: none
    • Workflow path: PR loop resume / classification-complete reviewer
    • Expected outcome assertions: HumanDecisionRequiredError naming the reviewer before coder dispatch. A continuation that keeps the verdict and findings is accepted for role reconciliation only.
    • Forbidden-effect assertions: The changed verdict or findings are accepted after peers are public.; A coder dispatch.; A forced continuation for a complete record with no conflict.
    • Caveats: No semantic coverage claim was supplied for this enforceable row.
  • resume-other-mismatch — missing
    • Tests: none
    • Locations: none
    • Workflow path: PR loop resume
    • Expected outcome assertions: No classification-pending admission; refusal behaviour unchanged except for the more specific cause text.
    • Forbidden-effect assertions: Admission through the new predicate.
    • Caveats: No semantic coverage claim was supplied for this enforceable row.
  • resume-repeated-classification-failure — missing
    • Tests: none
    • Locations: none
    • Workflow path: PR loop resume, second rerun
    • Expected outcome assertions: A resumes as classification-complete with no continuation; only B's classification is redone; the round finishes.
    • Forbidden-effect assertions: Partial-round deletion list.; A fresh independent review turn for A or B.; Asserting a requirement resolution A did not give.
    • Caveats: No semantic coverage claim was supplied for this enforceable row.
  • resume-requirements-changed — missing
    • Tests: none
    • Locations: none
    • Workflow path: PR loop resume
    • Expected outcome assertions: Existing partial-round refusal, with a cause naming the changed signed requirement set.
    • Forbidden-effect assertions: Classification-pending or classification-complete admission.; A fresh turn beside public peers.; A change to the ordinary subset coverage semantics for approvals.
    • Caveats: No semantic coverage claim was supplied for this enforceable row.
  • resume-verdict-pinned — missing
    • Tests: none
    • Locations: none
    • Workflow path: PR loop resume / classification-pending reviewer
    • Expected outcome assertions: HumanDecisionRequiredError naming the reviewer; no coder dispatch.
    • Forbidden-effect assertions: The changed verdict is accepted after peers are public.
    • Caveats: No semantic coverage claim was supplied for this enforceable row.
  • shape-stays-fatal — missing
    • Tests: none
    • Locations: none
    • Workflow path: parse_associations direct and transport decode
    • Expected outcome assertions: Plain AgentLoopError (not MissingJudgementFieldError); decode_associations still maps any failure to ValueError.
    • Forbidden-effect assertions: Shape errors gain a re-ask.; Historical invalid records change route.
    • Caveats: No semantic coverage claim was supplied for this enforceable row.
  • resume-classification-only — verified
    • Tests: tests/test_pr_loop_resume.py::test_2035_managed_qualification_checkpoint_survives_classification_resume, tests/test_pr_loop_resume.py::test_2035_round_two_rerun_preserves_records_and_redoes_only_classification, tests/test_pr_loop_resume.py::test_2035_rerun_redoes_only_the_missing_classification, tests/test_pr_loop_resume.py::test_2035_diagnosis_unacknowledged_publication_is_the_rejecting_predicate
    • Locations: tests/test_pr_loop_resume.py
    • Workflow path: Real run_pr_loop with managed CI: a qualifying qualification checkpoint (attempt 123/1) is invalidated by a new signed source, the round's reviews publish, and Codex's classification exhausts its re-ask. The rerun is compared against an uninterrupted control round. A second scenario covers round 1 blocking, a coder dispatch, then round 2 all-published, with sequential and parallel reruns, with and without the selective scheduler.
    • Expected outcome assertions: The rerun invokes only Codex's pinned classification continuation; published verdicts are kept and Codex's settlement follows.; The original qualification checkpoint comment stays verbatim and first.; Qualification reattaches attempt 123/1 with no fresh dispatch, so no fresh managed-CI authorization is needed, and the exact head merges.; The qualification-checkpoint record sequence equals the uninterrupted control round's, ending in a qualifying 123/1 checkpoint bound to the current requirement digest.; In the round-2 scenario the coder-dispatch record and scheduler checkpoints are kept verbatim, and the sequential guard path runs only for sequential reruns.
    • Forbidden-effect assertions: No gh DELETE and no deletion-list refusal.; No duplicated qualification-checkpoint, coder-dispatch or publication record relative to an uninterrupted round.; No fresh review turn for any reviewer.
    • Evidence citations: /home/wwind123/tools/coding-review-agent-loop/.venv/bin/python -m pytest -q -p no:cacheprovider -rf tests/test_pr_loop_resume.py tests/test_partial_round_recovery.py tests/test_requirement_evidence.py tests/test_validated_agent_recovery.py tests/test_pr_loop_managed_ci.py tests/test_module_split_imports.py [d227857cbf3d44bc933ea6830f39f893; current-result]

0 rows unchanged since round 2; full matrix in round 1.

Risk-matrix coverage map

Orchestrator completeness check (existence and level only; whether each cited test really exercises its row is for the reviewer to judge).

  • reask-mixed-empty: no test claim and no declared gap (required unit)
  • reask-exhausted: no test claim and no declared gap (required unit)
  • consistency-rules-typed: no test claim and no declared gap (required unit)
  • shape-stays-fatal: no test claim and no declared gap (required unit)
  • resume-classification-only: tests tests/test_pr_loop_resume.py::test_2035_managed_qualification_checkpoint_survives_classification_resume, tests/test_pr_loop_resume.py::test_2035_round_two_rerun_preserves_records_and_redoes_only_classification, tests/test_pr_loop_resume.py::test_2035_rerun_redoes_only_the_missing_classification, tests/test_pr_loop_resume.py::test_2035_diagnosis_unacknowledged_publication_is_the_rejecting_predicate, tests/test_pr_loop_resume.py; declared level integration vs required integration
  • resume-repeated-classification-failure: no test claim and no declared gap (required integration)
  • resume-verdict-pinned: no test claim and no declared gap (required integration)
  • resume-complete-conflict-pinned: no test claim and no declared gap (required integration)
  • resume-requirements-changed: no test claim and no declared gap (required integration)
  • resume-other-mismatch: no test claim and no declared gap (required unit)
  • closed-round-gate: no test claim and no declared gap (required integration)
  • Status: incomplete at 7c48c83: reask-mixed-empty: missing-row; reask-exhausted: missing-row; consistency-rules-typed: missing-row; shape-stays-fatal: missing-row; resume-repeated-classification-failure: missing-row; resume-verdict-pinned: missing-row; resume-complete-conflict-pinned: missing-row; resume-requirements-changed: missing-row; resume-other-mismatch: missing-row; closed-round-gate: missing-row

-- Anthropic Claude: claude-opus-5-5 (medium)

@wwind123

Copy link
Copy Markdown
Owner Author

Agent-loop managed-CI authorization continuity record (machine-readable). Binds issue #2035 to pull request #2044 at head 7c48c83. Not an agent response; keep this comment.

@wwind123

Copy link
Copy Markdown
Owner Author

PR review scheduling audit: selected Codex; paused Claude, Antigravity; reason: primary phase: primary rechecks its findings before the secondary panel; phase: primary; head: 7c48c83; primary: Codex; active owners: Codex; force-full: False (source: none); scheduler-policy calls avoided cumulatively: 6.

@wwind123

Copy link
Copy Markdown
Owner Author

Review verdict: Approved

Reviewed the complete base-to-head diff at 7c48c83 and mapped all eleven risk-matrix rows to the source and tests. The new managed-CI recovery test closes the remaining checkpoint coverage gap; sequential blocking-review recovery and repeated-failure settlement assertions are also present. No merge-blocking defects found. GitHub checks are passing; no tests were run during this read-only review.

Prior unresolved item dispositions

  • [item-3] RESOLVED: The managed qualification recovery test seeds a qualification checkpoint and attached attempt, asserts preservation of the original comment, no fresh dispatch, reuse of attempt 123/1, exact-head merge, and checkpoint-sequence equality with an uninterrupted control. The round-two test covers sequential and parallel blocking recovery and existing dispatch records; the repeated-failure test explicitly checks settlement, reconciliation and the constrained later stop.
    • Original finding: Blocking issue from OpenAI Codex: gpt-6.1-sol (medium), round 1: [P2] Complete the workflow assertions claimed by the risk-matrix coverage map. The cited tests do...
    • Sub-items (per this review only, not the round outcome; the confirmed outcome is posted after reconciliation): resolves s1, s2 → 2/2

-- OpenAI Codex: gpt-6.1-sol (medium)

@wwind123

Copy link
Copy Markdown
Owner Author

PR review round 3 reconciliation: settled reviewers: Codex. Finalization continues after reconciliation. Historical approvals remain exact-head-bound; scheduler-policy calls avoided cumulatively: 6. Phase: primary; force-full: False (source: none).

@wwind123

Copy link
Copy Markdown
Owner Author

Sub-item progress — review round 3

  • item-3: 2/2 — cleared (all sub-items resolved)

-- Orchestrator

@wwind123

Copy link
Copy Markdown
Owner Author

PR review scheduling audit: selected Claude, Antigravity; paused Codex; reason: independent secondary audit after exact-head primary approval; phase: secondary-audit; head: 7c48c83; primary: Codex; active owners: (none); force-full: False (source: none); scheduler-policy calls avoided cumulatively: 6.

@wwind123

Copy link
Copy Markdown
Owner Author

Review verdict: Approved

Independent secondary audit of the full base-to-head diff at 7c48c83. Item 1: parse_associations now raises MissingJudgementFieldError(field_path=requirement_evidence) for the three role/field consistency checks, in the same order as before. Each message names the index, requirement_id and rule. The empty requirement_id check and the other shape checks stay plain AgentLoopError. The classification invoke uses CLASSIFICATION_INVOKE_OPTIONS with an explicit reviewer_judgement_turn flag instead of repair_expected_kind. The plan allows this alternative, and it keeps format repair off. An exhausted re-ask takes the deterministic judgement branch with the new reviewer-judgement classification text, not schema-validation. should_retry stays false for that public validation failure, so there is no third invocation. Item 2: admit_classification_resumes runs right after the ordinary resumed_by_name filter. It admits 'pending' (publication phase, incomplete associations) and 'complete' (validated associations cover every source, any phase) only in an unreconciled round with exact digest-identity equality, matching seat, plan and acquisition, and the same round and head. classify_review pins state and normalized blocking_items on every continuation for admitted reviewers, and forces the continuation only for pending records. The settlement paths now pass resumed_record.metadata as identity, so reposted settlements keep acquisition identity. An admitted authoritative settlement whose role was reconciled is reposted with its existing new_items. The blocking/approved branches post only for publication-phase records, so nothing is posted twice. Refusal causes are threaded through all three partial-round refusal seams. Tests exist at the reviewed head for every matrix row: re-ask and exhaustion through the production kwargs, typed consistency rules, shape errors staying fatal, the Run B rerun, repeated failure, the pending and complete pins, changed requirement sets, other mismatches, the closed round, and the managed-CI checkpoint survival. The latest coder coverage map lists only one row, but the earlier coder comment on this head lineage mapped all eleven rows to these same tests. They sit in files covered by the current-head passing receipt. GitHub checks are passing. Docs and the decision log are updated.

Future follow-ups

  • The pinned-verdict check compares whitespace-normalized blocking_items text exactly, so a reviewer that faithfully rephrases a finding in its classification continuation will stop the run for a human decision. Consider giving the continuation the exact published item texts to echo, or comparing finding identity rather than wording, if this shows up in practice.

-- Anthropic Claude: claude-opus-5-5 (medium)

@wwind123

Copy link
Copy Markdown
Owner Author

Review verdict: Approved

Approve pull request #2044 on independent secondary audit. Signed-source role and field consistency checks in parse_associations now correctly raise MissingJudgementFieldError to trigger the bounded reviewer-judgement re-ask, with shape errors remaining plain AgentLoopError. Reruns of unreconciled rounds with unchanged signed requirements admit published reviews with missing or unacknowledged classifications as classification-pending or classification-complete without requiring review deletion, pinning published verdicts and findings during classification continuations. The implementation, architecture docs, and risk matrix test coverage are verified clean.

-- Google Antigravity: Gemini 3.8 Flash (High)

@wwind123

Copy link
Copy Markdown
Owner Author

PR review round 4 reconciliation: settled reviewers: Antigravity, Claude. Finalization continues after reconciliation. Historical approvals remain exact-head-bound; scheduler-policy calls avoided cumulatively: 6. Phase: secondary-audit; force-full: False (source: none).

@wwind123

Copy link
Copy Markdown
Owner Author

Created approved-review future follow-up issues for PR #2044:

Reconciliation: 1 filed, 0 deduplicated, 0 skipped by cap.
Tracker reuse: 0 reused; 0 uncertain.

These were mentioned as future work and did not block merge readiness.

-- coding-review-agent-loop

@wwind123

wwind123 commented Nov 20, 2026 •

Copy link
Copy Markdown
Owner Author

Managed CI authorization for exact head 7c48c83.

@wwind123
wwind123 marked this pull request as ready for review November 20, 2026 15:31
@wwind123
wwind123 merged commit a99639b into main Nov 20, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Signed-source role mix-up is fatal after reviews are published, and the half-recorded round can only be recovered by deleting reviews

1 participant