Repository navigation
fix(reporting): keep snapshot finality source-owned - #1924
rachitm022 merged 2 commits into
Conversation
There was a problem hiding this comment.
Ladon verdict: Approve
Approve.
Clean revert-plus-hardening PR. No blocking findings, no medium findings, no escalation triggers.
What I checked against the decision table:
- Rows 1 (critical/high): no findings of any severity. Pass.
- Row 2 (gated_paths): gated_paths is false — row does not fire.
- Row 3/5 (high_risk): high_risk is false — rows do not fire.
- Row 4 (medium data-loss/schema/infra): no findings.
- Row 6 (sticky escalate): no prior decision.
- Row 7 (no-auto-approve team): no team match.
- Row 8 (≥3 medium findings): zero medium findings.
- Row 9: falls through to approve.
Reviewer notes confirm the semver signal is correct: the removed symbols (official_stability_duration, require_finality_evidence, finality_evidence_ref/finality_stable_since) were introduced in 885af86, never shipped (targeted beta.21; latest tag v10.0.0-beta.20), and adcp.reporting is not in the enumerated public re-export tree — so the fix: prefix is correct and no released contract breaks. Unchanged-content reuse now verifies immutable rows/source fingerprints under lock in both memory and PostgreSQL stores, with an atomic repair-revision path committed inside the account lock/transaction and staged rows discarded in finally. Decode compat (from_wire pops require_finality_evidence; manifest retains stable_since for round-trip) is tested, and the catalog-scope test correction aligns with the already-dropped unique constraint/index.
There was a problem hiding this comment.
Ladon verdict: Approve
Approve — clean pass, no blocking findings.
The delta since the prior approve is confined to tests/conformance/reporting/test_reporting_core_lifecycle.py: test-expectation adjustments tracking the shipped "suppress unchanged revisions" behavior (revision chain stays at 1 after identical-content reads; revisions_committed drops the suppressed automatic revision; restatement parents directly on the original). Expectations are internally consistent with producer.py:449 and the commit_revision_from_manifest predecessor-retain path. No new production code, and the prior run's zero findings do not re-surface.
Decision rule trace: no critical/high/medium findings (row 1, 4, 5, 6, 8 don't fire); gated_paths is false (row 2 n/a despite review_decision REVIEW_REQUIRED); high_risk is false (rows 3, 5 n/a); no no-auto-approve team match (row 7 n/a). Falls through to row 9 → approve.
Why
Reliable Reporting 1.2 is snapshot-only and declares
emit_only_on_content_change: true. The SDK must not infer an official close or require source-stability evidence for that offering, and unchanged scheduled reads must not mint revisions.What Changed
official_close_lagbehavior unchangedVerification
Closes AI-22538