5e4191578a
fix(tolaria): C5D search-source pipeline — build+verify source from current Tolaria before rebuild
...
- SearchSourceBuilder: deterministischer Vault->Source-Snapshot (read-only),
atomar (temp->validate->fsync->replace), Secret-Scan fail-closed
- verify_integrity: exakte object_id/path Set-Equality (stale Source kann
nie APPLIED), Canary implizit ueber erwartetes Objekt-Set
- apply_commit: Source-Build+Verification vor Rebuild; VERIFYING_SEARCH-Resume
(kein Doppel-Rebuild) — C5E-Replay-Crash-Fall abgedeckt
- Fix: source_object_count ist keine 0-Fehlerbedingung (echter Defekt)
- SearchSourceBuildError + RC_SEARCH_SOURCE_BUILD_FAILURE (Human Gate)
- c4b: source_path env-konfigurierbar, indexed_object_ids/paths Read-Back
- Testsuite: 17 neue Tests (Test-Plan A-O + Realistic C4-Integration)
2026-08-26 12:12:45 +00:00
22e1cd0d44
C5E: FAILURE/REPLAY/RECOVERY + OBSERVABILITY (fail-closed Library)
...
- rq_c5e.py: C5EEngine (Recovery-Entscheidung RESUME/RETRY/WAIT/HUMAN_REVIEW/ALREADY_APPLIED,
deterministisches Replay, Partial-Commit-Recovery via object_progress), C5EReconciler
(read-only, kein blindes Repair), observability(), health_contract() (HEALTHY/DEGRADED/BLOCKED),
failure_evidence(). FAIL-CLOSED: Standard allow_writes=False, kein Polling/Daemon.
- rq_c5a.py: +4 Reason-Codes (SEARCH_UNAVAILABLE, NETWORK_TIMEOUT, MALFORMED_RESPONSE,
INTEGRITY_FAILURE) -> REASON_CODES 13->17 (minimal, dokumentiert, regressionsgetestet).
- rq_c5_cli.py: +7 C5E-Befehle (recover, replay, reconcile, observability, health,
evidence, guarantees), alle fail-closed.
- test_c5a.py: Assertions auf 17 Reason-Codes angehoben.
- test_c5e.py: 58 Tests (Restart/Retry/Replay/Partial/Ordering/Drift/Health/
Observability/Persistence/Reconciliation/No-Doppel-Writes).
Regression: C5A 25/25, C5B 40/40, C5C 35/35, C5D 35/35, C5E 58/58. Alle gruen.
Keine produktive Aktivierung.
2026-08-26 06:54:10 +00:00
ed374ac4ce
fix(c5b): persist human-review object changes + read content_before from parent commit
...
- C5A objects table: object_id nullable + reason_code column so HUMAN_REVIEW/
SECRET_DETECTED object changes (object_id=None) are persisted, not silently
dropped (was: object_id TEXT NOT NULL, no reason_code field)
- C5B poll_once: content_before now read from parent_sha (state BEFORE the
change) instead of sha, so MODIFIED changes classify as CONTENT_UPDATE
instead of being misclassified (Checker-Befund)
- test_c5b: add test_modified_content_reads_parent regression test
C5B 40/40, C5A 25/25, real-repo dry run: 233 object changes (109 IN_SCOPE
with valid id, 124 HUMAN_REVIEW), idempotent.
2026-08-26 05:20:38 +00:00
