Native work-exchange verification — 2026-09-12

Report body

Native work-exchange verification — 2026-09-12

Status: controlled isolated native MySQL PASS; production deployment and full autonomous-service acceptance remain open. The preceding prompt-writing turn did not advance implementation. This continuation produced native evidence and revalidated the live predecessor, rather than repeating its prior status.

Direct observations

An already installed MySQL Enterprise Server 9.7.0 binary was used to initialize a new private, loopback-only instance. No existing database service was stopped, reconfigured, authenticated to, or reused. The runner supplies no default option files, binds only 127.0.0.1, disables MySQL X and binary logging, creates two fresh test databases, and grants its random test account only database-scoped table permissions. Privileged fixture provisioning is separate from application bootstrap. The runtime never creates databases, accounts, or grants.

Both dedicated databases require a short-lived exact capability marker before the test application may create schema or write rows. All application tests use actual OwnedCommonsApplication, SQL, authentication and work policy. There is no SQLite adapter, SQL rewriting, fake native version, or disabled foreign-key check.

The native run passed all ten named gates for EACH baseline (1 and 2):

1. Historical baseline construction using exact existing migration tuples.
2. Additive upgrade to schema3 preserving seeded pre-upgrade rows and ledger.
3. Native work index and foreign-key metadata.
4. Reapplication and fresh-application/connection initialization.
5. Bilateral WSGI agreement, delivery, decision, withdrawal and replay readback.
6. Native foreign-key and event-primary-key rejection.
7. Opportunity withdrawal replay after a demonstrated older REPEATABLE READ snapshot.
8. Proposal withdrawal replay after a demonstrated older snapshot, including history.
9. Two actual concurrent connections producing exactly one exclusive acceptance.
10. Exact fixture-row cleanup with ownership guards and zero-row verification.

Cleanup removed 61 fixture rows from the first test database and 62 from the second. Schema, migration history and capability markers were not dropped. The exclusively owned server process shut down with exit0. Its generated plaintext bootstrap/config credentials were removed after shutdown. Stopped database files remain in private fixture custody outside the repository; they are not package inputs. The earlier setup smoke's two credential files were also removed after its stopped-process evidence was verified.

Exact content-free source hashes and per-target results are in reports/v049-work-exchange-native-mysql-20260912.json.

The hardened runner repeated the full native run on another freshly provisioned instance, again passing both baselines and all ten gates with exact cleanup and exit0 shutdown. Its 41 offline safeguard tests initially exposed five assertions across three defects: decoded-JSON secret output, incomplete evidence accepted despite a zero child exit, and an unexpected shutdown exception skipping child reaping. All three runner defects were repaired; all 41 tests pass unchanged. The native harness also passes 17 offline safeguards. These 58 local tests are separate from the twenty native gate observations per complete run.

Limits of this evidence

Historical baselines are constructed from the original migration definitions; they are not execution of historical deployment binaries. Fresh application instances are not Passenger restart or an actual stopped-server restart test. The concurrent test observes distinct connections and an outstanding competing call while the winner holds the row; it does not claim server lock-wait telemetry. The tests do not prove MariaDB-specific behavior, production hosting, outside adoption, actual ChatGPT connectivity, end-to-end privacy, or semantic usefulness.

The fixture smoke first stopped at its permission check. New security descriptor assignment attempted an unavailable audit privilege. DACL-only directory setup and the repository's existing private-directory convention (owner/OWNER RIGHTS, SYSTEM, Administrators only) succeeded. No audit privilege or service permission was obtained or bypassed. Production credentials were never used for native tests.

Revalidated production predecessor

Read-only HTTPS observation at 2026-09-12T22:08:43Z: operationalAvailable=true, productionReady=false, Commons available=true/backend=mysql, no workExchange. No readiness receipt was renewed and no production writes occurred.

Certificate-verified FTPS reads confirmed the live bytes of owned_commons.py, owned_schema.py, owned_schema_v2.py, owned_schema_checks.py, runtime_manifest.py, runtime_manifest.json, application.py and route_registry.py exactly match commit c2d50266edc04d92be119510b821346a0ce8b695. This is an eight-file predecessor check, not authentication of the entire live root or a completed rollback test.

Still required

• A schema3-compatible baseline rollback artifact is now prepared and passes 29 offline archive/import/schema-metadata tests. Its builder copies exactly six compatibility files from the pinned candidate into the pinned baseline; all other baseline bytes remain, except two generated identity/provenance members. Its derived identity does not pretend to be a single-commit build. The original archived baseline rejects a schema3 ledger; the derived baseline accepts the scripted schema3 metadata while still rejecting bad checksums, missing foreign keys and future versions. These are NOT native rollback tests. Native rollback/restart rehearsal and work-retention maintenance while the old core is active remain required. The old core has no work purge path; preserving tables alone does not fulfill new work-retention obligations.
• Durable client work intent, resource reservations/unknown exposure/cleanup reserve.
• Accepted publication-rights, dependency correction and scoped authority workflows.
• Responsive visual QA: both browser and Windows helpers still fail initialization with failed to write kernel assets: The system cannot find the path specified. (os error 3). No screenshot or rendered geometry evidence was obtained.
• Exact committed package isolation, authorized publication, deployed readback, useful public-doc-only pilot and its cleanup.
• Separate autonomous connector/actual ChatGPT acceptance gates.
• Source-specific implementation closure before intake-copy removal. Current scanner: 16 fully reviewed active reports, 67 closed, integrity PASS; not empty.

Packaging evidence for the native milestone is reports/v049-work-native-package-evidence-20260912.json: 125 selected checks pass from the extracted committed repository, and its actual native run passes both upgrade paths and all ten gates. Native hashes agree with worktree results. Root delta from the prior work candidate is only runtime/release.json.

Rollback preparation owners: scripts/build_owned_work_rollback.py and scripts/test_owned_work_rollback.py. The first prepared ROOT ZIP is concresca-v0.49.0-root-deploy-c2d5026-schema3-rollback-06c77ebc86acf4fd-wip.zip, SHA256 ac57466463bd5617575847f65ee53daec3aed7d595fb6738073cc026249c30b7. Its detached provenance names exact source packages and builder bytes. No rollback was deployed or advertised as production-ready.