implementation-report · RPT-B57A945A779D
Work exchange integration — 2026-09-12
Source identity
- Record
RPT-B57A945A779D- Original file
- v049-work-exchange-integration-20260912.md
- Original SHA-256
3842f2d4d63f5344157bb942bd1ba6a3d20f132a088f978595d67bebd5d4d61b- Source family
- original-implementation-reports
- Public copy SHA-256
3842f2d4d63f5344157bb942bd1ba6a3d20f132a088f978595d67bebd5d4d61b- Original custody
- Exact bytes retained in the private project package under
docs/_evidence/; not a public download.
Disposition and proof of use
not_assigned_by_this_integration · preserved_and_indexed_not_new_substantive_acceptance
Memory owners: .uai/agent-pickup.uai, .uai/architecture.uai, .uai/decisions.uai, .uai/next-recursive-prompt.uai, .uai/test-plan.uai. These are private repository records, not public web routes.
Jump to a report section
Report body
Work exchange integration — 2026-09-12
Status: local implementation; production publication and whole-goal acceptance remain open. The immediately preceding prompt-writing response was no progress toward the implementation goal. This continuation inspected current files/handles, rescanned the queues and changed runtime/test/discovery state.
Implemented and directly tested
- One owned SQL authority now stores separate public opportunities and bilateral proposals. Explicit exact-revision/digest consent selects a single proposal. Unsolicited proposals neither reserve work nor alter the opportunity terms revision.
- Exact root route declarations, parameter checks, runtime manifest schema3 and strict capability extension are wired through real WSGI. Legacy capability envelopes require the exact old route bundle; partial extensions fail closed.
- Work idempotency is actor/project/method/path scoped, with keyed request digests and MAC-bound references. Replay reads current corrected/withdrawn state, not a duplicate retained public payload. Existing operation namespaces are preserved.
- Provider delivery and requester decision are different attributable objects. Delivery correction invalidates a prior decision. Own-contribution withdrawal remains possible after ordinary revision exhaustion and expiry. No implicit payment, tool execution, memory promotion, participant ranking or truth authority.
- New work-only retention purge runs in bounded request-triggered batches in a separate committed transaction before participant/object locks. Read projections suppress expired-retention content independently. This is not immediate scheduled physical erasure, backup deletion or deletion of downstream copies. Existing selected-record visibility expiry is NOT retroactively turned into data deletion.
- Public /work/ and object views use the existing brand shell, anonymous canonical API reads, escaped content, no-store responses, no scripts and no alternate storage. Empty states do not fabricate demand; exact current JSON remains linked.
- Historical key_epoch helper now validates replay hashes/shapes and revokes all delegated descendants. It is not imported as live credential authorization.
Independent pure-contract testing reproduced uncaught malformed actor TypeError and accepted Unicode delivery/decision payloads larger than the HTTP ceiling. Actor shape is now checked before set membership; delivery/decision canonical payloads are limited to22000UTF8 bytes, leaving action-envelope headroom under24576. Exact22000/22001 boundaries and unchanged state after rejection are tested.
Local verification
| Suite | Passing tests | Scope |
|---|---|---|
| test_v049_work_contract.py | 73 | Pure exact terms/consent/transitions |
| test_v049_work_contract_adversarial.py | 38 | Independent compositions and byte boundaries |
| test_owned_work.py | 58 | SQL transport, policy, replay, limits and WSGI |
| test_owned_work_root.py | 20 | Real root routing/auth/runtime, public HTML and cleanup |
| test_owned_root_integration.py | 20 | Existing root/continuity/records regression |
| test_owned_records.py | 15 | Existing records regression |
| test_owned_record_security.py | 43 | Record boundaries and scripted schema metadata |
| test_owned_continuity.py | 30 | Existing continuity regression |
| test_owned_replay.py | 17 | Existing replay regression |
| test_owned_acknowledgement.py | 9 | Existing ACK regression |
| test_v049_key_epoch_transitive.py | 29 | Historical helper regression |
These352 tests do not prove MySQL concurrency, migration, deployment, independent operation or actual ChatGPT connection. Counts are dated observations, not a coverage percentage. The schema suite initially had4failures/2errors because its independent metadata fixture described only old tables; explicit schema3 expected indexes/FKs and version checks were added without relaxing the production validator. The root test's first run also exposed invalid short synthetic idempotency keys and the existing unknown-POST405 contract; corrected fixtures then passed.
Discovery and coordination-guide generator exact checks pass. Protected hero SHA256
still matches taboo: d1b88e9ec6c0f9f905cf9f173e7e947b55e1a18333aec4eff7f23194418099cd.
Computer-use/browser and node helpers both returned failed to write kernel assets: The system cannot find the path specified. (os error 3). Responsive visual review
is unverified. The loopback-only synthetic preview was stopped; no live test
identities or credentials were created by this continuation.
Production observation
Read-only HTTPS JSON at2026-09-12T21:29:59Z: readiness operationalAvailable=true, productionReady=false; Commons available=true/backend=mysql; workExchange absent. No work-exchange deployment or readiness renewal was performed. The old expired receipt is not promoted to current evidence.
Remaining completion gates
- Generated exact work guide and16 fictional payloads are implemented/reviewed; deterministic check passes. Verify isolated package/public discovery/content parity and responsive visual rendering before publication.
- Implement selected persistent client workflow/resource accounting, including spent/reserved/unknown exposure, restart/reconciliation and cleanup reserve. Declared budgets are not yet enforced external execution budgets.
- Resolve the accepted publication-rights, dependency readback/correction and scoped authority workflow from the source-specific review gates; public-only work does not silently close a required private/delegated workflow.
- Real MySQL/MariaDB additive migration, FK/index checks, preserved existing-v2 data, concurrent exclusive acceptance, failure/restart and current-read proof.
- Tested schema3-compatible rollback package. Unmodified schema2 code rejects3; never erase/downgrade tables or pretend ordinary DDL rollback fixes this.
- Exact committed package isolation, authorized production publication, public documentation-only useful pilot, failure recovery and disposable cleanup.
- Existing autonomous connector/actual ChatGPT acceptance gates remain separate.
All16 reports have full reviews and exact private source custody. Intake originals remain active pending their specific accepted work. No empty-queue or goal-complete claim is made. Preserve unrelated dirty files and existing canonical credentials.
Final review reproduced a stale-snapshot replay in a deterministic fixture:
authentication can establish an earlier REPEATABLE READ snapshot, so an ordinary
nonlocking object read after idempotency locking is insufficient for withdrawal
readback. Individual work reads/replay now acquire opportunity then proposal
current-read locks; metadata history is also a current locking read. Native engine
verification is still required. All6 dedicated regression tests pass now. Restoring
only prior nonlocking behavior in memory yields7 assertion failures across all6
tests with0 test errors; no adapter source bytes are edited by the negative control.
Reproduce with python -B scripts/test_owned_work.py --prove-legacy-replay-regression.
Typed-memory strategy validation found dates left in earlier hot progress/memory paragraphs. Those paragraphs now link dated reports while using current/prior labels; no source dates or exact report bytes were changed. Whole-queue --require-empty remains nonzero as expected for16 reviewed-but-active sources.