fix(session): complete created-session persist evidence

[cursor]

Superseded / split into durable session and recovery lanes

Close this predecessor without merge. Its session work has been carried forward through the repaired persist/load/start lineage, while its one unique late recovery change is isolated on #177.

Fresh evidence immediately before closure:

• fix(session): complete created-session persist evidence #106 exact head: 3a0c526073c76d29a58f07cad1c1dead7d12cb55
• current durable-session HTTP line feat(api): start and reload sessions over persist-backed HTTP #232 exact head: c7b45145c9b693951fae498e0779d58cbe170f86
• dedicated recovery line test(recovery): preserve inbox claim deadline evidence #177 exact head: 8a0db4611a684285185f9edadf086a1808444ab9
• protected main: a7637351be8f0f90c12651d3bcafd959bc52ac81

Current reviews already direct the Created-session work away from #106 toward later persist/load/start successors. A fresh commit comparison confirms #232 contains the later session lineage and is 23 commits beyond the common session base; the only #106-side delta not present on #232 is tests/postgres_recovery_invariants.rs (27 additions / 3 deletions). That is exactly the claim_deadline_at COPY/restore concern deliberately isolated on #177.

#232 carries the repaired #218 persist/load/command-history/start-lock path plus public start/reload HTTP; #177 carries the required recovery deadline seed and binary-COPY equality evidence without coupling it to session persistence. Keeping #106 open would therefore retain a competing session writer only to preserve work that now has a dedicated recovery vehicle.

There are no inline review threads on #106 to resolve. Successor behavior remains Active PR only until its own unchanged exact head passes all live CI/security/review gates.

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[cursor]

Stale comment

Review (PR #106, head 60b090d)

No blocking persist/replay defect on this head. This is the correct landing vehicle for the #61 created-session slice.

Verified

• Created-only insert, ON CONFLICT DO NOTHING, then exact-identity classify under READ COMMITTED.
• Rebinding participant, release, instrument_version_ref, digest, locale, state, or creation time fails closed.
• Public persist contract names instrument_version_ref and states that AssessmentSession::new validates references while persist stores them without validating them again.
• Happy-path PostgreSQL test asserts the committed instrument_version_ref.
• TRACEABILITY / CHANGELOG / as-built schema / ERD stay Active PR #61, not Implemented.
• Migration 0014 does not collide with #82 0020. Do not rename it.

Residual (not blocking this slice)

• Persist is write-only. A buyer still cannot resume or start HTTP against a stored session because there is no load/reconstitute path, and AssessmentSession::new would re-check current publication eligibility. That is the next slice, not a defect in this one.
• After this lands, close or retarget #61. Do not land draft #100 (0020) over this slice.

Merge gate
Do not merge until required checks are terminal-success on exact head 60b090d and independent last-push approval is recorded. I am not approving on queued checks.

  

_{Sent by Cursor Automation: Fix Issues}

[cursor]

Stale comment

Successor

Head 60b090d remains the persist-only slice. Created-session load without re-checking current publication eligibility is now on #109. Prefer landing #109 after its checks and independent last-push approval. Close or retarget this PR after #109 lands so persist-only and persist+load do not race.

View PR

  

_{Sent by Cursor Automation: Fix Issues}

[seonghobae]

Exact-head RCA after 60b090d6: Runtime CI failed at postgres_recovery_invariants (integration_consumption_claim_deadline_shape). Head now includes the #81 recovery seed repair (claim_deadline_at) while keeping the created-session persist evidence on Active PR #61. Please perform a bounded independent current-head review of this exact SHA against protected main a7637351. Do not update the branch or merge.

[cursor]

Review (PR #106, head 3a0c526)

No blocking persist/replay defect. Created-session insert, exact replay, and fail-closed rebinding remain sound. This head is not the landing vehicle for Created persist/load/start.

New commit 3a0c526

• Seeding claim_deadline_at on the processing recovery fixture is required after #81. maintain_inbox_claim_deadline is BEFORE UPDATE only, so a processing INSERT without that column violates integration_consumption_claim_deadline_shape.
• Asserting exact source/restored claim_deadline_at equality after COPY ... FORMAT BINARY is the correct restore invariant for database-authoritative claim expiry.
• Keep this recovery repair off the persist stack. Prefer a main-based recovery-only successor so #72 evidence stays valid if #106 is closed in favor of #164.

Landing

• Prefer #164 (b47b035) for Created persist/load/start-from-stored-release.
• Close or retarget #106 after #164 lands. Do not land #61 at 7f7ea0f or draft #100 (0020) over this slice.
• Later-state lock landing remains #154. HTTP POST /v1/sessions stays #87/#149/#161 after #164.

Residual (not blocking persist)

• Recovery tables still omit assessment_session. That belongs after persist lands, not in this evidence-only slice.
• TRACEABILITY/CHANGELOG/as-built/ERD correctly stay Active PR #61, not Implemented.

Merge gate
Do not merge until required checks are terminal-success on exact head 3a0c526 and independent last-push approval is recorded. I am not approving on queued checks, and I am not approving this automation's own persist lineage.

  

_{Sent by Cursor Automation: Fix Issues}