feat(analysis): bind retrospective-edge refusals to an analysis-run profile

[seonghobae]

Consolidation status

fold_into_landing_vehicle — this cutoff-safe retrospective_edge_v1 profile targets #416 (feat/copy-identity-analysis-run-gap-004), not protected main. It changes the same Analysis Run/Validation shared Cargo/lib/docs surfaces as #416, so it remains a Draft fold child rather than independent architecture authority.

Current exact source head: 847d96f913bb261803ac0bd751ad7e4f51324cee.

Preserved profile evidence

• Profile/contract retrospective_edge_v1; schema tepp.retrospective_edge.v1.
• Inference status retrospective_report_is_not_transition_and_not_translation.
• Source-specific implementation crates/analysis_engine/src/retrospective_edge_artifact.rs plus cutoff/execution contracts and doctoring.
• Request cutoff equality is typed through KnowledgeCutoff::instant().
• MAX_EVIDENCE_UNITS bounds execution admission and artifact validation.
• Retrospective reports remain distinct from forward reports and are refused as both state transitions and translations without erasing compatible forward relations.
• Terminal summary.validation_status is validated; the domain claim remains in inference_status.
• identity_recovery_rate remains library-side and inspect payloads remain metric-free.

Leakage repair

Fresh source review found the same historical-cutoff leak as #482: census_admitted_assignments inserted every assignment identity into duplicate detection before excluding assignments whose available_time was after the knowledge cutoff. A future-unavailable row could therefore make an earlier historical run fail DuplicateEvidence, meaning future evidence existence affected an earlier decision even though the row was inadmissible.

• RED 32394a327148ad57578f40436233f051cd31eb54 adds future_duplicate_identity_cannot_change_a_historical_cutoff_result with a future ForwardReport reusing an eligible retrospective identity.
• Causal repair 847d96f913bb261803ac0bd751ad7e4f51324cee applies the availability cutoff before duplicate-identity admission. Duplicate identities among evidence actually available at the cutoff still fail closed.

This ordering is part of TEPP's leakage-safe AvailableTime/knowledge-cutoff contract and must survive the eventual #416 fold. Hosted exact-head evidence must be reacquired on this new source head; predecessor checks are stale.

GitHub still reports this child as non-mergeable against #416 because the shared Cargo/lib/docs surfaces require a real conflict-resolving fold. Do not land this child independently and do not discard its unique source/tests/doctoring or RED/repair lineage.

ADR 0076 remains implementation lineage pending repository-wide normalization under #437; implementation maturity does not create branch-local architecture authority.

Merge authority

Only the eventual surviving #416/successor exact head counts after the fold. Reacquire all live required workflows, resolve current review threads, and satisfy the live organization ruleset's qualifying independent-review requirement on that surviving head. No self-approval, force push, routine admin bypass, or stale evidence transfer.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}