refactor(bayn): totalize evidence core

[gregkonush]

Summary

• Totalize owned canonicalization and reference replay identities with local typed failures while preserving every canonical hash input and historical golden.
• Replace the generic EvidenceStore lifting boundary and test-only live constructor with pure row decisions, explicit error mapping, and dependency-record initialization.
• Prove exact replay no-write behavior, typed divergent conflicts, duplicate-row failures, and transaction rollback in focused and PostgreSQL integration tests.
• Remove eight exported one-to-one wrappers/aliases and reduce Effect generator orchestration without splitting cohesive modules.

Related Issues

None.

Compatibility and scope evidence

• Base: 3a2d6aad649a5ec935b6efc291273f0818628ca5.
• Exact canonical bytes remain {"a":{"10":1,"2":2},"z":[null,true,false,0,1.25,"é","😀"]} with SHA-256 eb5c6929cd35d9590df65fb9db0caf13177df754b1d99e17c8d58e30735486c9.
• Independent reference replay and qualification-audit golden tests pass unchanged, including immutable historical protocol evidence.
• PostgreSQL exact replay preserves the complete evidence graph byte-for-byte; divergent deterministic reuse returns typed PersistenceMismatch without writes; terminal constraint failure rolls back every evidence table.
• Rechecked gh pr diff --name-only for refactor(bayn): make qualification decisions explicit #13342, refactor(bayn): extract autonomous cycle reconciliation core #13344, and test(bayn): add real Alpaca sandbox contract proof #13325 immediately before commit: zero changed-path overlap. No execution, broker, config/application, risk/capital-authority schema, strategy/market-data, workflow/Nix/GitOps, qualification, cycle, or reconciliation path is changed.
• The final tree has exactly the same tracked directory set as base and changes no depsSource manifest or patch input, preserving the existing Bayn fixed-output dependency hash without touching excluded Nix files.

Metrics

Changed production surface, before → after:

• LOC: 3,803 → 3,946.
• Largest file: 481 → 487 LOC; no changed production file exceeds 500 LOC.
• Exports: 135 → 143; added exports are typed identity/failure data and explicit initialization. Exported wrapper/alias family: 10 → 2.
• Effect.gen: 16 → 13.
• throw: 0 → 0; OperationalError: 0 → 0; Result.try: 2 → 2, limited to typed hostile-reflection/crypto containment.
• Removed dependency edges from replay modules to the throwing hash facade and from four PostgreSQL programs to evidence-store/boundary; added one cohesive replay identity edge and consolidated typed database-error mapping.

Testing

• bun run lint
• bun run lint:oxlint
• bun run lint:oxlint:type
• bun run lint:effect — 331/331 files, zero errors, warnings, or messages.
• bun run tsc
• bun run build — 679 modules bundled.
• bun run test — 820 passed, 0 failed, 122 PostgreSQL-gated tests skipped.
• BAYN_TEST_POSTGRES_URL=postgresql://postgres@127.0.0.1:55432/bayn_total_evidence_test bun run test:postgres against an isolated local-forwarded test database — 114 passed, 0 failed: cycle observability 5, evidence store 66, accounting persistence 30, cycle store 13.
• Final focused exact-head run — 89 passed, 0 failed across canonical hashing, independent replay, qualification audit, evidence recovery, qualification row decisions, and startup composition.
• Pre-commit Oxfmt/Oxlint and commitlint hooks passed.

Breaking Changes

None. The removed canonical JSON aliases and EvidenceStore test constructor had no active production callers; the retained canonical hash compatibility alias still serves active out-of-scope callers.

Checklist

• Testing section documents the exact validation performed.
• Screenshots are not applicable; Breaking Changes is filled in.
• No documentation or release-note change is required; compatibility and migration behavior are covered by tests and this PR evidence.

[gregkonush]

[gregkonush]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Bravo.

Reviewed commit: 23de744561

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[gregkonush]

Exact-head review/CI disposition before merge:

• Codex reviewed 23de744561 and found no major issues.
• Bayn pr-checks, Effect compatibility, broker sandbox contract, changed-area validation, repository CI, and PostgreSQL integration all pass on the exact head. The PostgreSQL rerun passed after one excluded cycle-recovery concurrency flake.
• Both image jobs fail at the pre-existing Bayn Bun fixed-output hash (specified sha256-TnXW...). The exact base SHA 3a2d6aad649a5ec935b6efc291273f0818628ca5 has the same failed image workflow: run 30340552093.
• This PR changes no Nix/workflow path, no package manifest/lock/patch input, and preserves the base tracked directory set. The image repair is explicitly outside this PR's ownership.