Separate metrics accounting and review evidence

This commit is contained in:
Meta Vens 2026-07-13 04:38:07 +01:00
commit 1f5f8d1d64
10 changed files with 51 additions and 15 deletions

View file

@ -1289,6 +1289,16 @@ with an explicit acceptance-receipt requirement. Operator projections label
artifact-acceptance receipts Maintainer-owned and the accounting lane
Bifrost-owned.
## Review-load evidence as Bifrost metrics ownership
`repo.metrics_request` named `Bifrost/Maintainer` and carried a
`bifrost_or_maintainer` authority gate. The request now assigns Self routing,
Bifrost accounting, and Maintainer review-load evidence separately. Model-spend
and review-load receipts are required observations; recording them grants no
spend, review-load, or ledger mutation authority. Metrics receipts and
accounting lanes are Bifrost-owned, while the contract catalogue explicitly
leaves review evidence with Maintainer.
The same inspection found 33 consecutive duplicate `#[cfg(test)]` attributes
on provider constructors/writers and their re-exports. One gate already makes
each item test-only; the second owned nothing. The duplicates are removed.

View file

@ -61,6 +61,7 @@
- Deployment requests split Self routing from independent Maintainer/Soul/Mind/Bifrost review and Idunn-only execution. Maintainer review cannot impersonate Idunn deployment; only Idunn deployment and aftercare receipts prove outcomes.
- PR requests split Self routing, Bifrost publication gating, Hands execution, and GitHub provider outcomes. GitHub receipt authorship does not grant publication-policy authority.
- Artifact acceptance requests split Self routing, Maintainer acceptance decisions, and Bifrost accounting. Acceptance receipts and accounting ledger rows no longer claim one composite owner.
- Metrics requests split Self routing, Bifrost accounting custody, and Maintainer review-load evidence. Spend/review receipts are observations and grant no spend or ledger authority.
- `repo.planning_brief` is deleted. It contained no candidate work-item records and let copied global catalog/schema/closure constants self-attest as per-item planning and readiness evidence. The live preparation chain is consensus draft -> interpretation request -> objective draft -> Mind adoption.
- There is no aggregate repo-swarm MVP gate. The former smoke manually aggregated fixture summaries into green rows and `demoReady=true`; it and its wrapper are deleted. Whole-organism readiness requires a live-fire path whose claimed consequences are observed at their owning boundaries.
- Persona owner: `collaboration-feedback` writes public, bounded feedback only after that provider receipt exists.

View file

@ -452,4 +452,9 @@ Bifrost records accounting, and the typed request requires the acceptance
receipt. Operator receipt and accounting rows report those separate owners.
Nine external-governance fixtures, 259 library tests, and all binaries pass.
Metrics/accounting no longer has a `Bifrost/Maintainer` composite or OR gate.
Self routes, Bifrost accounts, Maintainer owns review-load evidence, and spend/
review-load receipts are required observations without authorizing spend or
ledger mutation. Contract catalogue and operator rows use the same ownership.
The last completed code pass had 259 library tests passing and all binaries compiling. Re-run focused checks for the next touched surface; use the full library/binary baseline before committing a new architectural cut.