Skip to content

Upgrade edge-review plan and implement bounded M5 source recovery - #214

Open
realmarcin wants to merge 12 commits into
fix/calibration-prerequisites-20261117from
implement/content-review-20261119
Open

realmarcin wants to merge 12 commits into
fix/calibration-prerequisites-20261117from
implement/content-review-20261119

Conversation

@realmarcin

@realmarcin realmarcin commented Nov 19, 2026 •

Copy link
Copy Markdown
Contributor

The edge-review plan omitted the implemented six-axis assessments, failed calibration rounds and current prerequisites. This updates it into a gated implementation plan and supplies bounded M5 source recovery plus typed adjudication inputs.

  • Reconcile claim, DOI/snippet literature, relation, subject-node, object-node and source-fidelity assessment requirements; preserve every existing calibration threshold and historical result.
  • Run three dedicated codex-work helpers in isolated M5 clones. Actual owner methods reproduce the first43 qualified 0% capability edge, first28 100% phenotype mapping and fresh14 type-field deposit link. Retain original artifacts, conditional taxonomy/field-meaning limits and unrecovered historical build/row linkage.
  • Bind adjudication to both original matrices, authored seals, comparison, completed runtime/readiness lineage and observed identities. Preserve the registered 20 initial canonical pairs and separately validated final drafts; the adjudicator still needs its own fresh-environment save/readback canary.
  • Record actual Valency metadata use, failed access routes and inspected public body extraction for the matching fresh05 paper. This extraction is not yet admitted as canonical paper proof and does not resolve mixed-row cell lineage. No M5 licensed-library retrieval success is claimed.

Validation: final head 64a1c65 passed 1,468 tests, 16 contract checks, Ruff and both package builds (CI); combined code at 131669b also passed independently. All three native M5 completion checks and independent source-artifact inspections passed. After failure-record repairs, the actual Python 3.10.11 runtime passed four negative smoke scenarios; a fresh M5 clone then passed first43 replay/readback before first28 through the same committed driver. Independent inspection verified all 67 new source artifacts and 22 distinct owner pins.

Separate read-only adversarial reviews and every finding/disposition are in the review record; exact runtimes, receipts, failed attempts and limits are in the execution record.

Remaining work is explicit in #220: evidence sufficiency, neutral concrete templates/configs with distinct relative reviewer namespaces, a new scientific package and fresh bound scientific sessions. No new calibration, retained-90 reassessment, 9,900 expansion or full graph-release verification occurred. Historical failures are unchanged. This PR is stacked on unmerged #211; pinned shared CLAW PR #1289 and full-text helper #102 remain unmerged.

Fixes #209.
Fixes #213.
Fixes #215.
Fixes #216.
Fixes #217.
Fixes #218.
Fixes #219.
Fixes #221.
Fixes #222.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant