R12 baseline and R13 admission refusal

This bundle retains two separate attempts. R12 has 44 completed and graded incumbent jobs (38 pass, 6 fail), 4 cancelled jobs with unknown grades, and 48 unrun candidate cells. R13 separately retains 48 unsubmitted candidate cells after the host-memory reserve guard refused container creation. No model comparison or promotion follows. Exact incumbent restoration and explicit minimal-thinking Pi acceptance are verified.

• Authoritative derivative summary, R12 full native report, R13 full native report.
• R12 closure/cancellation audit extract, R13 refusal, prospective 44/4 population and frozen candidate.
• R12 projection ledger, R12 author-side projection validation, supplemental provenance, configured R12 identities.
• Sources, friction, publication summary, artifact manifest, manifest source.

The four projection families retain R12 native schemas and all selected records. Shared batch telemetry is retained once. Private originals remain authoritative; original-native hashes are commitments to different bytes and cannot validate this sanitized graph. R13 full prepared plans remain private, so complete R13 graph validation is not claimed. Missing cancellation output remains missing; complete interval telemetry does not prove complete per-request accounting.

Restoration and bounded restoration evidence retain the exact restore and single minimal-thinking Pi pass, with 29/29 independent checks. The historical thinking-off failure remains open. Reviewer R1's false negative and offline R2 correction are retained; there was no live retry. The earlier 127 reviewed artifacts remain separate and unchanged. Independent content reviews passed; the publication PR binds the final inventory and named ADR-0027 size exception. Artifact finalization does not make the incomplete model comparison complete.