KEY: keys/apr-release-minisign.pub on main — the release-asset signing key R-5 signs with and R-6 pins (R-5 blocked_by KEY)

[noahgift]

Track R (parent #2963), epic #2873. Created by the R-5 delta of 2026-09-08. Blocks R-5 (#2908) and, through it, R-6 (#2909), R-8 (#3019) and TAG-0.66.0 (#2932).

Deliverable: keys/apr-release-minisign.pub on main — the one public key (S0-19: the minisign public-key scheme, the .github/pr-review.pub class; no third scheme) that the R-5 release-assets.yml manifest job signs against and the R-6 installer pins. The secret half never enters a session: the operator generates the keypair and lands the public half by PR.

Until it exists on main: the manifest job and the C13 gate (scripts/check_release_assets.sh) refuse — exit non-zero, never a pass — and R-5 stays blocked_by: KEY.
DONE-IF: git cat-file -e origin/main:keys/apr-release-minisign.pub exits 0 → R-6 pins its fingerprint.

A:

• git cat-file -e origin/main:keys/apr-release-minisign.pub exits 0
• the file is a minisign public key (two lines; line 2 base64-decodes to 42 bytes beginning Ed); its key id and sha256 are recorded in the receipt docs/audits/impl-PMAT-1079-receipt.md — the fingerprint R-6 pins

Mutation → RED: remove the file on a branch → the manifest job and the C13 gate refuse.

Owner: noahgift · expiry: 2026-10-02 ([U]: R-5's 2026-10-09 − 6 d − 1, the slack rule) · quorum: operator · DAG row: KEY in docs/specifications/pp-066-dag.yaml (spec §5 R-5, v2.1).

---

Correction 2026-09-08 (orchestrator, agent/pp-066-spec @ cc462bfab): the ticket for this row is PMAT-1079, not PMAT-1078. PMAT-1078 was not free — agent/R-5's contracts/apr-publish-cascade-v1.yaml:4 already stamps it for the publish cascade (unminted in any roadmap). An id is taken the moment a branch writes it (pmat#1169 class).

[noahgift]

The consumers now all read one path, so the single command in this issue is unambiguous. agent/R-5 (pushed) changes:

• .github/workflows/release-assets.yml — minisign -Vm "apr-$TAG.sha256" -p ../keys/apr-release-minisign.pub
• scripts/check_promotion_receipts.sh — default PUBKEY=$ROOT/keys/apr-release-minisign.pub
• contracts/apr-release-assets-v1.yaml — pubkey: keys/apr-release-minisign.pub

It previously wrote and verified against .github/release-assets.pub, which would have given the release two candidate roots of trust and made this issue's command create the wrong one. grep -rn release-assets.pub .github scripts contracts is now empty.

So the action is exactly as filed, and nothing else on the path needs a decision:

minisign -G -p keys/apr-release-minisign.pub -s ~/.minisign/apr-release.key
git add keys/apr-release-minisign.pub && git commit && push

R-5 stays blocked_by KEY until that file is on main: the manifest job and the C13 gate both refuse without it — deliberately, since an unverifiable signature is not a verified one.

[noahgift]

Deferred out of 0.66 — D-13 (operator, 2026-09-08: "skip it and deprioritize").

0.66 ships assets that are checksummed, not signed. Nothing is asked of you for this release; this issue is now the first row of the 0.67 provenance track.

Done on the DAG: KEY → lane 0.67; R-5 and R-6 drop KEY from their blockers and are unblocked; C13's command loses + minisign signature; 0.66 lane 34 → 33 rows, invariants PASS.

What the release must now say, so the smaller claim stays honest (recorded as the decision's obligations on #2873):

1. the notes' install section states the assets are checksummed, not signed;
2. R-6's installer prints the same at install time, so a curl | sh user is told rather than left to assume;
3. no artefact, script, contract or note calls a 0.66 asset "signed" or "verified" — check_release_assets.sh drops its signature arm and a guard asserts the vocabulary so the claim cannot creep back as prose;
4. this issue stays open for 0.67.

The gap, plainly: sha256 proves an asset matches the manifest the release job produced, and the four host receipts prove that sha256 is what was tested. It does not prove who produced it — the manifest and the assets sit in the same GitHub release, so whoever can replace one can replace the other. That is what 0.67 closes.