roadmap: dedupe twelve duplicated ids; PMAT-1060 README CPU-only claim vs wgpu default, for 0.65.3 (Refs PMAT-1060)

[noahgift]

H4 of the BSE-001 convergence brief (repair once, commit): twelve epic-subtask stubs collided with full top-level items; the full item was kept and the stub dropped each time; pmat work validate now reports zero duplicates. PMAT-1060 files the 0.65.2 post-publish dogfood NO-GO (README says CPU-only, the plain install ships the wgpu backend) for 0.65.3; per spec v1.3 §3 the freeze is satisfied by the dogfood run and recorded.

Pmat-Ticket: PMAT-1060

[github-actions]

§13.11 rung 1 — quorum shadow verdict

S13-SHADOW pr=3027 head=94ba1984b9efa1b3890a2c77a5d325be00ebdadd verdict=REFUSE class=Q1 arm_rc=1

Shadow mode: this records a verdict and merges nothing. A refusal
to arm is not a block (§13 adds zero rows to §7) — the pull request is
exactly as green as it was.

[noahgift]

FYI from the PP-066 driver session (instance A): #3030 carries the identical byte-for-byte dedup of these twelve ids (opened at 18:10Z without re-listing PRs — my I0 miss; both deletions merge cleanly in either order). Three things worth knowing here:

1. This PR's guard-runner-labels fails at G-6 ("a PR's roadmap diff only adds entries or edits lifecycle fields") because subtasks was not a lifecycle key. fix(roadmap): PMAT-1072 — main RED: drop 12 legacy nested subtask records pmat 3.39.0 refuses as duplicate ids; in-repo pin-independent id-uniqueness guard #3030 adds it (scripts/lib/roadmap_diff.py LIFECYCLE_KEYS + self-test row 3b), so once fix(roadmap): PMAT-1072 — main RED: drop 12 legacy nested subtask records pmat 3.39.0 refuses as duplicate ids; in-repo pin-independent id-uniqueness guard #3030 merges, a gh pr update-branch 3027 should turn this green; until then the guard refuses it as written.
2. Root cause of the main red this repairs, measured: the fleet's PATH pmat moved 3.37.0 → 3.39.0 at 17:26–17:33Z (intel + lambda) and sovereign-ci's roadmap-valid runs bare PATH pmat; main @ 3792afa push run 34049248452 is RED on exactly these twelve ids. Tickets main RED: ci / security roadmap-valid rejects 12 legacy nested subtask records as duplicate ids under pmat 3.39.0 (fleet PATH pmat moved 3.37.0 → 3.39.0 at 17:23–17:33Z; the 3.37.0 pin passes) #3028 (aprender) and paiml/infra#468 (pin declaration vs hosts). fix(roadmap): PMAT-1072 — main RED: drop 12 legacy nested subtask records pmat 3.39.0 refuses as duplicate ids; in-repo pin-independent id-uniqueness guard #3030 also adds an in-repo, pin-independent duplicate-id guard so the runner's pmat version is never the only thing standing between main and a duplicate.
3. PMAT-1060 minted here collides with PMAT-1060 hand-minted on docs(PP-066): session 2026-09-06 — SPEC-2.0 (v2.0, 18-row scope, claims ratchet, executable criteria), DAG rows/edges/amendments, decision receipts, roadmap mints, rescope quorum record #3025 (R-0b). docs(PP-066): session 2026-09-06 — SPEC-2.0 (v2.0, 18-row scope, claims ratchet, executable criteria), DAG rows/edges/amendments, decision receipts, roadmap mints, rescope quorum record #3025 will renumber its own; nothing to do on this side.