fix(release): the train's identity is derived — release scripts carried 0.68.2's V/MS/EPIC/LAST_TAG and an out-of-repo AP (PMAT-3618)

[noahgift]

Closes #3618. Refs #3592 #3459 #3617.

The in-tree release scripts (ported from the 0.68.2 bundle, #3599) still carried V=0.68.2, MS=12, EPIC=3477, LAST_TAG=v0.68.1 and an AP under the operator box's RAID mount, which check_hardcoded_paths.sh cannot see (#3592). Cutting 0.69 through them would have targeted the old version and an out-of-tree directory.

done_when, item by item

#	Criterion	How
1	V/T/MS/EPIC derived or passed	The version is passed (first argument of all four scripts, refused unless X.Y.Z); T="v$V"; milestone = the one titled exactly V; epic = the one epic item in it titled EPIC: release train V …; last tag = highest v* below T. 0 or 2+ answers → rc 2, named. Checked against history: for 0.68.2 the derivations return milestone 12, epic #3477, last tag v0.68.1, which are exactly the literals they replace. For 0.69.0 they return milestone 6, epic #3080, last tag v0.68.2.
2	AP derived from the repository	<main checkout>/target/release-train/<T>, where the main checkout is the parent of the repo's common git dir, so every worktree resolves the same AP (prepare_bump writes release_notes.md, autopilot reads it). RELEASE_AP overrides at run time and is never a literal.
3	/mnt blind spot closed or own assertion	Own assertion: scripts/check_release_scripts_derive_identity.sh. R1 is no absolute path under /mnt /home /Users /opt /srv /media /root on any line (comments included). R2 is no literal assigned to V T MS EPIC LAST_TAG AP. It runs in guard-tree (--no-cargo dry-run lists it). #3592 stays open for the analyser.
4	Falsifier	On a copy of the real scripts/release/, planting AP=/mnt/nvme-raid0/agent-wt/rel-0690-autopilot in autopilot.sh turns R1 and R2 RED (rc 1), and planting MS=0.69.0 in prepare_bump.sh turns R2 RED. The case table has 14 rows. Each of 5 mutants turns it RED. Its first real finding was this PR's own library comment quoting the old path.

Also fixed on the way

• autopilot.sh ran python3 "$AP/ledger.py", a leftover of the bundle layout that would not exist in a derived AP. It now reads scripts/release/ledger.py.
• publish_strict.sh read publish-order.txt from the bundle dir. It now reads it from the tag's own checkout, $WT/scripts/release/.
• prepare_bump.sh: branch release-<V>, Pmat-Ticket from the derived epic. The /mnt fallback for arm_pr_automerge.sh is removed (the in-tree script is required). The launch line prints the in-tree autopilot with its version.
• ledger_from_run.sh's "tag": "v0.68.1" is now the v* tag the tested sha carries.
• 06x-release-schedule.md §4 now names the in-tree scripts and their one input.

Not changed: publish_strict.sh's crate count (74) is a 0.68.2-era constant of the same shape but outside #3618's five names. Its 4 bashrs errors are pre-existing false positives, identical on main.

Measured

Guard 14/14 + PASS on the tree · check_tag_step_gated PASS + self-test · check_t2_preflight_verdict PASS · t2_preflight --self-test OK · sourced-libs option-neutral (9) · bashrs-gate PASS · pipe-into-grep-q at ceiling · hardcoded-paths +0 · apr pinned · guards-are-wired · readme claims · roadmap aggregate + sorted/unique/additive · fmt · deny · aprender-contracts --lib 1666.

🤖 Generated with Claude Code

[noahgift]

quorum-review (AD-04): three PASS — agreed (auto_merge: checked=true was_armed=false disarmed=false)

{
 "ticket": "PMAT-3618",
 "head": "e9ca589997b042bb38572975c812fda8806be123",
 "width": 3,
 "executor": "agy",
 "agreed": true,
 "auto_merge": {
  "checked": true,
  "was_armed": false,
  "disarmed": false,
  "note": "auto-merge not armed"
 },
 "lanes": [
  {
   "lane": 1,
   "verdict": "PASS",
   "findings": 3
  },
  {
   "lane": 2,
   "verdict": "PASS",
   "findings": 3
  },
  {
   "lane": 3,
   "verdict": "PASS",
   "findings": 4
  }
 ]
}

[noahgift]

Cross-inspection of docs/audits/quorum-PMAT-3618.json lanes (non-author: aprender-04, 06:13Z)

lane	conversation	status	verdict	findings	refs to sibling lanes / $WORK	duration
lane 1	d7c2f7c4	SUCCESS	PASS (structured_output)	3 (3 cited, 0 with commands)	0	156 s
lane 2	bcaf70ba	SUCCESS	PASS (structured_output)	3 (2 cited, 0 with commands)	0	48 s
lane 3	3e46312b	SUCCESS	PASS (structured_output)	4 (0 cited, 4 with commands)	0	624 s

Distinct agy conversation ids: 3/3; no lane references a sibling lane or $WORK. Read from /mnt/nvme-raid0/agent-wt/rel-3618/docs/audits/quorum-PMAT-3618.json.lanes on this box. Note: release scripts take the version as their one input; milestone/epic/last tag derived (history proof: 0.68.2 → 12/#3477/v0.68.1, the exact literals replaced); AP derived from the repo; new guard plants the /mnt AP and goes RED; operator-facing change: autopilot.sh 0.69.0 .

Verdict line: 3/3 PASS, independent. Arming.

[github-actions]

§13.11 rung 1 — quorum shadow verdict

S13-SHADOW pr=3655 head=9ee985d416a1c71523c2ea31c56c705c2d947d13 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]

Disarmed by the cop (07:35Z): this PR is integrated in the 0.69 release batch #3669 (one CI run, one queue slot, roadmap.yaml resolved once), at the operator's direction to batch. Its receipt is in the batch tree unchanged. It closes as landed-in-#3669 when the batch merges. Don't push here; changes go to release/0.69-batch.

[noahgift]

Landed in #3669 (squash a877fa056, merged 2026-09-21T10:50:29Z). This PR's receipt stands as the constituent review; the batch folded its commits verbatim. — cop