Canon / spine lockstep red on main: constitution hash drift since #1041 + rsr spine dogfood red #1094

What is red

Canon / spine lockstep (Gate A, scripts/check-canon-lockstep.sh) fails on standards main itself (5fb9ad19), not only on PRs that wake it. Surfaced by #1088 (run 36737173709, job 109961819824); reproduced locally against a clean checkout of main and a fresh clone of rsr-template-repo at 8256a6e7:

FAIL  constitution  0-canon/constitution/
       declared e0f2c790f01b05bd…
       actual   be48496f7f786d9a…
FAIL  dogfood-gate is 'failure' at spine@8256a6e
GATE A FAILED — the canon and the spine are not in lockstep.

The gate is path-filtered, so main never runs it and the drift stayed invisible until a PR touching the filtered paths woke it.

Cause (measured)

1. Constitution hash drift. canon.lock was last bumped by release(canon): canon-v2.1.1 — ADR-0003 ratification (lifecycle flip) #861 (canon-v2.1.1, 2026-09-19). 0-canon/constitution/ was then edited by docs(canon): open CHANGE-PROCEDURE step 1 for "Always leave it working" (PROVISIONAL — merging opens the proposal, does not adopt it) #1041 (2026-09-23, CHANGE-PROCEDURE step 1 for "Always leave it working") without a matching canon.lock version bump and hash rewrite.
2. Spine dogfood red. rsr-template-repo main at 8256a6e has its dogfood gate in failure. Related: julia-library archetype: children cannot go green (invalid actions.lock ⇒ startup_failure, linter, allowlist, canon lockstep) + arrival pack still CLADE-only rsr-template-repo#211 (archetype children cannot go green, canon lockstep among the listed causes).

Acceptance criteria

• canon.lock carries the current 0-canon/constitution/ hash under a bumped canon version, per the canon CHANGE-PROCEDURE (not a bare hash rewrite).
• The rsr-template-repo dogfood gate is green on its main, or the lockstep gate's spine pin moves to a spine commit whose dogfood gate is green.
• bash scripts/check-canon-lockstep.sh --canon . --spine <rsr-template-repo checkout> --base origin/main exits 0 on standards main.
• Positive control: the same command, with one byte of 0-canon/constitution/ changed and canon.lock left alone, exits non-zero naming constitution.
• Follow-up decided: should this gate also run on push to main (or on a schedule), so drift is caught at the commit that causes it instead of by the next unrelated PR?

Per the standing stopping rule this is an issue, not a blocker for #1088.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QYY8Gp4v4x2J7iSNn1vZ57