docs(handoff): #1406 round-1 audit fixes are COMMITTED, and the branch now carries a TRUE merged tree

[ZacxDev]

Interim update to handoff-cairn-oss-multi-instance.md, forced by one sentence that had become false in the dangerous direction.

The correction. The doc said "SIX FILES ARE UNCOMMITTED in that worktree — the round-1 audit fix round, which exists nowhere else." They are now f9ff05c4, and ~/workspace/devrc-flake-pin is clean. A reader acting on the old sentence would treat committed work as at-risk, or re-do it.

The base moved. feat/cairn-flake-pin had drifted 3 ahead / 17 behind origin/main. git merge-tree --write-tree exits 0 — no textual conflict, read by EXIT CODE and not by a marker grep — but main had touched SECRETS.md, three scripts/tests/ files and two skills, all sharing ratchets with that diff, so the merge is in (b79cf63a, origin/main is now an ancestor) and the gate is being re-run on the merged tree rather than on the branch.

Two new investigation blocks, both appended rather than replacing:

1. tekton/devrc-pytests is red on feat(nix): deploy cairn from the PINNED flake package, not a fork in this checkout #1406 at the now-stale head f98be263, naming a test the diff does not touch. Ruled out that the tier is broken (feat(bar): AIO pump + case fan RPM pill, and the staged sudo script that keeps its driver #1411 passed it in the same window, collected=21163 failed=0) and that the other five red devrc SHAs share a cause (four distinct verdict classes, two of them KILLED/BROKEN GATE congestion). Not ruled out and recorded as the honest gap: the run has been pruned, so which assertion fired is unrecoverable — the same shape as rank 6.
2. Round 1's two behavioural findings, with a both-ways mutation matrix. The decoy cairnPackage = pkgs.hello; + cairnUnused = cairn.packages.${system}.cairn; SURVIVED at f98be263 and is killed now; a legal multi-line nix/home.nix header was RED at base with a false diagnosis and is green now. 6/6 mutants killed by their intended guard's own message.

No clawgate-task: field: clawgate_handoff.sh resolve exits 5 with its positive control green, and a wrong session id also answers 200 with an empty array — that is not a clean bill of health.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NTRPN6dhkdf8kZJch3mne5

[ZacxDev]

Closing as superseded, not abandoned — and the reason is the finding.

While this sat open, four commits touched claudedocs/handoff-cairn-oss-multi-instance.md on main, including c756efe0, which makes the same "six files uncommitted is now FALSE" correction this PR exists for. Another session got there first.

main's copy is now 1,189 lines; this branch's is 1,129. Merging would have replaced the newer document with the older one and silently dropped the other sessions' content — the stale-base hazard handoff_doc.py refuses at status=stale-base, arriving here by the back door because a PR merge does not run that check.

Nothing is lost: the content unique to this branch — the round-1 mutation matrix and the Tekton investigation block, neither of which is on main — is re-landed on top of current main in #1425, together with the round-3 and merge outcomes.

⚠ The transferable bit: a docs PR left open across a busy day is a stale-base hazard, not a harmless one. Land a handoff delta in the same session that writes it, or re-derive it against current main before merging.