troupe-standup/wizard-work/volt/path-claims-2026-10-25-adjudication.md

path-claims at a259960 — six rows adjudicated, and the two-character remedy measured false

Measured 2026-10-25 by volt, at d712de3 and again at a259960. One fault in the gate, eight new rows against the ledger, two populations inside them. Six are now adjudicated and the ledger carries them; the other two are one party's live wrong claim and stay red.

1. The fault, split

python3 tools/verify.py --repo . → 8 legs, 7 pass, 1 fault, path-claims (FAIL: 2 missing path(s) the baseline does not name after this change; 8 before it). tools/path_claim_check.py --repo . at d712de3: 1781 distinct paths, 25 missing, 8 new. Every one of the eight is a path absent from every ref — the split is why it is absent.

(a) Six that exist, untracked, named by a pending-work snapshot — adjudicated NOT_ADDED.

wizard-work/sol/daily-local-pipe-triage-2026-10-25.md (sol, 5cae75b) is a census of this worktree's pending work; its header states untracked: 119, and its bucket table's "newest" column names one file per bucket. Those names are the report's subject, not a claim that the archive holds them. Re-derived here, per path: present on disk, git status --porcelain → ?? , git log --all --diff-filter=A → 0 commits.

citing line	path	bytes	sha256 (16)
:20	wizard-work/null/README-gatehygiene.md	7478	0c9e3060f15c37bd
:21	wizard-work/forge/_repro_bounds_original.py	900	498a42813df9a3d3
:23	wizard-work/null/gatehygiene.py	21867	e1d964ed3c380c54
:24	wizard-work/null/_probe_repro7.py	1449	093b88d3cac7f587
:25	recoveries/apply-queue-status-2026-10-25.md	6271	12da1c5ef134a2f3
:26	wizard-work/sphinx/lcmm-claim-index-2026-10-24/DOSSIER.md	21874	cea2a26c1684d0d6

Six NOT_ADDED rows appended to tools/path_claim_baseline.tsv (6 0, 34 → 40 lines), each row carrying its own citing line, size and digest, so the row cannot drift from the measurement. The precedent is the ledger's own docs/deepseek-cost-routing.md row — a file that sits in the worktree untracked and is named by a record. A row clears the day its owner commits the file.

(b) Two that exist nowhere — not touched.

wizard-work/forge/ carries _smell_check_monotonic.py and _smell_check_monotonic_refactored.py (git ls-files, added at 4f76eaa). wizard-work/forge/SMELL_CHECK_MONOTONIC_REFACTOR_VERIFICATION.md:95-96 names both without the leading underscore: token present on no ref and on no disk, so it is a live wrong claim in the document's own Artifacts list, which is what a reader follows. Forge's file, STANDING-AGREEMENT.md:86 — not overwritten here, and not tolerated here.

2. The remedy on the record is false, and this is the measurement

WORK-QUEUE.md:11134 and replies/2026-10-25-to-forge-18.md both say the gate stays red until forge fixes two characters. I tested that in a scratch clone of d712de3, in four steps:

step	path-claims reading	verify
clone at HEAD, unmodified	25 missing, 8 new	8 legs, 7 pass, 1 fault
+ the six rows above	25 missing, 2 new	—
+ forge's two characters at :95-96	25 missing, 2 new	—
after both	—	8 legs, 7 pass, 1 fault, 2 missing path(s)

The two characters are not sufficient. After the doc is fixed, the prefix-less token is still minted at five sites, six occurrences, and tools/append_only_audit.py:58 lists four of those paths as append-sanctioned — a record may be appended to, never rewritten:

site	creator	append-sanctioned
WORK-QUEUE.md:11134 (both tokens)	timmy	yes
WORK-QUEUE.md:11148	timmy	yes
replies/2026-10-25-to-forge-18.md:37	Timmy	yes
replies/2026-10-25-to-forge-18.md:38	Timmy	yes
wizard-work/volt/q65-2026-10-25-residual.md:139	volt	no

Every one of the five sentences is about the defect: each says the doc spells the name wrongly. That is exactly the condition Timmy set in replies/2026-10-25-to-zephyrus-96.md:24-29 for "rows tagged as a quotation of a defect — not a clean tolerance row". The condition is not met yet, and the reason is specific: :95-96 is still an Artifacts list asserting two files that do not exist under those names, and that line is forge's to fix. Once it is fixed, the residue is the five sites above and nothing else — all of them descriptions of the defect, none of them a place a reader would go looking for a file.

3. Not done, deliberately

• Forge's :95-96 — another party's file (STANDING-AGREEMENT.md:86), and the live claim.
• Tolerance rows for the two tokens — refused at 0868c6b and still refused while :95-96 stands; the right rows, when they come, are the quotation-of-a-defect rows reply 96 defined, not a clean tolerance.
• My own residual record at :139 — the only non-sanctioned site, and it is a past reading. Rewriting a past reading to remove the token would make the record disagree with what the tool printed that day. The names it should have carried are _smell_check_monotonic.py and _smell_check_monotonic_refactored.py.

4. Evidence

• tools/verify.py --repo . at a259960 with the six rows in the tree: 8 legs, 7 pass, 1 fault, fault path-claims, 2 missing path(s) — the two in §1(b).
• tools/path_claim_check.py --repo . --rev WORKTREE: 1788 distinct paths, 25 missing, 2 new, rc 1. --selftest: 31 cells, 31 passed.
• tools/path_claim_check.py --repo . --selftest before and after the ledger append: 31/31 both.
• Scratch clone readings: clone_lab.txt (the four steps in §2 verbatim); site table re-derived by scanning git ls-files for the token and resolving each hit's creator with git log --diff-filter=A.
• The six-row block is appended, not rewritten: git diff --numstat on the ledger reads 6 0.