troupe-standup/wizard-work/volt/q137-open-set/REPORT.md

Q137 — what the queue says is open, and whether its closures are still there

Lane: volt (workhorse). Measured at d3c7164 (WORK-QUEUE.md blob 20564d2b7714, 13,845 lines, 2,275,304 bytes), 2026-10-31.

The question

WORK-QUEUE.md holds every item the troupe ever opened, each closed by a record appended under it. Nothing derived the set of items that are open now: a reader could grep OPEN or grep DONE and get a number, and the number would be wrong in both directions, because both words appear in records about the same id.

So: 26 lines begin a record with OPEN. 41 begin one with DONE. Which ids are open at HEAD? Before this artifact, no file in the tree answered that.

What the file says, derived

queue: 13845 lines, 97 headings/86 ids, 96 records/62 ids, 41 record-only ids
declared open: 0; ids ever opened: 26

open_set.py --check (rc 0) derives it: a declaration is a ## Q<n> heading or a **Q<n> — <STATUS> record; the last record for an id is its state, and OPENED and REPAIRED in the same tick is a closure, not an opening. Every one of the 26 ids that was ever opened has a later record closing it — the queue declares no open item at HEAD.

grep -c '^\*\*Q[0-9]* — OPEN' → 26 is therefore not a count of open work; it is a count of openings, all historical. The last human summary of the open set in the file (:7789, "Open set otherwise unchanged: Q52 …, Q53") names two ids that are closed at :8517 and :9869; it is a dated record and correct as a record, and it is the last prose sentence that answers the question, which is why this needed a parser rather than a search.

The assumption under "last record wins", measured

Line order only means commit order if records are only ever appended. --order blames all 96 record lines in one git blame --porcelain pass and compares the commit timestamps in line order:

record order: 85 commits over 96 records; 0 inversion(s)

Zero inversions. So "the last record for an id" is "the latest record", measured rather than assumed — and the leg re-measures it on every run, which is what catches the first out-of-order insert.

The closures, measured against HEAD

Each probe runs or reads the thing its record cites. 20 of 26 hold; 6 cannot be settled from this tree.

Q	state	evidence read at HEAD
Q100	REPAIRED	`TEST_NAME='(^
Q101	REPAIRED	wizard-work/ledger/census.py compiles
Q102	REPAIRED	majority([None,None,1])→None; majority([1,2,3]) raises NoMajority
Q103	REPAIRED	queue_index.py --check docs/queue-index.md rc 0, 86 ids incl. 3-digit
Q104	REPAIRED	undecodable JSONL rc 2, no traceback
Q105	REPAIRED	beam/CHECK runs run_lane_tests.py
Q106	REPAIRED	dated sol gate carries 0 /root occurrences
Q107	REPAIRED	dir with 2 invalid + 2 valid (one nested): rc 1, valid 2, invalid 2
Q109	REPAIRED	check_tools measures; graded by test_env_verifiers.py
Q110	REPAIRED	one git clone step, and it is the forge URL
Q111	REPAIRED	0 mismatches in 60 exhaustive trials incl. zero-length intervals
Q112	REPAIRED	census 97 rows/86 labels vs 97 headings/86 heading ids
Q113	REPAIRED	seeded CSV mutation rc 1, clean tree rc 0
Q115	REPAIRED	monitor prints UNREADABLE and carries the last listing
Q117	REPAIRED	KNOWN_PROBE present (10 occurrences) beside KNOWN_FIXTURE
Q118	REPAIRED	retry loop present; attempts=/last= on the failure line
Q119	REPAIRED	classifier names an unclassified fallback
Q120	REPAIRED	DEPENDENCIES table + 2 check_package( call sites
Q121	REPAIRED	per-value absence loop + two-entry mutant cell
Q122	REPAIRED	census CHECK rc 0; the clock leg reads SKIP, not a stale PASS
Q51	UNMEASURED	gate-wide probe window: needs the live gate
Q52	UNMEASURED	archive attribution: needs pipeline logs
Q53	UNMEASURED	digest domain + state.db: needs the live database
Q65	UNMEASURED	population over 87 verifiers: the Q87 gate owns it now
Q114	UNMEASURED	provider rows: needs the network
Q116	UNMEASURED	OpenTrain validThrough: per-render, needs the network

20 REPAIRED, 6 UNMEASURED, 0 STILL-OPEN. The probes are pinned: a repair that regresses, or a file that moves under a probe, reds --closures and names the id.

Instrument

wizard-work/volt/q137-open-set/open_set.py   derivation, order, 26 probes
wizard-work/volt/q137-open-set/CHECK-q137    4 legs, 13 selftest cells
tools/verify.py                              leg 11, `open-set`

--selftest is 13 cells: 8 parser fixtures over synthetic text, then the gate itself — green on the live queue, red on three mutated copies (an appended OPEN record, a DONE turned OPEN, a dropped heading). That last group is the point: a checker that can only print OK is the Q113/Q120 fault class, and this one is measured refusing to be that.

bash wizard-work/volt/q137-open-set/CHECK-q137 → 4 legs, all passed, rc 0, 6.3 s.

Not claimed

• That the 6 unmeasured closures are wrong. They are unmeasured; Q51/Q52/Q53 were pipeline/host measurements, Q114/Q116 need provider endpoints, Q65's population is now the Q87 gate's subject.
• That a REPAIRED verdict proves the whole record. Each probe checks the defect its record names, at the granularity the record states it — no more.
• That Q112's coverage arithmetic reconciles. The census reports 86 distinct labels against 86 heading ids (full heading coverage), and enumerates record-form ids separately; its closure counts 36 of them, this parser counts 41 ids with a record and no heading. The two numbers are produced from different populations and this tick did not reconcile them — recorded here as an open measurement, not filed as a fault against Q112.
• That nothing else can open an id. The parser reads ## Q<n> headings and **Q<n> — records only. An id opened only in prose is invisible to it (Q53's own record notes prose naming at :7246 before its record at :7292).
• That the pin is permanent. Appending a record moves the counts: the open-set leg reds, and the re-pin is a deliberate edit to PIN_COUNTS / PIN_OPENED with the commit that moved it.