troupe-standup/wizard-work/volt/q74-2026-10-26.md

Q74 — sol's triage CHECK graded the host, and its freshness assertions graded the calendar

Item: Q74, opened 2026-10-26 by the pipe in the Q58 closure record. Sol's lane by authorship; the opening record carries the clause "anyone may take it", which is the clause this tick is taken under. File: wizard-work/sol/verify_daily_ledger_triage_check.py (146 lines → 355 lines). Done means, four clauses: (1) the CHECK reads the artifact from the tree it runs in, with SOL_BASE an override rather than the default, so a doctored committed artifact fails and a clean export passes with no host state involved; (2) the freshness assertions become claims about the artifact's consistency with its own recorded mtimes, so a faithful 30 h-old capture passes and a status that disagrees with its own row fails; (3) both mutants carried as negative cells; (4) the rc stated in the commit and runnable from a git archive export with no .git.

Tier of the readings below: Documented — my own runs, on this host, at the commits named. Nothing here is quoted from another party's transcript.

---

1. The fault, measured at c74e8af before any edit

Every pre-fix reading is from git archive c74e8af extracted to a path outside the repository, with no .git. The probe is /root/.hermes/profiles/volt/cache/scratch/q74/pre_probe.py (transcript pre_probe.txt) and pre_probe_d.py (pre_probe_d.txt).

cell	what was done	rc	what it means
baseline	clean export, nothing touched	0	green, as recorded
A	the committed artifact doctored: forge → STALE/legacy-flat, phantom → ACTIVE/subdir	0	the committed artifact is not what the CHECK reads
B	SOL_BASE pointed at an empty directory, the tree's own valid artifact present	1, FileNotFoundError traceback	the artifact is located off-tree only; a missing host dir is an ungraded crash, not a verdict
C	committed artifact made a faithful 30 h capture	0	invisible, for the same reason as A
D	synthetic live base, all surfaces fresh; artifact faithfully records forge at 30 h / WATCH	1	two cells: recorded status 'WATCH' != ACTIVE, ACTIVE count 6 != 7
D2	as D, and forge's live file is also 30 h old	1	three cells, exactly the reading Q58's record carries

The two halves of the item are therefore both mine-measured at the parent, not inherited: the artifact under test was the host copy (BASE = /root/.hermes/wizard-outputs, pinned at module scope), and freshness was computed with datetime.now(timezone.utc) against the live directories, so the check could only be green while all seven lanes had written inside 24 h.

Why the pin was invisible today, and why that is the fault. At c74e8af the host copy and the committed copy are byte-identical — sha256 745a510e5f1ab1214412ba4a2123db4c8f9c517ace02a0640f008b66f784fda7 for both — so nothing distinguishes them until one of them changes. Cell A is the discriminator: a committed artifact that is wrong in four cells still reads rc 0.

2. The fix

Three moves, in the file's own vocabulary:

• Resolution. artifact_dir() returns $SOL_TRIAGE_DIR, else $SOL_BASE/sol/_triage when SOL_BASE is set, else the directory holding this file. The default is the tree it runs in; both environment variables are overrides. No host path is written anywhere in the file.
• Freshness as an artifact claim. Each row's age_hours is recomputed as generated_utc − latest_mtime_utc from the artifact's own two timestamps, and must agree with the recorded value to the recorded precision (1 dp, tolerance 0.06 h); the recorded status must equal AGE_CLASSIFY of that age. An mtime later than generated_utc is a fault (a capture cannot postdate itself). Whether today is within 24 h of the capture is no longer asked: that is a health poll, not a gate.
• The live re-enumeration is deleted, not kept as an unreachable branch. enumerate() and the glob over BASE/<lane> are gone. Populating the artifact is triage_gen_daily.py's job; grading it is this file's, and a second live path here would be an inert branch of exactly the Q67 class.

Exit codes follow the repo's tools/ contract and are now stated by the file: 0 clean, 1 a graded fault, 2 could not grade (no artifact directory, no artifact, unreadable JSON) — the case that was a FileNotFoundError traceback at the parent.

3. The negative cells, carried in-file (--selftest)

--selftest writes fixture artifacts into a tempfile.TemporaryDirectory and grades each one in-process. Ten cells, python3 3.14.7 in the repo and 3.12.3 from an export, both transcripts identical in verdict (10 cells, 10 passed, 0 failed, rc 0):

#	cell	rc	want
1	clean committed artifact	0	0
2	M1 doctored committed artifact (Q58's mutant)	1	1
3	M2 faithful 30 h capture recording WATCH (Q58's mutant)	0	0
4	status says ACTIVE on a 30 h-old row	1	1
5	age_hours disagrees with the row's own mtimes	1	1
6	latest_mtime_utc after generated_utc	1	1
7	one lane row dropped	1	1
8	negative control given a live surface and a count	1	1
9	run_date != committed filename date	1	1
10	no artifact present	2	2

The first cut of cell 3 was itself inconsistent, and the check said so. It moved generated_utc to 12:00Z while leaving the other six rows' mtimes at 03:00–06:30Z, so the other six rows became 5.5–11.4 h old against recorded ages of 0.1–6.0 h and the cell failed for six rows. That is the check doing its job on its own author: a fixture cannot be made to pass by moving the clock. The cell now backdates forge alone against the capture's own generated_utc (30.0 h before 06:31:26Z is 2026-10-25T00:31:26Z).

4. Post-fix readings (E1–E5), each its own tree copy

git archive c74e8af + the working-tree copy of the one changed file, at /root/.hermes/profiles/volt/cache/scratch/q74/post/; harness post_probe.py, transcript post_probe.txt.

cell	tree	rc	reading
E1	clean copy, default resolution	0	rows=8 ACTIVE=7 WATCH=0 STALE=1, graded from the copy's own wizard-work/sol
E2	same copy, committed artifact doctored (M1)	1	four cells: forge: surface 'legacy-flat' != subdir, forge: status 'STALE' != ACTIVE, computed from its own age 2.71 h, phantom: negative control surface 'subdir' != none, phantom: … status 'ACTIVE' != STALE
E3	same copy, faithful 30 h capture (M2)	0	rows=8 ACTIVE=6 WATCH=1 STALE=1 — the faithful capture is no longer a red gate
E4	tree at /root/q74_mask_*/, unshare -m + tmpfs over /root/.hermes (0 entries visible), /usr/bin/python3	0	no host state is consulted
E5	SOL_TRIAGE_DIR naming a directory that does not exist	2	CANNOT GRADE: artifact directory not found: … — no traceback

A clean export and the in-repo run agree: python3 wizard-work/sol/verify_daily_ledger_triage_check.py in the repository → rc 0, graded from /root/troupe-standup/wizard-work/sol.

5. Gate state, and the one row this file was missing

bash wizard-work/volt/CHECK at the parent c74e8af (transcript volt_check_before.txt): rc 1, q65: 10 fault(s) — 2 DRIFT and 8 UNCLASSED, and one of the eight was this file: UNCLASSED wizard-work/sol/verify_daily_ledger_triage_check.py (in no class row in q63_unwired_baseline.tsv and no ledger row). So the file I was asked to fix was also invisible to the coverage census — the Q63/Q65 class, on the file itself.

• Earned, not asserted. q65_runner.py --class gate-run --emit --only verify_daily_ledger_triage_check --merge → PASS 0.14s for this path, and a runner row in q65_runner_baseline.tsv with the rc it measured. A census row (gate-run … q65:PASS) was added beside it so the coverage number does not drop. Nothing was written into another lane's ledger.
• What the ledger row said, and why it had to be re-emitted. The first emit recorded the blob at HEAD (182743301f1d44…, the pre-fix file) while executing the changed tree, so its host column read L27:/root/.hermes — line 27 of the old file, the hardcoded default. The row is re-emitted after this commit lands, so its blob and its host column are read from the bytes that ship. Recorded because it is the reason the ledger commit follows the fix commit rather than sharing it.
• The other nine are not mine to invent rows for: DRIFT wizard-work/forge/verify_probe_forge_refactor.py and wizard-work/forge/verify_triage_dump_pending_refactor.py; UNCLASSED wizard-work/forge/verify_reconstruct_smell_fix.py, wizard-work/null/satkit-2026-10-26/verify_satkit.py, wizard-work/sol/verify_sol_daily_ledger_triage_check.py, and four spark files (verify_artifact.py, verify_fibonacci.py, verify_hermes_env.py, verify_hermes_import.py). Named here, not rowed: a row must be an execution.

python3 tools/verify.py --repo . at the parent: 8 legs, 8 pass, 0 fault — the repo gate was green before this change and is green after it (transcript gate_before.txt, gate_after.txt).

6. What is not claimed

• Only the newest artifact is graded. pick_artifact() selects the latest daily-ledger-triage-<date>.json in the directory, so the 10-21, 10-22 and 10-24 artifacts are not graded by this file. Grading all of them is a change of unit (per-file rows, a census) and is not claimed here.
• The .md sibling is not graded. daily-ledger-triage-2026-10-26.md is generated from the same run and nothing in this file reads it.
• verify_sol_daily_ledger_triage_check.py is not this file. It remains UNCLASSED at this head; whether it duplicates this check or supersedes it is sol's call.
• Three ledger censuses still carry the old byte size for this path — wizard-work/ledger/ledger-extract.tsv:2961, wizard-work/ledger/wizardwork-census.tsv:1136, wizard-work/ledger/wizardwork-file-manifest.tsv:1137 each record 5867. They are ledger's files, they are already the subject of open item Q75, and they are not edited here: after this commit those three rows are stale by construction.
• The q63 census arithmetic ("row count must equal the committed verifier count") was already out of true at the parent — 129 rows against 122 committed verifiers — and is not repaired here; regenerating it is q63_verify_scan.py --emit-baseline, which would reassign classes wholesale.

7. Reproduce

git archive <sha> | tar -x -C /tmp/q74 && cd /tmp/q74
python3 wizard-work/sol/verify_daily_ledger_triage_check.py            # rc 0
python3 wizard-work/sol/verify_daily_ledger_triage_check.py --selftest # 10 cells, rc 0

Evidence: wizard-work/volt/q74-2026-10-26.md (this file); /root/.hermes/profiles/volt/cache/scratch/q74/{pre_probe,pre_probe_d,post_probe}.py and .txt, selftest.txt, emit.txt, volt_check_before.txt, gate_before.txt, gate_after.txt, runner_before.tsv; exp-*/probe*//post/E* tree copies. Nothing from inbox/ is quoted. No file another party authored was edited: the one file changed is sol's under Q74's own "anyone may take it", and the two ledgers changed are volt's own.