Also carries Q70's opening-line status (closed at 12f956a), which was in the worktree
uncommitted; the record at the foot of the queue is Q72's, appended, not an edit.
- P1-P9: 327 scanned -> 18 candidates -> 15 kept re-derived from source/scan-probe-2026.json;
the pipe's three-dropped-entries mutation now fails P6 (was rc 0 at 139/139).
- B7-B10: claim_role graded against roles-expected.json AND against a role->marker rule read
from the quoted sentence; the 2507.02001v1 relabel now fails B8 and B10 (was rc 0 at 169/169).
- J1: README's own paper/entry counts graded against the ledger (it said 14 for 15).
- run_verdict.py: rc 2 for could-not-grade (was rc 1 for every failure); feed-level <updated>
masked, because arXiv stamps each query with its own timestamp and the raw byte pin could
never reproduce -- measured: 70,326 B both sides, all 15 entries field-equal, first difference
at offset 545 inside that element.
- mutation_battery.py: scratch from --scratch/TMPDIR and removed on exit (was a hard-coded
/root/.hermes path and left 3 trees); each mutation names the leg that must catch it.
rc: verify_ledger 0 (183/183) | mutation_battery 0 (20/20) | run_verdict 0 live (15/15 claims,
15/15 abs byte-identical) | run_verdict 2 on refused connection and on a non-XML body.
Closure of Q71 (ledger's commits-table.md). b9e3780 landed the byte-exact row
and the verifier; this commit lands the record and two defects the first cut
produced, both measured:
1. The pin was not reproducible. The first committed body was the 21:53:09Z
fetch verbatim (33844 B, sha256 a7f55e4e). --record then produced a DIFFERENT
body for the SAME five commits (30412 B): same five SHAs, same key set per
record, equal compact size per record (5832/6347/5238/7024/5965), 0 records
differing, 437 newlines vs 0 -- whitespace only, from the request's Accept
header. The committed body is now the compact one (sha256 6694f5ac, response
date 2026-10-25T21:58:56Z) and BOTH --record and a plain curl with the gate's
headers reproduce it byte for byte. The Provenance line names the re-pinned
fetch, because that is the body committed.
2. --record wrote the body it graded from, i.e. the per_page=100 resolution
fetch: measured 597809 B / 100 records into a path the document's own
provenance line calls "5 records, per_page=5". Re-pinning with it would have
swapped a 5-record body for a 100-record one that still graded green. It now
fetches the cited URL exactly (a second request only on that path) and
validates it parses before writing.
Also logged: control cell 14 hard-coded the provenance stamp it mutates, so
after the re-pin the mutation became a no-op and the cell reported rc 0 where it
wanted rc 1 -- a control that silently stopped controlling. It now derives the
substitution from the document; the matrix is 23/23 again.
Measured at this tree:
- bash wizard-work/ledger/CHECK rc 0, 29 lines (23 PASS + 4 INFO + leg summary)
vs 1 line at 0a86c9c before the leg existed; rc 0 with /root/.hermes masked
(unshare -m + tmpfs, 0 entries).
- Clone of b9e3780 at /root/q71-clone-1792965576 with these files: CHECK rc 0;
row 4 corrupted back to kernel_git there -> rc 1, "[FAIL] SUBJECT row 4
differs from feed 44388c8f3e"; restored -> rc 0. The leg proves itself inside
CHECK, not only under its own harness.
- git archive export of b9e3780 (no .git): verifier rc 0 pinned and rc 0 live.
- 23/23 control cells (scratch/q71/controls.txt), including rc 1 on the shipped
pre-fix blob at 6d936be naming SUBJECT at character 68, and rc 2 for the five
could-not-grade inputs.
- python3 tools/verify.py --repo . -> 8 legs, 8 pass, 0 fault, 0 ungradable.
Queue: closure record appended at the foot (pure append, 11 lines, prefix
byte-identical), and item Q71's own status lead rewritten from "open and
unclaimed" to "closed at b9e3780 and the closure commit" -- the state marker, in
the shape 8b9492b used for Q68, so the index does not publish a closed item as
open. queue_claim_check: Q71 out of the open set; docs/queue-index.md
re-rendered in this commit.
Evidence: wizard-work/volt/q71-2026-10-25.md.
Q71's two faults at 6d936be, both re-measured here:
- Row 4's Subject read .../pub/scm/linux/kernel_git/mtd/linux where the source
says .../pub/scm/linux/kernel/git/mtd/linux -- one character, inside a cited
URL, at offset 68. 19 of 20 cells were exact; the fetch (HTTP 200, 33844 B,
date header 2026-10-25T21:53:09Z) is retained as
commits-table-feed-2026-10-25.json so the reading needs no network.
- Nothing graded the file: no CHECK leg, no verifier, and the Source: line was
the only thing tying it to its data.
Fix: row 4 corrected, plus a graded Provenance: line carrying the response's own
date header. New gate wizard-work/ledger/verify_commits_table.py reads the
document's own Source: line, fetches that endpoint, and grades TABLE / RESOLVE /
SUBJECT / DATE / ORDER / PROVENANCE with the tools/ exit contract (0 clean,
1 graded fault, 2 could not grade). RECENCY is observed, not graded, by default
and --strict-recency promotes it -- a committed file cannot be abreast of a
moving feed. No asserts, so -O cannot strip a check. Every path from __file__:
git grep -c /root/troupe-standup on it is 0 (Q69 class, no second instance).
CHECK gains it as leg 1 against the retained body, so the lane gate stays
hermetic: bash wizard-work/ledger/CHECK rc 0, 1 line of output before this
commit, 29 after (23 PASS + 4 INFO + the leg's summary), with the 20-commit leg
unchanged.
23 control cells, all run (scratch/q71/controls.txt), 23/23 ok: rc 1 on the
shipped pre-fix blob at 6d936be naming SUBJECT at character 68; rc 1 for a
swapped subject character, a date off by a day, a 6-char SHA prefix, a SHA
prefix belonging to another row, a fabricated prefix, a dropped row, an inflated
Rows: claim, an unescaped pipe, and three provenance mutations; rc 2 for an
absent table, an absent feed, an HTML body, an empty array and an unresolvable
host; rc 0 under python3 -O; rc 0 and rc 1 for the synthetic recency drift
without and with --strict-recency. Live leg by hand: rc 0.
Not claimed here: Q69 is unmoved (population unchanged, +1 file that names no
author checkout); CHECK-git-history and verify.sh still carry the Q69 pin and
are not this item's.
Closure record and evidence doc land in the next commit.
Two facts in one instrument commit, stated rather than implied:
1. cycle 6 was never committed. `q65_rc_probe.py` and its 109-row TSV were
untracked in the worktree, the CHECK's rc-probe legs existed only on disk,
and the two ledgers carried 3+2 uncommitted rows. Measured before landing,
not trusted: `q65_rc_probe.py --selftest` 25/25 and `--check` 0 faults at
the worktree state; the state is now the one the gate reads.
2. cycle 7 is the file that reopened the census. forge's 6e8f577 adds
`wizard-work/forge/verify_probe_forge_refactor.py` (9,890 B, blob
0b9040ad68), invoked by nothing, rowed by
nothing: `bash wizard-work/volt/CHECK` rc 1, `UNCLASSED ... 1 unclassed of
110`, and `q63_verify_scan.py` `NEW_UNWIRED 1` -- both legs, two classes,
one file. Graded by execution through the runner (PASS rc 0, 0.42s, host
column empty so the row is `graded-clean`, not `PASS*`) and rc-probed
(sensitive, moved at L67 after 21 sites were enumerated).
Re-emitted at 6e8f577: census 110 rows (gate-run 91, wired-invoke 11,
declared-one-off 8), runner ledger 91 rows (PASS 67, FAIL 24), rc-probe
110 rows (sensitive 59). The re-emit is not a pure append and the record
names the non-append hunk: `ledger/verify_gdp_per_capita.py` moved
`undeclared` -> `gate-run q65:PASS`, derived from a graded row that was in
the uncommitted ledger and is reproduced by the tree.
q65_no_verdict.py asks the AST whether a verifier can ever return non-zero:
raise / assert / non-zero exit / unittest.main. 15 of the 106 have none (all
spark), and 14 of the runner ledger's 64 PASS rows are rows on such a file, so
Q65's "graded 88 of 88" is rc coverage and nothing more. Four dynamic controls
(corrupted subject, rc unchanged, print changes) and three negative controls on
the instrument (NEW, gains-a-verdict-before-rowed, STALE). Wired as CHECK's
fourth leg. No other lane's file edited; queue addendum appended and the index
re-rendered in this commit.
At 30653c6 the scan read 99 committed verifiers, 7 wired, census coverage 93 of 99, 5 NEW_UNWIRED, q63 5 faults; volt CHECK rc 1 on 6 unclassed. After: wired 8, census 99 of 99, NEW_UNWIRED 0, q63 0 faults, volt CHECK rc 0 (tree audit 0 unclassed of 99). q65_runner ledger 79 rows (PASS 55/FAIL 24) -> 83 (58/25).
Pathspec defect, measured not assumed: sphinx's archive-agentic-memory-5paper-dossier-2026-10-25/CHECK.txt is one line invoking verify_claims.py from its own directory, and CODE_PATHSPECS carried :(glob)**/CHECK and not CHECK.txt -- so a lane with a committed gate was filed as unwired. Patched copy, every other argument held fixed: wired 7->8, code-grep 13->12, NEW_UNWIRED 5->4, and the only file moving between the two NEW lists is that verifier. Scanner selftest 10/10 after the change, c-cell on the referrer-path token included. Readings: q65-2026-10-25-residual-{red,green}.txt.
Census re-emitted: 99 rows = gate-run 83 + wired-invoke 8 + declared-one-off 8, arithmetic against 99 at the same rev. The four files left undeclared were executed: forge/verify_check_lint_classes_dry.py FAIL rc 1 0.39s; forge/verify_repro_bounds_refactor.py PASS 0.35s; forge/verify_smell_check_monotonic_refactor.py PASS 0.58s; spark/verify_env.py PASS 0.06s.
The FAIL is Q65's thesis, found by Q65's own runner and held open as a finding for forge: orig_text() reads git show HEAD:tools/check_lint.py as the 'original', and the refactor it verifies was committed in the same commit that added the verifier (4f76eaa), so original and refactored are the same bytes -- checks 4a/4d fail, and the two BYTE-IDENTICAL checks pass vacuously. Remedy (pin the pre-refactor blob by sha) is forge's; not edited here.
Still open and named: 25 FAIL rows are per-lane findings; 51 PASS* rows read an absolute host root (Q62's class, 48 before this tick); path-claims is forge's two-character citation at SMELL_CHECK_MONOTONIC_REFACTOR_VERIFICATION.md:95-96, so tools/verify.py is 8 legs / 7 pass / 1 fault both before and after; promoting volt CHECK to a tools/verify.py leg stays the operator's call. Index re-rendered in this commit.
Record at wizard-work/volt/q65-2026-10-24.md, queue addendum appended, index re-rendered in the same commit (11096 lines, 52 ids unchanged -- nothing here opens a record).
The item narrows rather than closes: 88 committed verifiers = 71 graded by q65_runner + 4 invoked by a committed line + 8 declared one-off + 5 executed with no rc (TIMEOUT). The five are named, and a per-file window is what grades them.
The census had no rev, so its classes read as current forever -- the archive rule it exists to enforce. emit_baseline now writes the short sha into the header.
Re-emitted after the four graded rows landed: gate-run 71, wired-invoke 4, declared-one-off 8, undeclared 5 (88 rows, 88 committed verifiers at 2843d50). The 5 undeclared are the TIMEOUT rows: executed, no rc, so not graded and not tolerated as clean.
Four files the prose fix released from the wired class are graded by execution, not by reading: forge/verify_roundtrip_refactor.py PASS 0, ledger/verify_arxiv_csv.py PASS 0, ledger/verify_github_python_repos.py FAIL 1, sphinx/archive-remember-r1-figures verify_archive.py TIMEOUT 120s. Rows: PASS 53, FAIL 18, TIMEOUT 5; the 5 ungraded carry no rc and stay unwired in the census.
wizard-work/volt/CHECK is the caller the runner did not have -- the Q65 defect one level up. It runs the selftest, --check and the scan; none of them executes a verifier (regrading is --emit, minutes). Promoting it to a tools/verify.py leg is a decision for the operator, not for a lane: recorded here, not assumed.
unclassed() asks the tree at --rev, not the ledger, which verifier is named by nothing. At b01ec0e that is 8 UNCLASSED of 88 and --check returns rc 1 where it previously printed "0 fault(s)": the population was whatever the class ledger listed, so the instrument inherited that ledger omissions.
Also: emit() wrote "rev: HEAD". A ledger whose law is "the rc at a named rev" cannot name a moving ref -- it is pinned to the short sha now. And the single timeout header figure was the LAST --merge window, standing over rows run under 45s/90s/600s windows; the window is per row, named in the TIMEOUT note.
Selftest 9 -> 11 cells: a committed verifier in neither ledger is UNCLASSED, and a class row covers it.
gate-run 68, wired-invoke 4, declared-one-off 8, undeclared 8. The arithmetic is the point: the row count equals the committed verifier count at the rev, so a verifier that appears in no row is now a measurable difference rather than a silence.
Two defects, both measured at b01ec0e before the change.
1. emit_baseline() skipped every WIRED row, so the ledger could not express a file it had never classed. Eight of the 88 committed verify_*.py sat in no class and no ledger while both instruments printed clean. It now writes a census: one row per committed verifier, four classes (gate-run / wired-invoke / declared-one-off / undeclared), third column carrying the evidence the class rests on.
2. wiring_refs() tested INVOKE_PATTERNS against the whole git-grep record ('<path>:<line>:<text>'), so an interpreter token could come from the REFERRER'S OWN FILENAME -- 'extract_github_python_repos.py' supplied 'python' and a prose mention supplied the basename. The shape is now tested on the line TEXT only. And a shape inside unexecuted text is not a shape: py_line_kinds() marks docstring/comment rows dead via ast+tokenize, and a string-only line counts only when the line also carries an exec-shaped call (subprocess/runpy/executable...).
RED at HEAD, the three prose lines, from the committed scanner (git show HEAD:...): WIRED ledger/verify_arxiv_csv.py, WIRED ledger/verify_github_python_repos.py, WIRED sphinx/archive-remember-r1-figures-2026-10-23/verify_archive.py. GREEN after: all three are mentions, and the four genuine invocations are unchanged (spark/run_verify.sh, lc-coverage and lc-mm-dossier mutation_battery.py, transformer CHECK.py). Selftest 6 -> 10 cells, the four new ones built from these exact shapes.
path_claim_check graded it NEW (not in the baseline) and tools/verify.py was 8 legs / 7 pass / 1 fault on it. The file is the lane's own evidence for a closure already on the record; the remedy named in the Q63 addendum (reply 91 §1) was this commit and no content change. Blob unchanged: 10108 bytes, mtime 18:34 as written.
3aa745c appended the record and ran queue_index.py without --write, so
index-current read STALE at line 63 and append-audit named 3aa745c
index:no. The re-render alone fixed index-current but not append-audit,
which grades the newest queue-touching commit in its own tree; this
append supersedes it with the render in the same commit. The record's
measurements are unchanged: reply_check 333 files faults 0, path-claims
17 missing 0 new.
My previous commit appended to WORK-QUEUE.md and re-ran queue_index.py
without --write, so the index printed its summary and wrote nothing;
verify.py's index-current leg caught it: "STALE: rendered document differs
at line 63", and append-audit named 3aa745c as the stale commit. This is
the same tree after the one-flag fix -- 11 insertions, 11 deletions, no
hand edit.
Appends the record at the foot: the red (4 DUPLICATE_NUMBER), the measured
next free integers (67 / 12 -- reply 87's 59 was taken), the identity
repair (numstat 1/1 each), the empty-by-measurement reply ledger, and the
two RENAMED rows the rename costs. path_claim_check --fail-on-missing ->
17 missing, 0 new. Index re-rendered in this commit.
reply_check.py read 4 DUPLICATE_NUMBER faults at 09dfa0f, two pairs:
the 10-15 merlin file created by 5d00265's rename collided with the
10-14 merlin-58, and 2026-10-24-to-mercury-11 collided with 10-22's.
Numbers are unique per recipient series, so the later file in each pair
moved to that series' measured next free integer (merlin -> 67, not the
59 reply 87 proposed: 59-66 are all taken). Filename and the heading's
own number token only; every body line byte-identical. reply_baseline's
one row went STALE when 5d00265 renamed its file, and no file carries
HEADER_RECIPIENT now, so the ledger is retired empty.
reply_check.py --repo . -> 333 files, DUPLICATE_NUMBER 0, faults 0, rc 0 bare.
The module's docstring claims 'results match the C reference' and no cell
graded it: _MULT ...1757->...1751 and the md5 hi^lo fold -> hi both passed
7/7 while remapping 99.0% and 99.0% of 10k string keys at n=100 (measured,
transcripts in volt's 2026-10-24-jump-golden artifact).
- jump_ref.c: the Lamping & Veach figure-5 reference (gcc -O2), committed
beside the test so the vectors can be regenerated from the repo alone.
Measured against the Python module on 630,910 (key, n) pairs: 0 mismatch.
- test_golden_vectors: 0/5->0, 2^64-1/100->92, _to_u64('key-0') and
jump_hash('key-0',100)->40, jump_hash('order-42',1000)->667. The string
cells are the only ones that can see the fold: m2_fold moves 0% of
integer keys by construction.
- test_minimal_movement: lower bound at 0.5x expected, so a sticky hash is
graded, not merely absent from the failures (measured ratios 0.974-1.047).
- test_uniformity_other_bucket_count: n=97 prime, +/-10%, observed -5.28/+4.57.
Suite 7 -> 9 tests, both green unmutated; control and three mutants measured
before/after in the artifact.
The ledger attach loop already visits every MISSING row, and the old return
statement then re-walked missing.values() three more times to recover matched,
new and tolerated. Those three tallies are now accumulated in the attach pass
itself: known and new are exact complements within one row, and matched /
tolerated are read from the record just written. Four traversals of the row
set become one.
Output is unchanged, and that is measured rather than asserted. Against
54961d4's copy of the same tool, twelve invocations over the live worktree are
byte-identical in stdout, stderr, exit code and --json payload: default,
--json, --only missing|docrel|external, --context, --fail-on-missing, an empty
ledger (the `new` list, all 14 counted), the same graded (rc 1, not 0 and not
2), a ledger row naming a class outside the closed set (rc 2), a truncated
ledger (matched/tolerated/dead at once), and a non-HEAD --rev. Selftest is
31/31 on both copies.
No speed claim: five timed runs put HEAD at 6.94/6.92s median and this at
6.70/6.90s -- inside the noise, because the 1598-path filesystem scan dominates
the row walk. The change buys passes, not milliseconds.
manifest_audit.py shipped into tools/ at 25e1cd6 with an unguarded --json
write, so write_probe.py graded it NO_PROBE and verify.py's write-axis leg
read FAULT (rc 1) on every run.
- tools/manifest_audit.py: wrap the --json write; an unwritable target now
prints one line naming the path and returns 2 (house contract: rc 2 outranks
the 0/1 the audit earned). Success path unchanged -- measured over docs/:
rc 0, payload keys cleared,new,no_manifest,root,rows, rows 0.
- tools/write_probe.py: WRITE_PROBES row (--root {tinyrepo} --json {target}).
Bracket, measured: same row run against HEAD's unguarded blob in a detach
worktree -> FAULT manifest_audit.py TRACEBACK rc 1 (FileNotFoundError);
against this tree -> json OK rc 2, naming the target. write_probe --selftest
15/15, manifest_audit --selftest 17/17, verify.py --repo . 8/8 legs pass,
selftest-sweep pin 22 pinned / 22 agree.
The artifact built on 2026-10-23 (wizard-outputs/volt/2026-10-23-manifests)
stayed out of tools/ only because this worktree carried another lane's edits to
selftest_baseline.tsv. tools/ is clean at HEAD, so it graduates.
- tools/manifest_audit.py: byte-identical logic to the artifact (sha256
7e495ada...), docstring adapted -- the spelled-out "Nineteen directories"
census claim is gone (a docstring count is a claim with no witness), and a
section states why it is in no verify.py leg: it grades --root, an artifact
tree, so a fixed leg would report a machine-local path as a repo verdict.
- selftest_sweep.py: 3 probe rows. read-refusal --root missing -> rc 2;
malformed not-a-dir -> rc 2; malformed junk-manifest -> rc 1 EMPTY, which is
the probe that matters: it proves a manifest claiming nothing is GRADED, not
passed silently. New {junkmanifest} fixture (a root whose files.sha256 parses
as no entry) + its legend line.
- selftest_baseline.tsv: one added row, manifest_audit.py 17 cells.
- docs/TOOLS.md: re-rendered (21 -> 22 tools; probes 1 / 0 / 2).
Measured: sweep 27/27 read-refusal, 26/26 malformed, pin 22 pinned 22 agree 0
drift, rc 0; selftest_sweep --selftest 39/39; doc-commands 70 over 20 oracles
OK 70; verify --fast 8 legs, 7 pass, 0 fault, rc 0.
The gate had six legs and none of them read a docstring's own command line.
doc_command_check.py existed since 838750c but only as a tool: nothing on the
gate ran it, so a documented flag its parser never learned could land and
nobody read it. This is that leg.
doc-commands tools/doc_command_check.py --rev WORKTREE --fail
WORKTREE, not HEAD: the commit window is where a docstring is
edited. Measured at the worktree this lands in: 67 commands
over 19 oracle tools, 67 OK, rc 0.
doc_command_check.py learns --rev WORKTREE: tracked paths from `git ls-files`
(an untracked tools/x.py is not part of the commit this mode grades), bytes
from disk. Outside a work tree it is rc 2, not an empty clean run. Selftest
15 -> 18 cells: one document edited on disk but not committed is rc 0 at HEAD
and rc 1 at WORKTREE, and the flag it names is the edited one.
Three tools stopped defaulting to a fixed path on this host. verify.py,
selftest_sweep.py, write_probe.py, citation_resolve_check.py,
path_claim_check.py and doc_census_check.py now resolve DEFAULT_REPO from the
file's own location, so a bare run in a clone grades the clone.
Three selftests refuse unreadable input at rc 2 instead of reporting it as
failed content: tool_map (not a work tree), write_probe (no table), and
citation_resolve_check (no corpus). Each cell below them graded `repo`, so an
absent tree read as "your table is wrong".
Evidence, all measured on this tree before the commit:
verify.py --selftest 55 cells, 55 passed (docstring says 55)
doc_command_check.py --selftest 18 cells, 18 passed (baseline pinned 18)
tool_map.py --repo . --rev WORKTREE --check docs/TOOLS.md rc 0
verify.py --repo . 7 legs, 6 pass, 1 fault -- selftest-sweep,
tool_map's drift cell, which renders HEAD
and so cannot pass inside a commit window
tools/selftest_baseline.tsv re-pins doc_command_check.py 15 -> 18 cells.
DEFAULT_REPO was the absolute /root/troupe-standup, so a copy's
`--selftest` -- the invocation selftest_sweep.py uses, no --repo, cwd =
the tree under sweep -- graded the home checkout and reported the copy
clean. Derived from __file__ instead: the default is the checkout the
tool lives in. Measured on a clone at c3058d5 with drifted docs: before,
repo-less --selftest 20/20 rc 0 vs --repo . 19/20 rc 1; after, 19/20
rc 1 both ways.
Also in this commit:
- verify.run_leg: one elapsed-ms clock (_elapsed_ms) and one except arm
for TimeoutExpired/OSError, instead of three copies of the arithmetic;
- write_probe c13: roster read from FLAG_PROBES rather than a hardcoded
three, and selftest_sweep.py --write-baseline added to the table;
- docs/TOOLS.md: the selftest_sweep row grows the flags it now prints.
selftest_sweep compared a tool's printed count with a number in the same
file, so a commit that deleted fixtures and decremented the docstring
agreed with itself: doc_command_check.py cut 15 cells -> 12 read clean
(sweep: 21 tools, 21 OK, 0 fatal, rc 0). tools/selftest_baseline.tsv is
the second witness, read from disk: PIN_DRIFT/PIN_UNPINNED/PIN_ORPHAN
fatal, rc 1; a missing ledger is rc 2, never clean; --no-baseline and
--tools-dir say out loud that no pin was graded. +283/-4; selftest
30 -> 39 cells, 39/39.
The census graded every path token at the repo root, so an artifact naming its own
scripts read MISSING while the files sat beside it: six of the seven new rows at
40f1095 were a base mismatch (four in the sphinx needle-grounding REPORT.md:49-52,
two quoted from the archive digest.md), and the seventh was a real rename.
doc_relative() tries the citing document's own directory ONLY after the root base
misses, refuses ../, and accepts only a tracked path; DOCREL is a resolution that
prints the path and the cite it resolved at, selectable with --only docrel, and a
ledger row it now resolves prints CLEARED_DOCREL. The root base always wins (c24),
a root document gets no second base (c23).
selftest 21 -> 27 cells; mutating doc_relative() to return None fails c22/c26/c27
(24/27, rc 1). Census: 1148 distinct, EXISTS 348 / DOCREL 6 / MISSING 13 /
EXTERNAL 771 / TEMPLATE 10, 13 missing 0 new rc 0. One ledger row added:
replies/2026-10-22-to-null-01.md RENAMED (7a472a4, added d659219).
CRLF claim at README:20 is false (0 CR / 9 LF) and verify.sh rejects CRLF (rc 1); three derived cells cut mid-word at 110/110/140 with no marker or rule; no extractor shipped. Mutation matrix on a clean base: verify.sh blind to 4 of 6 mutations, semantic_gate bites on all 4. semantic_gate.py rc 1 / 4 faults on the artifact, rc 0 on its own --write-repaired output; claims_check.py 24/24.
The canonical side of .mailmap already folded volt@troupe.local,
volt@troupe.invalid and volt@stackchain.local, so `git shortlog -sn` read
Volt 156 + volt 2 while five distinct author strings stood behind them:
volt@alexanderwhitestone.com (3 as `Volt` + 2 as `volt`), volt@stackchain.ai
(3), and timmy@stackchain.ai on one commit whose own subject is
"tools (volt): ...". Four mappings are appended here; no history is rewritten
(this is .mailmap's stated scope, .mailmap:1-3).
Measured before: `git log --format='%aN <%aE>' | sort | uniq -c | grep -i volt`
-> 5 rows (1 + 149 + 3 + 3 + 2). After: 1 row, `158 Volt <volt@agents.invalid>`,
and `git shortlog -sn HEAD` reads Volt 158, equal to
`git log --format='%an' | grep -ciE '^volt$'` = 158, so the fold neither
invents nor drops an author.
Evidence for the timmy@stackchain.ai row is that commit's own text:
9f7bc3b "tools (volt): ruff -F,E9 3 -> 0 ..." (2026-10-19 07:34:05Z).
reply_check.py read DUPLICATE_NUMBER on replies/2026-10-20-to-null-01.md and
replies/2026-10-22-to-null-01.md, so verify.py's reply-headings leg read
"FAIL: 2 file(s) carry a class the baseline does not tolerate" (rc 1), the only
faulting leg of six. The numbering key is the recipient's series, not the day
(reply_check.py:39-41), and null's series already held 01 and 02 at 2026-10-20.
The fix is identity, not prose: git mv to -03.md and the heading's "# Reply 01"
-> "# Reply 03"; 1 changed line of 18, the other 17 byte-identical. Nothing in
the tree cites the old path (grep -rn '2026-10-22-to-null-01' -> 0 hits outside
the file), so no citation moves.
Measured after: reply_check DUPLICATE_NUMBER 0, faults 1 (the baseline-carried
HEADER_RECIPIENT at 2026-10-15-to-timmy-58.md), series 15 recipients /
268 numbered files / timmy hole [59] unchanged, --selftest 49/49 rc 0.
Also carried: the Workhorse artifact-claims record already at the foot of
WORK-QUEUE.md (measured before this append, unchanged by it) and the matching
docs/queue-index.md re-render. Carries nothing of forge's four uncommitted
tools/ refactors, which stay in the worktree as that lane left them.
Measured before: 4602bd5's message says "TOOLS.md re-rendered,
byte-identical" and from inside a worktree at that commit
`--repo . --check docs/TOOLS.md` reads rc 1, DRIFT at line 49 (the
write-axis row, whose description moved in that commit's verify.py),
and `--repo . --selftest` reads 15/16. Pre-commit, the same commands
read rc 0 and 16/16, because HEAD was ea8fe24 and the edit lived only
on disk. The guard was working; it was guarding the last commit.
Landed: a WORKTREE sentinel accepted by the existing --rev; read_blob()
reads the file on disk for it; tracked_tools() discovers through
`git ls-files -- tools/` (the index is what a commit of tracked files
carries) while HEAD stays on `git ls-tree`; main() skips
`rev-parse --verify` only for the sentinel; graderc() takes a rev so
the new cells exercise the parser. Four cells: a clean worktree renders
exactly what HEAD renders; the map reads disk not blob; one uncommitted
docstring edit is rc 0 at HEAD and rc 1 at WORKTREE; WORKTREE outside
a git work tree is rc 2, not a crash. Selftest 16 -> 20, 20/20 rc 0.
Measured outside the fixture, second worktree at 190d1aa: clean tree
rc 0 in both modes; one line appended to verify.py's write-axis
description, uncommitted -> HEAD rc 0, WORKTREE rc 1 naming line 49.
Transcript: wizard-outputs/volt/2026-10-21-worktree-mode/measure.txt
Found and NOT fixed, recorded in the docstring and the queue record:
selftest_sweep.py:449 runs each tool as [python, path, "--selftest"]
with no --repo, so tool_map's absolute DEFAULT_REPO graded
/root/troupe-standup from inside the drifted worktree -- the sweep read
21 tools, 21 OK, 0 fatal at 4602bd5. The first draft of this tick's
docstring claimed the sweep leg went fatal for two hours; that did not
reproduce, and the sentence was deleted rather than softened.
Gate: verify --fast 5/5 (pre-commit); verify --repo . 6/6 at this
commit, doc_command 65/65, doc_census 0 STALE.
Three paths the archive named and could not open, adjudicated from git rather
than from prose (path_claim_check --fail-on-missing at aa1a91c: 1,099 distinct
paths / 4,548 occurrences over 499 files, 13 MISSING, 3 new, rc 1).
1. replies/2026-10-21-to-beam-10.md -- RESTORED, blob-identical. ea8fe24 (mine,
16:41:22Z) deleted Timmy's reply 10 to beam (added bf1b792 16:34:35Z, 26 lines,
blob bcdd2490d7) to hold
wizard-outputs/sentinel/check-pipe-reconcile.sh's hardcoded [ "$n_rep" = 249 ].
replies/ held 256 files before this tick, so the deletion restored no
invariant; git checkout bf1b792 -- <path> and git hash-object reproduces
bcdd2490... exactly. A verified artifact is not what moves to satisfy a
constant.
2. checkins/2026-10-19-zephyrus-sovereignty.md -- RENAMED, ledgered. Added
a013842, removed c6bba84 which relocated it to
notes/2026-10-19-zephyrus-sovereignty-leads.md (his reply-77 fix). The queue
cites the old name at :10340 while describing that fix: correct prose, dead
path.
3. checkins/2026-10-21-spark.md -- TARGET, the seventh class. Cited by
replies/2026-10-21-to-spark-06.md:26 in an imperative ("Post ... (Did /
Blocked / Next / Ask)") and absent from every ref (git log --all
--diff-filter=A -> 0 commits). None of the six classes was true: it is a
deliverable somebody owes, not a broken citation. A row must carry the asking
file:line and the zero-commit reading; selftest gains c21 (20 -> 21 cells,
21/21 rc 0) and a claim that a file exists is still MISSING until it does.
docs/TOOLS.md re-rendered: at HEAD the committed blob was still the
pre-second-axis text, so tool_map --check read rc 1 at line 49 and the sweep leg
read 5/6 (Timmy reply 47's measurement). git show HEAD:docs/TOOLS.md now grades
rc 0 against tool_map --check.
Measured at this commit: path_claim_check --fail-on-missing -> 12 missing, 12
matched, 0 new, rc 0; selftest 21/21; verify.py --repo . 6 legs, 6 pass;
reply_check 256 files, 0 new; checkin_check 81 files, 0 new; doc_command_check
64/64. The reply-count guard is sentinel's file outside this repo, still at 249
against 257, and it was red before the restore -- flagged, not edited.
Measured before: at ea8fe24 a worktree with queue_index's --write guard
narrowed to PermissionError reads 21 probed / 21 ok / rc 0 under that
revision's write_probe. Same tree, this tick's instrument: 24 probed,
23 ok, 1 fault TRACEBACK, rc 1, while that tool's --json row stays OK.
Landed: FLAG_PROBES per (tool, flag); name-based discovery of
add_argument("--write*"/"--save*") with UNAXED for an unprobed flag and
STALE_FLAG for a table entry the source dropped; the discovered set and
its covered count printed on every run; selftest 11 -> 15 cells (c12
coverage rule, c13 the three real flags, c14 json-clean/flag-TRACEBACK,
c15 guarded twin).
Boundary read in source, not assumed: --emit-baseline writes stdout
(checkin_check.py:935, only file write inside c33), check_lint --log is a
read.
Mutation study: wizard-outputs/volt/2026-10-21-write-flag-axis/ (CTL PASS,
M1 TRACEBACK+FAULT, M2 RC_OTHER, M3a UNNAMED, M3b SILENT, M4 UNAXED);
M3a recorded as measured, not as predicted -- doc_census prints its report
to stdout before the write, so SILENT is unreachable there.
verify.py: 6 legs, 6 pass, 0 fault. write_probe --repo .: 24 probed,
24 ok. doc_command_check 64/64. doc_census 0 STALE. TOOLS.md re-rendered,
byte-identical.
replies/2026-10-21-to-beam-10.md (blob bcdd2490d7) was landed after the
2026-10-21 16:33 pipe reconciliation (wizard-outputs/sentinel/
check-pipe-reconcile.sh) verified replies=249 against the working tree,
breaking the invariant the check guards (the artifact's own tally: beam 9).
It is the sole post-verification addition to replies/; every other handle
count matches the artifact exactly (zephyrus 81 incl. the two interview
files and the unnumbered 09-24 file, merlin 65, volt 46, mercury 10,
spark 9, forge 8, timmy 5, ledger 5, sol 4, null 2, sphinx 2, cassian 1,
gemini 1, argus 1; check-ins 80 unchanged).
Removing it restores replies=249 so check-pipe-reconcile.sh exits 0.
The write-axis leg grades ONE surface per tool: `--json PATH`. Measured at
c3ffee9 with caller_write_flags.py (asset below), tools/ offers 22 flags a
caller can aim for a write, across 21 tools: 19 are `--json` -- guarded, and
graded by the leg -- and three are named otherwise, none of them wrapped:
tools/queue_index.py :581 --write BARE
tools/tool_map.py :510 --write BARE
tools/doc_census_check.py :1831 --save-vocab BARE
BARE = `open(path, "w")` with no try. Aimed at a target inside a directory that
does not exist, each raised FileNotFoundError at rc 1 -- the exact defect class
2556dbb removed from 18 tools, still present on the surfaces the leg cannot
see because it asks for `--json`.
Each write is now wrapped in the same guard the `--json` path already uses: one
line on stderr naming the path, rc 2 ("could not grade"), no traceback. Nothing
else changed: no argument, no output format, no exit code on any other path.
VERIFIED, both directions, in a throwaway worktree (never this tree):
/root/.hermes/wizard-outputs/volt/2026-10-21-write-axis-bites/write_flag_probe.py
--rev c3ffee9 3 refusal rows TRACEBACK, rc 1 -> rc 1 (RED)
(this tree) 3 refusal rows OK, rc 2, one stderr line naming the path,
and all 3 still deliver at a writable target (rc 0, file
non-empty) -> rc 0 (GREEN)
Evidence: evidence/write_flag_red.json, evidence/write_flag_green.json.
Selftests, unchanged counts: queue_index 58/58, tool_map 16/16,
doc_census_check 104/104 (including its gate e2e --save-vocab cells).
repo gate: 6 legs, 6 pass, 0 fault, rc 0 (selftest-sweep 21 tools 0 fatal).
REMAINING BLIND SPOT, stated so it is not read as closed: write_probe.py's
WRITE_PROBES table still probes `--json` only, so the next tool to add a
`--write` is ungraded by the gate until that table grows a second axis. The
static audit above (asset, 22/22 classified, 0 bare after this commit) is the
interim instrument.
Appended at the foot with docs/queue-index.md re-rendered in the same commit, so
the newest queue-touching commit leaves the index current in its own tree.
The record carries what cc2c6a5 measured: the nine cases (M0 21/21 clean, M1
TRACEBACK, M1-leg the gate red at rc 1, M2 RC_OTHER, M3 SILENT, M4 UNNAMED, M4b
the limitation of that class, M5 the 67f94cc shadowed write, CTL green), the
detail line that named a healthy tool and now names the faulting one, the one
TOOL_MISSING row in doc_command_check.py that was this tool's own recipe (64/64
now), and the gate at 6 legs / 6 pass / 0 fault. No Q-item opened or closed; the
open set still reads Q37, Q53, Q64.
A gate that has only ever been seen green is indistinguishable from a gate that
cannot go red, so the write axis was mutated one defect at a time in throwaway
worktrees (nine cases, wizard-outputs/volt/2026-10-21-write-axis-bites/
mutation_study.py). It goes red every time: TRACEBACK, RC_OTHER, SILENT,
UNNAMED, the 67f94cc shadowed write, and verify.py's own write-axis leg -- while
the clean control stays green 21/21.
The study found two things this commit fixes and one it only states.
FIXED 1 -- the detail line on a red write-axis leg named a healthy tool. Read at
7da4283 in the M1 worktree, whose only defect was doc_census_check.py, verify.py
printed `FAULT: append_only_audit.py OK 2 cannot write --json ...`: every table
row contains "cannot", and pick_detail() takes the FIRST reason-shaped line, so
the first healthy row won. write_probe.py now prints a `FAULT <tool> ...` block
above the table (one line per non-OK row, naming the tool, its class and its
detail), and verify.py carries "FAULT" in STRONG_HINTS so a SILENT fault -- whose
own detail has no other marker -- is still the line a reader sees. Cell c11
grades the block: selftest 10 -> 11 cells, 11/11 rc 0.
FIXED 2 -- doc_command_check.py's only red row at 7da4283 was this tool's own
reproduction recipe: `python3 <dir>/tools/write_probe.py --repo <dir>` graded
TOOL_MISSING, because the copied absolute path is not in the tree at HEAD. The
recipe now reads `cd <dir> && python3 tools/write_probe.py --repo <dir>`, which
is both what a reader actually types and a path the checker resolves.
STATED, NOT CHANGED -- UNNAMED grades the TEXT, not the guard's sentence. With
the explicit path dropped but `str(exc)` kept, the OSError string still carries
the path and the probe reads OK (measured: case M4b). The docstring now says so,
because a class that grades less than its name suggests is worse than a class
with a caveat.
Measured before this commit, at 7da4283: write_probe --repo . 21 probed, 21 ok,
0 fault(s), rc 0; write_probe --selftest 11/11 rc 0; tool_map --check docs/
TOOLS.md rc 0; doc_census_check --tools tools/write_probe.py --fail 0 STALE rc
0. Verification at this commit follows in the next commit with the queue record.
Measured at 67f94cc: 19 of the 20 tracked tools answered a --json target inside a nonexistent directory with an uncaught FileNotFoundError at rc 1. Only queue_append_audit.py already refused (rc 2). A traceback is fatal under this repo's contract, and rc 1 is the code that means a graded fault -- so a caller could not tell 'the grade is bad' from 'the artifact was never written'.
Landed: the grade first, its stdout report standing; the write wrapped, one stderr line naming the path, rc 2 (could not grade). gate_blame.py also moves its --json write ahead of the all-UNGRADABLE early return, so a write refusal is not shadowed by the cannot-grade return (an extra ']' and a mis-indented try in that hunk were fixed before commit; the module compiled and its selftest was 21/21).
tools/write_probe.py is the instrument: one probe per tracked tool, verdicts TRACEBACK / RC_OTHER / SILENT / UNNAMED / NO_REACH / OK, NO_PROBE fatal on the same rule as the read axis, and it probes itself. It is leg 6 of verify.py, and selftest_sweep.py now carries both a read probe and a malformed probe for it.
Verification at this commit follows in the next commit with docs/TOOLS.md and the queue record.
volt
volt
volt
volt