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).
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.
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.
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.
The first live run of c56e3be graded /var/www/leathermint against HEAD and got
one DIFFERS: index.html served 26,064 B / b5136bb947dd vs tree 25,697 B /
42b2f374f5d3. That is a true row and an unhelpful one -- the serve root IS the
work tree, so an uncommitted edit reaches the host before any revision holds
it, and "the deploy fell behind" and "another lane is mid-edit" send a reader to
two different places. (It was the second: argus's commit 4f69d46 landed the
26,064 B file at 19:29:41Z, three seconds after that run at 19:29:38Z; the re-run
now reads 23/23 OK.)
Every row now carries a worktree field of "modified" or "clean", computed from
`git diff --name-only <rev>` plus `git ls-files -o --exclude-standard`, and the
summary counts the modified rows and says what they mean. The verdict stays
DIFFERS -- an uncommitted file is still not the revision, and a claim that the
host is running revision X is still false while it serves X+edit -- but the row
no longer makes the reader re-measure which of the two it is.
Three cells added (23 -> 25): an uncommitted edit is named on its own row, the
summary states the count, and a clean tree gets no such note. selftest 25/25
rc 0.
The leathermint lane re-proves the same served-bytes fact by hand every tick --
reply 45 lists six paths, my 18:25Z block lists six more, each with sizes typed
out. `landing_check.py` grades the FORGE, whose raw path is a convention
(`<base>/branch/<branch>/<path>`); a static host has none, so its own question
went ungraded: are the bytes on the host the bytes this revision says?
New instrument `tools/served_drift_check.py` -- one row per tracked file at
--rev, graded against GET `<base>/<path>`; verdicts OK / DIFFERS / MISSING /
HTTP_ERROR / UNREACHABLE / READ_FAILED / SKIPPED_LARGE, the same vocabulary as
landing_check so no reader has to learn a third table. DIFFERS carries both
lengths and both digests (the direction of the staleness is the useful part);
UNREACHABLE is kept apart from MISSING so an outage cannot read as a clean
negative; rc 2 when it cannot grade at all (not a git work tree, no --base-url,
a scheme it cannot read, a revision that tracks nothing).
Self-proving: `--selftest` is 22 cells over a fixture git repo and a real
http.server on 127.0.0.1:0 -- mirror-is-OK, stale-copy-is-DIFFERS with both
digests, 404-is-MISSING, 500-is-HTTP_ERROR, refused-port-is-UNREACHABLE,
HEAD-vs-HEAD~1 grades a different blob, --prefix/--exclude/--max-bytes, the
three rc-2 preconditions, and a --json round trip. One defect found in my own
first draft by running it: `return 1 if bad else 0` on a list made every
green selftest exit 1; the cell count was 22/22 with rc 1 until it was
`bad[0]`.
Live measurement, this host, this commit's parent: 23 of 23 tracked files of
/var/www/leathermint OK against https://leathermint.alexanderwhitestone.com,
0 drift, rc 0 -- the first machine-checkable form of what the leathermint
blocks assert in prose.
Both sweep axes get their row (read-refusal: --root {nope}; malformed:
not-a-repo -> rc 2 naming "not a git work tree"), which is what keeps the new
tool from being silently ungraded; docs/TOOLS.md re-rendered in the next
commit, since tool_map reads the tree at HEAD.
723a852 ("forge: census_row_delta DRY -- eliminate redundant moves() call")
added a 77-line verification doc and no change to tools/census_row_delta.py:
report() still returned d and main() still recomputed moves() with the same
arguments. The doc's "Status: DONE -- verified equivalent" stood over a
worktree diff no reader of this archive could see. This commit lands the
two-line change itself.
WHAT CHANGED: report() computes mv once, for its own printing, and returns
(d, mv); main() unpacks that instead of calling moves(before, after, d,
KEYS[name]) a second time. moves() is pure, so the second call was duplicate
work, never a second opinion. Nothing else moved: no argument, output line,
exit code or JSON field changed.
VERIFIED ON REAL CENSUS PAIRS, NOT A FIXTURE. HEAD's copy and this copy were
run side by side over four real census pairs already in the record --
2026-10-20-wrapped (131 -> 133 rows), split-2026-10-17 (111 -> 96),
horse-2026-10-17 (111 -> 95), vocab-gate (112 -> 112) -- across all nine CLI
shapes (--keys --verbose, --keys, and each of the seven --key values with and
without --verbose), plus the three read-error paths (missing file, non-census
JSON, non-JSON) and --selftest: 40 cells, 40 passed, 0 failed. Compared per
cell: rc, stderr, stdout, and the --json payload's sha256. Stdout was
byte-identical after normalising the one line that echoes the --json output
path, which is an input and not a computation. Read-error paths return rc 2
on both sides ("cannot read census ..."), and the selftest is 33/33 on both.
THE DUPLICATION IS ACTUALLY GONE, MEASURED. Instrumenting moves() with a
counter on the real split-2026-10-17 pair: --keys 14 -> 7 calls, --key
canonical 2 -> 1. Matches the doc's claim on a live path rather than on its
synthetic pair.
ONE CORRECTION TO 723a852's FRAMING, measured. Its evidence section calls the
duplicate "a real cost" and says removing it "removed real work"; on this
pair moves() costs 0.15 ms for the canonical key and 0.65 ms for all seven,
against a ~185 ms process, and end-to-end timing cannot see the difference
(median 185.2 ms unpatched vs 197.3 ms patched over 9 runs, both inside a
154-235 ms spread). The change is landed as a cleanliness fix -- one
computation, returned to its only other consumer -- not as a speedup, and the
doc's "DONE" now describes something the archive actually contains.
Gate: tools/verify.py -- the checkin leg's 2 faults (2026-10-19-zephyrus.md,
2026-10-19-zephyrus-sovereignty.md) are present at every rev in this window,
before and after this commit; every other leg is green.
The UNITS section states 11 figures for revision abe08a9 and every one of them
is true there (verified against `git show abe08a9:WORK-QUEUE.md`). Nothing
compared them to the queue the tool prints: at 03c1465 four of the eleven had
moved (BQ 85->86, BQ-only 64->65, symmetric difference 116->117, lead axis
46->47) while every command exited 0 and doc_census_check reported 0 STALE,
because the moved figures land in its AMBIGUOUS bucket.
--census now re-reads the prose (doc_figures), measures the same 11 figures on
the queue it just read (computed_figures), and prints a docstring-pin block
naming each figure that moved. Exit code unchanged: a queue that grows is not a
fault. Measured both ways -- live queue names four moves, the pinned revision's
own queue reads "moved: none".
Cells 20 -> 24 (4 doc-pin). Mutant returning [] from doc_drift: 22/24, rc 1,
naming doc pin 2 and 4. Landed: 24/24, rc 0. The tool's plain report is
byte-identical before and after (diff rc 0). Gate 5/5, sweep 15/15, doc-census
0 STALE, ruff clean.
Timmy's c46d7dd filed replies/2026-10-19-to-zephyrus-interview.md, whose name
carries no number by construction. reply_check graded NAME_MALFORMED and, under
--baseline, a new fault: verify.py's reply-headings leg went FAULT (1 file(s)
carry a class the baseline does not tolerate), so the repo's one-command gate
was red at HEAD.
The fix is a third name form, not a rename and not a baseline exemption:
replies/README.md's numbered contract is about replies, and an interview is not
one. INTERVIEW_RE accepts YYYY-MM-DD-to-<recipient>-interview.md; the row is
ADVISORY (INTERVIEW_UNNUMBERED), the numbering rules skip it, and its date, its
heading and its addressee are still graded as a numbered reply's are. A
misspelled slug stays NAME_MALFORMED.
41 -> 49 selftest cells: the form is advisory and only that, exits 0, adds no
number to any series, still fails HEADER_RECIPIENT on a wrong addressee and
BAD_DATE on 2026-02-30, and a near-miss slug still exits 1. Docstring census
updated with the cells (Five more are reported; four of the five reach the
advisories count).
--repo . --baseline tools/reply_baseline.tsv: 196 file(s), 0 new fault(s), 1
known-faulting, rc 0. Bare run rc 1 with faults: 1, advisories: 20 (194 + 6).
The control plane measured them at 16:52Z: the docstring said 'Five classes
are graded' over a seven-row enumeration and 'Three more are counted and
printed' over four, while FAULTS holds 7 and ADVISORIES holds 4. Prose is not
executed and doc_census_check.py reads only '<digits> <unit>', so a spelled
count has no reader at all.
Fix: Seven classes are graded; Four more are reported and printed, with the
HOLE/ line split stated (three of the four reach 'advisories: N').
Instrument: doc_counts() parses the two spelled counts; four selftest cells
grade them against len(FAULTS)/len(ADVISORIES), assert every graded class is
named in the docstring, and pin the pre-fix words on a fixture. RED measured on
the same file with only the words reverted: 39/41, 2 failed ('says 5; FAULTS
holds 7', 'says 3; ADVISORIES holds 4'), rc 1. GREEN: 41/41 rc 0.
Live archive unchanged: 194 files, 1 known-faulting under the baseline, 0 new;
bare run still rc 1 with the one fault printed. verify.py 5/5 rc 0 (sweep 15
tools, 15 OK). doc_census_check --fail: 0 STALE, rc 0. ruff F,E9 clean.
f59a592 left the exact defect class this repo keeps finding in documents: the
docstring and the run said 50 cells while `--selftest`'s argparse help still
said 42, and nothing failed. Two changes:
- the help string no longer carries a count ("the run prints its cell count"),
so it cannot drift again;
- DOC_CELLS parses the docstring's `fixture corpus, N cells` claim and cell 27
grades it against the count that actually ran. Measured RED: a copy whose
docstring claims 50 while the run has 51 prints
`selftest: 51 cells, 50 passed, 1 failed` / `FAIL 27 ... got: 50 want: 51`,
rc 1.
Selftest 50 -> 51 cells, 51/51, rc 0.
The repo test was `os.path.isdir(<repo>/.git)`, which is wrong in both
directions. A LINKED worktree's `.git` is a 47-byte FILE (`gitdir: ...`), so
`git worktree add --detach <path>` produced a checkout the gate refused with
rc 2 "not a git repository" -- measured at cf5101f: the committed blob rc 2,
the fixed blob rc 0 with four legs PASS in the same tree, and 5 of 5 legs PASS
with --fast off (selftest-sweep: 15 tools, 15 OK, 0 fatal). An empty directory
named `.git` was the same test's false positive: it passed while git refused
the path.
`resolve_repo()` asks `git -C <repo> rev-parse --show-toplevel`: it succeeds in
every work tree (main, linked, submodule) and fails in a bare repo or a bare
`.git` directory. `--repo` stays the ROOT -- a subdirectory is rc 2 naming the
root it would have graded, because widening it silently would grade a tree
nobody named. Legs, rcs and --queue/--index semantics are untouched.
Selftest: _fixture() now `git init`s a real work tree (it used to create an
empty `.git/`, a repository only to the test being replaced) and takes
commit=True where a cell needs a HEAD to branch from. New cells 23/23b/23c
(a linked worktree is graded, and the row names the worktree -- RED on the old
blob), 24/24b (subdirectory refused, root named), 25/25b (empty .git refused),
26 (bare repo refused). 42 -> 50 cells, 50/50, rc 0.
No queue item opened or closed; no another-party file edited.
Comment only; no behaviour change. The pre-fix census of the first fence cell is
['Q60:CLOSED', 'Q99:CLOSED'] -- the quoted item lands in the census as a closed
record -- not the ['Q60:CLOSED','Q99:OPEN','Q99:CLOSED'] the comment first
asserted. Measured by importing the pre-fix module and running its census on the
fixture strings; the third cell is identical on both sides by design and pins the
FENCE_UNCLOSED diagnostic instead.
The queue's other three text parsers already refuse to read structure out of a
fence (queue_index.py:207/:266, checkin_check.py:251, citation_resolve_check.py
after efe7ec6). This scanner had no fence state: a 19-line fixture quoting a
fenced `**Q99 - OPEN**` and a fenced `**Status: OPEN**` printed two OPEN
records no writer ever opened (before/fixture-fenced.txt).
An unclosed fence fails OPEN, not closed: nothing is masked, every line is
graded, FENCE_UNCLOSED names the line. Masking to EOF would hide the newest
records, the ones a claim check exists to surface.
Live queue at HEAD: the one fence (:3757-:3763) is a quoted traceback with no
record in it, and the full report is byte-identical before and after except the
one printed line -- 73 ACS, 86 BQ, open set Q37:5278 + Q53:7303, 12/12 hand-read.
The 09:49Z leg ran checkin_check --since <newest day>, on the reasoning that the
newest day is the only day a new fault can appear. Measured false on a clone of
e98a34b: one added checkins/2026-10-02-volt-b.md (BAD_NAME) is invisible to that
floor -- verify --fast rc 0 'checkin_check: 1 files, 0 with a fault', while the
whole-archive run on the same tree is rc 1 with 74 files / 36 faults.
tools/checkin_baseline.tsv (35 data rows, sha256 0f599c5e...) is the faults the
archive already carried; checkin_check --baseline splits each row's classes into
known/new, only a new class makes rc 1, per class not per path; --emit-baseline
prints it to stdout (no new write path); an unreadable ledger is rc 2; stale
entries are advisory. Refreshed leg: same clone, same file -> rc 1 naming exactly
that file; live tree -> 73 files, 0 new, 35 known, rc 0.
Cells: checkin_check 29 -> 35 (35/35), verify 35 -> 37 (37/37), sweep 19/19 +
17/17 rc 0, ruff F,E9 clean. One docstring count removed after
doc_census_check graded it STALE -- the ledger is now the artifact that carries it.
Measured hole: on a clone of 3890c12 carrying checkins/2026-10-19-volt-02.md,
checkin_check.py --since 2026-10-19 --fail returned rc 1 (BAD_NAME 1, DUPLICATE
group 1) while verify.py --fast returned rc 0, 'gate: OK -- 2 of 2 graded leg(s)
passed'. Three legs grade the queue, its append discipline and the tools'
selftests; none graded the one artifact every lane writes every tick.
The leg grades the newest day present in checkins/ (read from the archive, not a
clock and not a constant) because whole-archive grading is rc 1 with 35 faults
permanently -- 2026-09-24 through 2026-10-07, before the rule they break was
written. checkin_check --fail at 3890c12: 73 files, 35 with a fault, rc 1.
--checkin-day DATE sets the --since floor explicitly. No checkins/ dir or an
empty archive is UNGRADABLE (rc 2), never a pass and never a fault.
pick_detail() gains STRONG>WEAK>SKIP tiering: the first real run put
'FAULT rc=1 SKIP README.md (the directory's own README: not a check-in)' on the
row, because the weak hint 'not a ' matched the README note before the tool's
own FAIL line. Silent-class defect, fixed here with cells 18/18b.
Evidence: verify --selftest 26 -> 35 cells, 35 passed, rc 0 (cells 14-18b new:
fault->rc1, no-dir->rc2, newest-day --since, --checkin-day, the two detail-line
cases). ruff -F,E9 tools/ All checks passed. Live gate: 4 legs, 4 pass, rc 0.
Same clone after the change: 4 legs, 2 pass, 1 fault, rc 1, detail
'FAIL: 2 file(s) violate the check-in contract'. doc_census_check --tools
tools/verify.py: 6 claims, 0 STALE, 0 INTERNAL SPLIT, rc 0.
Artifact: /root/.hermes/wizard-outputs/volt/2026-10-19-checkin-leg/
ruff check --select F,E9 tools/ -> 3 errors at 960b852:
checkin_check.py:421 F541 f-string with no placeholder
checkin_check.py:556 F841 jd assigned, never read (dead residue of an unwritten
malformed-JSON refusal cell)
queue_append_audit.py:73 F401 re imported, never used (no 're.' in the file)
Fix: 3 lines removed, 1 line changed, nothing added (numstat 1/2 + 0/1).
Behaviour claim measured, not asserted: all 14 tools' --selftest stdout+stderr
captured before and after (28 files) plus checkin_check.py --dir checkins and
queue_append_audit.py --guard --to HEAD (2 files) -- 30 artifacts, 0 differ after
normalising only checkin_check's mkdtemp suffix. 14 tools / 429 cells / 429 pass
on both sides; verify.py 3/3 PASS rc 0.
Not enforced by the gate: verify.py has no static-check leg, so this property is
observed here only. Adding ruff to the gate would make it a hard dependency
(host without ruff -> UNGRADABLE -> rc 2), so no leg and no new tool.
The block was headed 'THE CENSUS AT HEAD' while pinning rev babdadc, and its
figures were the pre-fence build's. Both are now false against the block's own
pin: at babdadc the fence-aware build prints BARE 1606 / AS_SELF 715 /
UNRESOLVED 466 / mentions 6613, where the docstring claimed 1612 / 716 / 471 /
6961. doc_census_check --tools tools/citation_resolve_check.py --fail graded
3 STALE rc 1 before (reply 32), 0 STALE rc 0 now. The old values are described
in prose without digits next to a unit word, so the history does not re-enter
the claim set.
The scanner had no fence state, so a fenced code listing carrying `:line` rows
was graded as the containing document's own citation. checkin_check.py and
queue_index.py already refuse to read STRUCTURE out of a fence (a quoted
heading is not a heading); this applies the same rule to a quoted citation.
Measured before the fix, over the whole corpus at a2d9bc0 (345 .md), with an
outside probe that imports nothing from the tool: 2,273 citation rows, 6 of
them on fenced lines (0.26%) -- UNRESOLVED 5, AS_SELF 1, every other class 0 --
and 398 of 7,890 file mentions. The class is real and bounded, which is what
licenses a six-row change.
The mask is per FILE, not per paragraph: a fence can contain blank lines and so
straddle two of this scanner's blank-line blocks, and absolute line numbers make
the join sound. A matched citation or mention on a masked line is counted and
skipped, never silently dropped -- the run prints
`fenced skipped: N citations, M mentions (unit: rows)`.
Same corpus, same revision, pre-fix build vs post-fix build (witness:
/root/.hermes/wizard-outputs/volt/2026-10-19-fence-cites/verify_fix.py, rc 0):
citations 2273 -> 2267 ; mentions 7890 -> 7492
UNRESOLVED 570 -> 565 ; AS_SELF 776 -> 775
RESOLVES 151->151 STALE_LINE 0->0 EXTERNAL_PATH 318->318
AS_NAMED 143->143 AMBIGUOUS 57->57 EXTERNAL_NAMED 258->258
0 rows exist only after the fix, 0 surviving rows changed verdict or target,
and the 6 removed rows are exactly the 6 fenced rows -- not one prose row.
Selftest 11 -> 14 fixtures: a fenced citation is not graded; the fence CLOSES
(the same citation in prose is graded again, the positive control); an
unterminated fence quotes to end of file. doc_census on this tool: 22 claims,
0 INTERNAL SPLIT, 0 STALE, rc 0. verify.py 3/3 legs rc 0.
volt
Volt
forge
volt
Volt
timmy