Re-taken today
docs/theory/IGLA-FORMAL-RESULTS.md P17 publishes a seven-row table of what t27c catalog-gate checks.
| Check |
published |
today |
mandatory-field |
83 |
109 |
widths-partition |
65 |
91 |
gf-closed-form |
21 |
30 |
gf-ratio-optimal |
21 |
30 |
gf-phi-distance |
21 |
30 |
source-agrees |
10 |
36 |
no-spurious-layout |
10 |
the check no longer exists |
no-spurious-layout was replaced by fields-fit-concrete-width in the W603 correction printed directly above the table. grep -rn no-spurious-layout --exclude-dir=target hits only this document and the two W602/W603 reports — never the source.
Three checks the table has no row for run today: emitted-agrees (436), getter-parity, gf-rule-unstated (17).
The findings column
Published: six zeros and one 5. Today the gate reports 3 findings and exits non-zero:
FINDINGS 3
[getter-parity] 109 CATALOG records but 83 getters
[fields-fit-concrete-width] gfternary: bits=2 is concrete, but s+e+m = 3
[gf-rule-unstated] 17 record(s) in cluster=GoldenFloat state no `rule=`
Error: 2 unexpected catalog finding(s) (1 allowed as debt)
A reader auditing the catalog against this table would look for the wrong checks over the wrong populations and conclude it is clean.
The repository contradicts itself in one build
bootstrap/src/main.rs:405 — "Verify specs/numeric/formats_catalog.t27, whose 109 records live in structured comments".
IGLA-FORMAL-RESULTS.md:4250 and :16967 — 83, and 16967's prose was written after the W603 rename, so 83 was carried forward into new text as a live figure.
How it drifted, dated
grep -c 'CATALOG:' at each commit that touched the catalog: 83 at f0d6a0f51 → 92 at 08adcc39f ("register the GF-T ladder in the catalog SSOT, all nine rungs") → 109 at b92872507.
The gf rows moved for a reason, not by drift: since #2792 the phi checks are gated on rule=phi-ratio rather than on cluster=GoldenFloat, so 30 is the population and the 17 GoldenFloat records stating no rule are reported separately.
Proposed
Correct it in the document's own convention — a quoted block beside the original, anchored to a commit, with the W602 reading kept as the record it is. Plus a grep-only test: mandatory-field is bumped once per parsed record with no predicate, so its population is exactly grep -c 'CATALOG:', re-takable with no build. The test also asserts the help string and the document do not disagree about the same number.
Where it came from
A read-only fan-out over docs/, which nothing audits — tri skill claims covers only .claude/skills/*/SKILL.md. 42 figures, 40 presented as current, 13 confirmed, 7 refuted, 0 agent deaths. Each confirmed finding carries a runnable command; the other twelve are a separate to-do.
Re-taken today
docs/theory/IGLA-FORMAL-RESULTS.mdP17 publishes a seven-row table of whatt27c catalog-gatechecks.mandatory-fieldwidths-partitiongf-closed-formgf-ratio-optimalgf-phi-distancesource-agreesno-spurious-layoutno-spurious-layoutwas replaced byfields-fit-concrete-widthin the W603 correction printed directly above the table.grep -rn no-spurious-layout --exclude-dir=targethits only this document and the two W602/W603 reports — never the source.Three checks the table has no row for run today:
emitted-agrees(436),getter-parity,gf-rule-unstated(17).The findings column
Published: six zeros and one 5. Today the gate reports 3 findings and exits non-zero:
A reader auditing the catalog against this table would look for the wrong checks over the wrong populations and conclude it is clean.
The repository contradicts itself in one build
bootstrap/src/main.rs:405— "Verify specs/numeric/formats_catalog.t27, whose 109 records live in structured comments".IGLA-FORMAL-RESULTS.md:4250and:16967— 83, and 16967's prose was written after the W603 rename, so 83 was carried forward into new text as a live figure.How it drifted, dated
grep -c 'CATALOG:'at each commit that touched the catalog: 83 atf0d6a0f51→ 92 at08adcc39f("register the GF-T ladder in the catalog SSOT, all nine rungs") → 109 atb92872507.The gf rows moved for a reason, not by drift: since #2792 the phi checks are gated on
rule=phi-ratiorather than oncluster=GoldenFloat, so 30 is the population and the 17 GoldenFloat records stating no rule are reported separately.Proposed
Correct it in the document's own convention — a quoted block beside the original, anchored to a commit, with the W602 reading kept as the record it is. Plus a grep-only test:
mandatory-fieldis bumped once per parsed record with no predicate, so its population is exactlygrep -c 'CATALOG:', re-takable with no build. The test also asserts the help string and the document do not disagree about the same number.Where it came from
A read-only fan-out over
docs/, which nothing audits —tri skill claimscovers only.claude/skills/*/SKILL.md. 42 figures, 40 presented as current, 13 confirmed, 7 refuted, 0 agent deaths. Each confirmed finding carries a runnable command; the other twelve are a separate to-do.