From 9897d55b671fd6618599b92e4fc673c4c320c02a Mon Sep 17 00:00:00 2001 From: slaguru666 <111923774+slaguru666@users.noreply.github.com> Date: Sun, 13 Sep 2026 19:45:13 +0100 Subject: [PATCH] =?UTF-8?q?docs:=20R-299=20=E2=80=94=20check-figures,=20th?= =?UTF-8?q?e=20twenty-first=20guard?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Records the blind spot (check-cited can only resolve markers that exist), why the net is narrow (an earlier draft found 282 candidates, nearly all prose), the opt-in rule and ratchet, and the two things it found before being wired in — 74.9% printed bare twice, and 'the longest fight is 120 rounds' where 120 is the exact field check-cited refuses to let anybody cite. Two rules each right, and together a hole: refusing the citation while the document printed the number left the least stable figure in the suite as the only one nothing held. Co-Authored-By: Claude Opus 5 --- docs/REVIEW_LOG.md | 67 ++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 67 insertions(+) diff --git a/docs/REVIEW_LOG.md b/docs/REVIEW_LOG.md index 893a5e4..4b1a3a6 100644 --- a/docs/REVIEW_LOG.md +++ b/docs/REVIEW_LOG.md @@ -7149,3 +7149,70 @@ And it lands on the one character whose sheet already says *you have come back t **State:** CLEAN GROUND v0.22, twenty guards green, every desk-reachable branch reached. The only untested thing left is a run with human beings. + +--- + +## R-299 — check-figures, the twenty-first guard + +**The claim R-298 left standing.** `check-cited` resolves every citation marker against the +artifact it names, and has exactly one blind spot, structural rather than a bug: **it can only +resolve the markers that exist.** A measured figure written into prose with no marker beside it +is not a failed citation. It is not a citation at all, and nothing looks at it again. + +**This reads the figures instead of the markers** — the fifth asymmetric counter in the repo, +and the first to run in the opposite direction to the reader it covers: + +``` +check-cited reads MARKERS -> is the figure beside this marker right? +check-figures reads FIGURES -> does this figure have a marker at all? +``` + +**Narrow by design, and the first draft proves why.** Matching any number within forty-five +characters of a measurement word found **282 candidates**, nearly all of them ordinary prose: +*down 140 steps*, *an engineer on his rounds*, *01:06*. A guard that cries wolf 282 times is a +guard somebody switches off. The eight shapes it ships with are the phrasings the documents +actually use when quoting the simulator, **each one read off a figure that is already cited +somewhere** — so the guard learns its vocabulary from the marked figures and applies it to the +unmarked. + +**The opt-in rule.** A document that uses citations has joined the regime and every measurement +figure in it must carry a marker. A document with no markers anywhere has not, and its figures +are held by a **ratchet** — countable, unable to rise. Turning three unguarded scenarios red is +how a guard gets switched off on the day it is written, and the ratchet still means they cannot +get worse. The moment such a document gains its first citation it joins the strict regime whole. + +### What it found before it was wired in + +1. **74.9% printed bare twice** while the same figure is cited correctly four times in the same + document — and that is the number that was published at **58.7%** for three versions until + R-295 found the truncated sweep. The two prints nothing held were prints of the figure with + the worst history in the case. +2. ⚠ **"the longest fight is 120 rounds"** — and 120 is exactly `packs cut.hollow6.longest`, + **the field `check-cited` refuses to let anybody cite**, because a sample maximum moves by a + third of its value on a re-seed. **Refusing the citation while the document printed the + number left the least stable figure in the suite as the only one nothing held.** The two + rules were each right and together they were a hole. The sentence now leans on the guarantee + that is actually strong — `check-packs` refuses to record a sweep in which anything reached + the cap — which is a better sentence as well as a held one. + +**So the guard has a third verdict**, beyond marked and bare: *bare, and the value matches a +field nothing may cite*. It names the field and says **do not cite it — stop printing it.** + +### Proved by breaking it + +Four ways, every restore byte-identical: + +| Break | Result | +|---|---| +| Strip the marker off a cited prose figure | fails, names the line | +| Plant *"a median 24 rounds"* — the original defect | fails, at its own line | +| Add a bare figure to a document that cites nothing | trips the ratchet, 8 → 10 | +| A threshold column header, *"Past 15 rounds"* | **does not fire** | + +**The last row is the one worth keeping.** It *did* fire in the first draft, because the +exclusion tested the matched text while the word that excuses the figure — *Past* — sits +outside the match. A guard whose first output includes a table header is a guard nobody trusts +by its third run, and the fix is in the file next to the rule it fixes. + +**State:** CLEAN GROUND v0.23, **21 guards**, 162 cited figures. Still the only untested thing +is a run with human beings.