docs: R-299 — check-figures, the twenty-first guard
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 <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 5
parent
91a104e682
commit
9897d55b67
@@ -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
|
**State:** CLEAN GROUND v0.22, twenty guards green, every desk-reachable branch reached. The
|
||||||
only untested thing left is a run with human beings.
|
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.
|
||||||
|
|||||||
Reference in New Issue
Block a user