docs: R-301 — the narrow repair to R-299, and what R-300 got right

A cell inherits the measurement status of its column: a different mechanism
from the shapes rather than more of them, because the document says what the
column is and the vocabulary has to guess how somebody will phrase a figure.

Records the two bugs in the repair, and why the first matters more than the
blind spot it fixed: scanning raw cell text reported ~130 correctly-cited rows
as unheld, and a guard that calls the good material broken teaches its reader
to stop reading the output.

Leaves the merge question open on purpose. Two readers that disagree are how
three of this session's defects were found.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

docs/REVIEW_LOG.md

@@ -7303,3 +7303,63 @@ Both halves mutation-checked: disabling the table-header class fails *"a bold fi
 measurement header is fatal"* and nothing else; accepting an empty waiver fails *"an EMPTY
 waiver is fatal"* and nothing else. End to end, restoring the **24** stops the build with
 both values named. Twenty-two guards green.
+
+---
+
+## R-301 — the narrow repair to R-299, and what R-300 got right
+
+**R-300 was correct and it was correct with a run rather than an argument.** `check-figures`
+reported OK on a bare `**24**` in a table whose own header row reads *Median rounds*, against
+a baseline holding 25. The scan read one line at a time, so measurement status living two
+lines above was invisible.
+
+**The repair is one idea: a cell inherits the measurement status of its column.** That is
+deliberately a *different mechanism* from the shapes rather than more of them — the shapes
+guess at phrasing, and a column header does not have to be guessed at, because **the document
+itself says what the column is.** Where the vocabulary has to anticipate how somebody will
+write a figure, the header only has to be read.
+
+**R-300's general point stands and is now written where the shapes are defined.** A vocabulary
+learned from the marked figures cannot contain the phrasing of the figure nobody marked; the
+defect sits outside the training set by construction. `check-unmarked` attacks that from the
+other end, using **bold as the corpus's own mark of a published figure** — a signal that comes
+from the documents' habits rather than from a guard author's imagination, which is why it
+catches what a shape list cannot.
+
+### Two bugs in the repair, both found before it shipped
+
+1. ⚠ **A citation marker is full of digits and none of them are figures.** `packs
+   line.hollow6.hurt` holds a 6; `fight-tail cut.over15` holds a 15. Scanning raw cell text
+   reported **roughly 130 correctly-cited rows as unheld** — the best-marked tables in the
+   corpus, every one of them fine. Comments are now blanked rather than stripped so every
+   offset still points at the right character of the real line.
+2. **"3.48 of 6"** — the 6 is the party size the mean is out of, not a measurement.
+
+**The first is worth keeping in view.** A guard that reports the correctly-marked material as
+broken is worse than one that misses a defect: the first run produces 130 lines of noise, and
+what a reader learns is to stop reading the output. The blind spot R-300 found cost one wrong
+number; a false-positive storm costs the guard.
+
+**Coverage: 25 figures, 7 marked → 120 figures, 102 marked.**
+
+### Proved by breaking it
+
+| Break | Result |
+|---|---|
+| The historical bare `**24**` in the median-rounds table | fails, at its own line |
+| Strip a marker off a cited prose figure | fails |
+| Plant *"a median 24 rounds"* in prose | fails |
+| Bare figure into a document that cites nothing | trips the ratchet |
+| Threshold column header, *"Past 15 rounds"* | silent |
+| All 102 correctly-cited cells | silent |
+
+Every restore byte-identical.
+
+**Two guards now cover one job**, and that is a decision rather than an accident: `check-figures`
+reads shapes and column headers, `check-unmarked` reads boldness and waivers. They agree on the
+tables and disagree usefully in prose, where one is fatal and the other counts. **Whether they
+should be merged is a question for whoever next has cause to touch either** — the answer is not
+obvious, and two readers that disagree are how three of this session's defects were found.
+
+**State:** CLEAN GROUND v0.24, **22 guards**, 168 cited figures, artwork prompts written and
+nothing generated. Still the only untested thing is a run with human beings.