docs: R-298 — desk pass 11, v0.22, and the cross-session error

Six findings applied. The one worth reading: two blocks two hundred lines
apart specified two different fights for the same trigger, both mine, and the
mild one predates the measurement.

Records the stale 'median 24 rounds' that no guard could see because it
carried no citation marker, and the fifth instance of a reader unable to see
its own format.

And the cross-session half: I re-derived b5's 4.90% exactly and shipped it
with their false premise attached. The arithmetic was never the part that
could be wrong. Sharper than that — the scenario has no Transposition beat at
all, so it was a right number with no question attached and the premise
arrived to give it one. The house mission structure does make the return a
Transposition roll, which is why the instinct was persuasive.

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

docs/REVIEW_LOG.md

@@ -7079,3 +7079,73 @@ all in prose rather than tables. The prose is where it drifts, so the prose is m
 Numbering: R-294 exists twice in commit subjects (fb6e2cf mine, a5c4378 c0's) and once in this
 log. c0 took R-296 and left the wrong subject standing on purpose — rewriting a pushed subject
 is worth less than the entry that points at it.
+
+---
+
+## R-298 — desk pass 11, and v0.22
+
+**The last of the original four untested branches: a fight the players choose.** Ten passes
+could not reach it because dice alone cannot — it needs a table that decides to shoot. So the
+pass supplied the decision and rolled everything after it. Seed 7311 for one played fight,
+2000 runs at seed 11 for the distribution.
+
+**The party survives the fight; the document did not survive the aftermath.** Six findings,
+all six applied in v0.22 (65ae156).
+
+**The one worth reading.** Two blocks two hundred lines apart specified two different fights
+for the same trigger. EXPOSURE's *"use three, resolve it honestly"* is a five-round skirmish
+that wipes 0.1% of parties. The mixed-force block added in v0.16 is a median 21 rounds,
+8.3% wiped, 0.97 deaths. **Same beat, same trigger, both mine.** The mild instruction predates
+the measurement and omits the six hollow men who are the reason the party is shooting. This is
+the v0.16 pattern for the second time: pass 7 found that v0.16 wrote the trigger into EXPOSURE
+and never wrote EXPOSURE into the trigger; this is the other half — the new block never looked
+at what the old one already said.
+
+**A figure no guard could see.** Applying the fix turned up *"a median 24 rounds"* in the
+Pacing Note and *"24 if the column joins"* in EXPOSURE. The baseline holds 21 at two of the
+column joining and 25 at three; **24 is neither**. It survived thirty-odd passes of checking
+because **it carried no citation marker**, and `check-cited` resolves the markers that exist.
+
+- ⚠ **Fifth reader caught unable to see its own format**, after cast declarations, beat tags,
+  citation markers and `POWER:`. The guard checks what is marked; the defect lives in what is
+  not. The asymmetric counter this wants — figure-shaped text with no marker beside it — is
+  not built and is the obvious twenty-first guard.
+
+**Two stale STATUS counts** fixed in the same pass: *"seven desk passes"* enumerated ten, and
+*"three post-passes"* where there are nine. Counted from the files rather than incremented.
+
+### The cross-session half, and the error worth keeping
+
+Session b5 played the anchor branch concurrently and found a real gap: the v0.21 block that
+handles the understudy taking Ashcroft assumes cooperation in every line, and never answers
+the obvious move — the party shoots it. They reported it with a measured figure, **4.90%**,
+for five untrained agents on the 1% floor without Ashcroft's Transposition 63.
+
+**I re-derived the 4.90% from `roster.mjs` and `content.mjs`, got it exactly, and shipped the
+finding with a false premise attached** (cd3be71): that shooting it strands the party. It does
+not. Line 136 — the crossing is *"open since the felling, widest at dusk, and it will not
+close on its own"*. b5 caught it themselves and sent the correction unprompted; corrected in
+a8f48df, inside the hour.
+
+**The mechanism is worth recording because verification did not prevent it.** The arithmetic
+was never the part that could be wrong. A verified figure lends its credibility to whatever
+sentence it is standing in, and the check that would have caught it is the cheapest available:
+*what does this figure need to be true in order to matter?*
+
+**It is sharper than that, in fact.** There is no `[CUS: Transposition — …]` beat anywhere in
+CLEAN GROUND — none of the document's twenty-two beats name it, and the 63 appears twice, both
+characterisation. The house mission structure *does* make the return a Transposition roll per
+agent (`tools/mission.mjs:343`), with one agent able to aim for everybody, which is why the
+instinct was so persuasive: **it is correct everywhere else in the system, and this scenario
+opted out of it by having a door that is already open.** So it was not a right number on a
+wrong premise. **It was a right number with no question attached, and the premise arrived to
+give it one.**
+
+**The corrected finding is better than the original**, which is the part worth keeping. *"You
+are stranded"* is a dead end a GM has to argue the table out of. *"You walk home with nobody
+holding the door, and you come back thin"* is a scene, built entirely from four lines already
+in the document — 136, 256–259, 1012 and 1494 — that had never been stood next to each other.
+And it lands on the one character whose sheet already says *you have come back thin before*.
+
+**State:** CLEAN GROUND v0.22, twenty guards green, every desk-reachable branch reached. The
+only untested thing left is a run with human beings.