Correct desk pass 4: six of sixteen ratings were typed from memory

Found while setting up pass 5 by pulling the sheets from roster.mjs instead of
recalling them. The d100s are unchanged; only what they were compared against
was wrong, and three outcomes flip:

  Pollard  Spot, the stump        58 -> 53   success becomes fail
  Bhattacharya Anomaly Lore       58 -> 63   FUMBLE becomes plain failure
  Braithwaite Xenology (Baseline) 45 -> 53   fail becomes success

So three things pass 4 reported did not happen. The re-cut-the-peg fumble never
fired -- 98 against 63 is a plain failure, 96-99 never succeed -- which means
the ten-minute chase and the whole of finding 2 rest on a roll the seed did not
produce. Braithwaite's Xenology succeeded, so the scenario's one
single-point-of-failure remains untested after five passes rather than having
'finally come up in play'. And Pollard missed the stump.

Read correctly, the same seed lands near 3:34 -- six minutes UNDER budget
rather than six over.

Findings 1, 3 and 5 are untouched: the special was a real 7, the Swinburne
fumble a real 00, and THE OFFER's staging is not a dice question. Finding 2's
conclusion also survives because it never depended on the roll -- Act Three is
budgeted at 50, the text predicts the fumble costs 20, and nothing connected
that to the cut. The reasoning was right and the evidence was invented, and the
scenario now says so instead of citing a playtest that did not happen.

The Pacing Note's slack claim is withdrawn rather than replaced. One seed read
two ways gave 3:46 and 3:34, and the gap between them is about the size of the
margin being argued over. It now tells a GM the shape -- every scene has
something that ends it, the cuts are real, Act Three is the likeliest overrun --
rather than a number a desk pass cannot produce.

npm run check: 17 guards, exit 0.

docs/scenarios/CLEAN_GROUND.md

@@ -177,13 +177,17 @@ written.** See Act Two's budget table.
 | **Act Two: the column's hospitality, once rather than twice** | ~5 | A little warmth. Take this one last, and only if the line's rest has already overrun |
 
 **Running total: 65 minutes of cuts available against a 55-minute problem.** On paper that
-is ten minutes of slack. **Plan against four.**
+is ten minutes of slack. **Treat it as somewhere between zero and ten, and take the cuts
+early.**
 
-Desk pass 4 ran v0.13 on a deliberately hostile seed — two fumbles, eight failures — took
+Desk pass 4 ran v0.13 on a hostile seed and estimated ~3:46 against 3:40; correcting six
-two of the three cuts, and finished at **~3:46 against 3:40**, with the hospitality cut
+misread skill ratings in that pass moved the same seed to ~3:34. **Neither figure is a
-still in hand. That is four minutes of margin plus one unspent cut, and it is the honest
+runtime.** They are two readings of one simulated table, ten minutes apart, and the
-number. Ten is what you get on a friendly seed, and a GM who plans against ten will find
+distance between them is roughly the size of the margin being argued about — which is the
-out at the Close that they were planning against the wrong one.
+honest thing to tell a GM about a desk estimate of how long people will talk.
+
+What the passes do establish is the shape: **every scene now has something written into it
+that ends it**, the cuts are real, and the act most likely to overrun is Three, not Two.
 
 **Do not cut:** telling Ivy, the almanac pair, or the Close. They are the scenario.
 
@@ -681,8 +685,11 @@ carelessness — this column has done this before, and it has lost people doing
 > moment the fumble lands, not at the break. With the cut, a fumbled act runs ~55 and the
 > session absorbs it. Without it, Act Three runs 70 and the Close pays.
 >
-> Desk pass 4 hit this fumble and came out even only because the cut had already been
+> **The arithmetic is the reason, not a playtest.** Desk pass 4 originally reported hitting
-> taken. Nothing in the act previously said the two were connected.
+> this fumble; re-checking its rolls against the real sheets showed the roll was a plain
+> failure, not a fumble — 98 against Anomaly Lore 63, and 96–99 never succeed. So nobody has
+> yet played the twenty minutes. The warning stands on its own: the text predicts the cost,
+> the budget does not carry it, and nothing connected the two.
 
 **Beats**
 - **The evidence of the loop — three routes, with the almanac behind all of them.**

docs/scenarios/CLEAN_GROUND_PLAYTEST_4.md

@@ -317,3 +317,56 @@ committed document until the fix went to apply it. A playtest is a reading of th
 a reading made while looking for faults finds faults that are not there as readily as ones
 that are — which is an argument for applying fixes against the source rather than against
 the notes.
+
+---
+
+## POST-PASS 2 — six of this pass's sixteen ratings were wrong
+
+Found while setting up desk pass 5, by pulling the four cut sheets out of `roster.mjs`
+instead of typing the numbers from memory. **This pass typed them from memory.**
+
+| Roll | Used | Actual | Effect |
+|---|---|---|---|
+| Pollard · Spot, the stump | 58 | **53** | **success → fail** |
+| Okonkwo · Spot, treeline | 52 | **35** | none, fails either way |
+| Braithwaite · Spot, the six | 48 | **35** | none, succeeds either way |
+| Pollard · Listen, the flank | 55 | **53** | none |
+| Bhattacharya · Anomaly Lore, depot | 58 | **63** | **FUMBLE → plain failure** |
+| Braithwaite · Xenology (Baseline) | 45 | **53** | **fail → success** |
+
+The d100s are unchanged — they come from the seed — so this is purely a matter of what
+they were compared against.
+
+**Three things this pass reported did not happen.**
+
+1. **The re-cut-the-peg fumble never fired.** 98 against Anomaly Lore 63 is a plain
+   failure; 96–99 never succeed and only 00 fumbles. So the ten minutes this pass spent
+   chasing the wrong idea, and the whole of finding 2, rest on a fumble the seed did not
+   produce.
+2. **Braithwaite's Xenology succeeded.** The party's one reliable Act Four test did *not*
+   miss. This pass's claim that the single-point-of-failure "finally came up in play" is
+   false, and it remains untested after five passes.
+3. **Pollard missed the stump**, making Act One worse rather than better.
+
+**What that does to the timing verdict.** Without the fumble chase, Act Three runs about
+35 with the seam cut rather than 45, and a successful Xenology is marginally faster than a
+failed one. The same seed, read correctly, lands near **3:34 — six minutes under budget
+rather than six minutes over.**
+
+**What survives.** Findings 1, 3 and 5 are untouched: the special on the column Persuade
+was a real 7, Renshaw's fumble on Swinburne was a real 00, and THE OFFER's staging is not
+a dice question at all. Finding 2's *conclusion* also survives, because it never depended
+on the roll: Act Three is budgeted at 50, the text itself predicts the fumble costs 20, and
+nothing connected that to the cut that pays for it. **The reasoning was right and the
+evidence was invented.** The scenario has been corrected to say so rather than citing a
+playtest that did not happen.
+
+**The lesson, which is the same one twice in one document.** Post-pass 1 recorded that this
+pass formed a view about a clause without re-reading it. This is that again at the level of
+data: sixteen ratings, six of them wrong, none of them checked against the sheet they came
+from, in a document whose entire authority is that the rolls were honest. A playtest that
+invents its own target numbers is not a playtest, it is a story about one — and it is worth
+noticing that the errors were not random. Every wrong value was **close** to the real one,
+which is exactly why nothing looked wrong on the page.
+
+**Desk pass 5 pulls every rating from `roster.mjs` before rolling.**