RingBRP/docs/scenarios/CLEAN_GROUND_PLAYTEST_3.md

CLEAN GROUND — desk playtest 3 (player chair, contrarian)

Scenario version: v0.7, at commit 796bb7d Date: 13 September 2026 Method: full session from the player chair, choosing against the scenario at every junction — go round the farm, refuse the column, skip the depot, refuse to test, and try to cancel the entry from the wrong side of the seam. Seeded roller, seed 5580, with one corrected branch at 7731 (see Act Four). Pass 1 was 4271, pass 2 was 9314.

Verdict: the failure cases and the pass-2 fixes all held. But playing against the scenario found two holes neither GM-chair pass could reach, and both are refusals rather than rolls — the two things players do that the document has no answer for.

---

Act One — we never go to the farm

Contrarian premise: Corrigan is the interesting thread, not the sheep. Chase the employer's paperwork instead.

Bhattacharya Research — Corrigan's paperwork    47 vs 63   success
Renshaw      Persuade — the foreman             30 vs 63   success
Braithwaite  Medicine — the body, grudgingly    62 vs 53   fail

• Research landed and the truth section supports it beautifully: casual labour, paid cash, unfiled — which is why the copy came apart in a week. A player reaching this from the employer's side rather than the body's side gets there just as well.
• ⚠ A Persuade success bought an interview with a man who does not exist. The timeline says the copy came apart in front of his foreman on 6 September, with police attending. That is the best eyewitness in Act One — somebody watched it fail — and there is no foreman in the NPC roster: no name, no essence, no idea what he says. I invented him at the desk, which means the document did not.
• Medicine failed and CG-H01's photocopy covered it exactly as written. Second pass in a row that failure case has earned its place.
• ⚠ Act One's failure case assumes a location this route never visits. It reads: if the party never goes near the stump, Swinburne takes them to it. She cannot — they are in a forestry office twenty miles away. The automatic notch is only automatic for a party standing at the stump, so the clue the act "cannot afford to lose" was lost here, and the backstop could not fire.

The refusal the scenario has no answer for

At dusk, having worked out that something comes through a gap on a farm boundary, the party did the professional thing: cordoned it and called it in. Nobody crossed.

• ⚠⚠ Acts Two, Three and Four are all on the far side, and the document does not contain a single line about declining to go. Not a failure case, not a nudge, not a reason the Bureau will not simply send a bigger team tomorrow. It is the most reasonable choice on the table for a cautious party and it stops the scenario dead.
• This is not a contrarian stunt. Agents who have just read a coroner's note about a man who walked himself to death have every professional reason to wait for daylight and authority. The scenario needs to answer it, and the answer probably is not "you feel compelled to step through."

For the rest of the pass I assumed they were ordered through, and carried on.

Act Two — we refuse the column

Renshaw  Insight — the child sent with bread    9 vs 63   special
Ashcroft Insight — THE OFFER                   89 vs 63   fail

• The written failure case fired — a child is sent over with bread — and it worked on this table as the text predicts it works on almost everybody. Refusing the column is not a viable refusal, which is the right design.
• ⚠ Minor: a special on Insight has no written result. The child is real and that is all the text has. A special ought to buy something — she has been told to smile.
• THE OFFER failed, so Ashcroft took the offered memory for a real one. Per fix 7 this quietly makes him the understudy's preferred target two acts later, and he was not told. Hold that thought.

Act Three — we skip the depot and try to fix the tree

Bhattacharya Knowledge — requisition number on the tin   50 vs 63   success
Ashcroft     Anomaly Lore — can we just re-cut the peg?  14 vs 63   success

• The ration-tin failure case worked. The depot was never opened, the number carried the truth, and the act did not stall.
• ⚠ The re-cut idea arrived on a SUCCESS, and the text only answers it as a fumble. The fumble result — total confidence that a fresh notch will hold the entry — is written and good. But re-pegging is the obvious idea, players propose it unprompted, and on a success Anomaly Lore must say something. The honest answer is that he knows it cannot be done and why: the entry is pegged to the older authority, and there is no older tree. That is information, not a dead end, and it is currently unwritten.

Act Four — we refuse to test him at all

Pollard Spot — just watching him, no examination   79 vs 53   fail

Nobody examined him, so rehearsal step 5 arrived on schedule, as beat 3 promises.

⚠ I then mis-applied my own fix, which is a finding in itself. I rolled step 5 against Okonkwo, the lowest POW in the room — because that is what the countdown row says. The Ashcroft exception, which had been armed since Act Two, lives only in beat 3's sub-bullets. If the author of the fix missed it eight minutes after writing it, a GM at hour three will miss it too. Corrected branch, seed 7731:

Understudy POWx5 — reaching for Ashcroft       13 vs 75   SPECIAL
Ashcroft   POWx5 at Difficult (he accepted)    18 vs 42   success

• Special beats success, so the understudy wins, and a PC is taken. The local opposed ruling resolved a special-against-success without a moment's hesitation, which is exactly what it was written for.
• Fix 7 is verified end to end, and it is better than I expected: the anchor who quietly accepted a comfortable lie in Act Two is taken in Act Four because he accepted it, at 42% instead of 85%, and the table can trace the causal line backwards through the whole session. That is the best outcome this scenario can produce and it is now reachable.

Close — we try to cancel it from the wrong side

The party, standing on the far side, asked to cancel the entry from inside it.

Bhattacharya Research — can we cancel from in here?   100 vs 63   FUMBLE
Ashcroft     Anomaly Lore — what happens to us then?   40 vs 63   success

• ⚠⚠ Both questions are unanswerable from the document. Nothing says whether cancellation can be performed from the far side, and nothing says what becomes of agents standing inside a duplicate when it stops being a place the record supplies. A fumble on the first and a success on the second and the GM has nothing either way — the fumble needs a consequence and the success needs a fact.
• Players will ask this, and the answer is load-bearing on the whole ending. If cancelling from inside kills them, the Close is a different scene: it means somebody has to go back through first, and the decision acquires a cost nobody has priced. If it does not kill them, say so, because a table that suspects it might will not choose it.

---

Player-hat verdicts

Archetype	Verdict	Score
Methodical note-taker (Bhattacharya)	Still the best-served sheet. Reached the truth from the employer's side without the farm at all, which says the case is genuinely multi-route.	9/10
Roleplay-first talker (Renshaw)	Bought an interview with an NPC that does not exist, and got a special the text has no result for. Two dead ends in one act, both in her lane.	5/10
Quiet tactician (Pollard/Okonkwo)	Refusing things worked: the cordon-and-call-it-in play is completely sound and the scenario simply has no reply. Satisfying and slightly hollow.	6/10
Lore-digger (Ashcroft/Braithwaite)	Best arc in three passes. Took the offered memory in Act Two and was taken for it in Act Four, traceably. Then asked the one question at the Close that the document cannot answer.	8/10

---

Fix list

Nothing applied.

1. SERIOUS — answer "we are not going through." Text hole. The most reasonable refusal available, and there is no line about it. Recommended: the Bureau's answer is that the seam is pegged to a tree that no longer exists, so it is not going to be there tomorrow in the same place — waiting costs them the crossing and the forty-one with it. That is a reason, not a compulsion, and it leaves refusing as a real choice with a real price.

2. SERIOUS — say what cancelling from the far side does. Text hole, and it changes the Close. Decide and write both halves: whether the entry can be cancelled from inside, and what happens to anybody standing in the duplicate when it goes. Then give the fumble a consequence.

3. Write the foreman. Text hole. One line of essence and three things he says. He watched the copy fail; he is the best eyewitness in Act One and currently a blank.

4. Rewrite Act One's failure case so it does not assume the farm. Text hole. The notch is only automatic for somebody standing at the stump. A party that never goes there needs a route that reaches them — Swinburne telephones the Bureau, or the police file includes a photograph of the cut.

5. Answer the re-cut on a success, not only on a fumble. Missing option. A success should say plainly that it cannot be re-pegged, and why.

6. Move fix 7's conditional target somewhere a GM will see it. Staging. Put the Ashcroft exception on the countdown row beside "lowest POW in the room", not only in beat 3. I missed it myself, minutes after writing it.

7. Give Insight and Anomaly Lore specials a result where they matter. Minor.

8. The four-player cut's mean hides a coin flip — VERIFIED, ready to apply. Missing context on a published number. Histogramming agents down rather than averaging it:

   down	0	1	2	3	4
   share of fights	23.1%	16.9%	10.9%	4.0%	45.0%

   68% of fights land at one of the two ends and only 11% land on the published mean of 2.31. The fight is decided by round five: none of the three down by then and 58.5% of those fights are wipes (68% of all fights); one down and it falls to 19.7%; two down and it is zero wipes in 84 fights. Six players against six of the column is not bimodal — 10 to 18% at every outcome — so this is a property of the cut, not the creature.

   • Provenance, because this took two attempts. The finding came from another session. Its first citation did not reproduce for me at all, and the cause turned out to be two different random generators — simulate.mjs had a private mulberry32, playthrough.mjs a look-alike LCG, and check-focus a third copy — so a seed named a different fight in each tool and every cross-tool citation was unreproducible by construction. Fixed at 8627fa3, one exported generator for all three.

   • Verified here, after the fix. Three commands, each doing a different job.

     The two ends — one targeting mode, two seeds. This is the honest illustration of a coin flip, and it cannot be misread as a claim about concentrating fire:

     node tools/playthrough.mjs --creature quiet_neighbours_npc --count 3 \
       --party ashcroft,bhattacharya,renshaw,braithwaite --seed 2 --mode random
     → [random] PARTY WIPED in 9 rounds · 4 of 4 down · 1 dead
     
     node tools/playthrough.mjs --creature quiet_neighbours_npc --count 3 \
       --party ashcroft,bhattacharya,renshaw,braithwaite --seed 5 --mode random
     → [random] PARTY WINS in 4 rounds · 0 of 4 down · 0 dead
     

     The shared stream, which is the thing that was broken — the same seed must name the same fight in both tools:

     node tools/simulate.mjs --creature quiet_neighbours_npc --count 3 \
       --party ashcroft,bhattacharya,renshaw,braithwaite --runs 1 --seed 2
     → 4.00 of 4 down · median rounds 9 · 100.0% wiped · 1.00 deaths
     

     --mode is required. Without it playthrough runs both targeting arms and prints two fights, which is exactly how the original citation went wrong.

   • ⚠ --mode focus is NOT the other face of the coin. At seed 2 it also wipes — 11 rounds, 4 of 4 down, 1 dead. The bimodality is across seeds, not across targeting, and any written version must not imply that concentrating fire flips this fight. (It is worth 8.3 points across the corpus per check-focus, which is a real and different claim.)

   • The buckets, so the figure is checkable: 1353 / 548 / 84 / 15 fights reach round five with 0 / 1 / 2 / 3 of the three down, wipe rates 58.5% / 19.7% / 0.0% / 0.0%. They sum to 2000, which is the run length. "Zero wipes" is zero of the 84, not zero of the 2000.

   • What to write: the honest warning is nearly half of all parties die, and the dishonest one is you will probably lose two people — because that is the outcome least likely to happen. Say the shape, not the average, and note that it is decided by round five.

FIXES APPLIED — v0.8

All eight applied. The two serious ones were refusals, and both needed a decision rather than a rewording.

"We are not going through" is now answered with a reason and not a compulsion: the entry is pegged to a tree that no longer exists, so a seam with no peg does not sit still — waiting does not make it safer, it makes it unfindable, and the forty-one go with it. Registry says that once and leaves the choice alone. A party that still refuses gets a real, bleak ending in ten minutes rather than four railroaded hours.

"Can we cancel it from in here" is now a design decision with teeth. Cancellation is a Registry act on our side — the entry is a record, the record is held here, and nothing done on the far side touches it. So it cannot be done from inside, somebody has to be back through before it happens, and anyone still in the duplicate when it goes goes with it, because they were never filed as being there. That is a cost the Close did not have: not only what they choose, but who is standing where when it takes effect. The fumble now costs a Coherence to correct.

Also in: Ray Teasdale, the clearance foreman who watched Corrigan come apart and whose own phrase is he stopped being able to do his hands; an Act One failure case that no longer assumes the party is at the farm (Swinburne telephones, and the police file carries a photograph of the cut); the re-cut answered on a success as well as a fumble, with the reason — the entry is pegged to the older authority and there is no older tree; fix 7's conditional target moved onto the countdown row where a GM will actually see it; a special result for reading the child; and the verified bimodality paragraph, with the focus-arm claim dropped.

POST-PASS — two findings from another session, handled differently

1. Ashcroft is not an attacker, and the cut is three. APPLIED in v0.9. Reading the two narrated fights showed Dominic attacking at vs 1, four times. Confirmed against the sheet rather than the log: weapons: [], and his only combat skill is Firearm (Pistol) 43 with nothing to fire. So the four-player cut fights with Renshaw and Bhattacharya on compact pistols and Braithwaite on a knife — a quarter of the party out of any fight.

A defect I introduced. "He carries no weapon, and the GM must not fix it" was written as characterisation and never checked against a fight. The fix is to say it, not repair it: his player is told before the game that Dodge, First Aid, Insight and the anchor are his set-piece, the scaling note states the three-attacker arithmetic, and EXPOSURE names it as part of why violence at four players is party-ending. Arming him would have cost the character for a problem that was really a briefing failure.

2. The bimodality has a mechanism — VERIFIED, deliberately NOT published. Both sides disable at half max HP, so every disable thins the return fire and makes the next likelier; nothing pulls the fight back toward the middle once it tips. First blood is a true coin in the cut and worth 43 points of wipe rate (49.7% / 50.3% at mean round 1.7 → 23.1% against 66.5% wipes), against a 14-point swing at six-vs-six. It reproduces exactly here.

It stays out of the scenario for now, and the reason is the point. It comes from tools/first-blood.mjs, a reading tool with no guarded baseline — and this house's rule is that a GM-facing page may not carry a figure that cannot be re-derived. The published "decided by round five" numbers came out of the guarded lineage and can be checked by anyone who doubts them. "Round five is where it becomes legible, not where it happens" is the better sentence, and it is not yet a citable one. If the measurement gets a guarded home, reframe the paragraph around the first disabling blow.

What three passes have now confirmed

• The handout and failure-case layer is the strongest thing in the document. CG-H01's photocopy, the ration tin, the child with bread and the almanac pair have all now fired under failed or skipped rolls, across three different seeds, and the case never stalled on a die.
• The Close is still finished. Its only problem is a question about physics, not drama.
• The scenario is multi-route in a way I had not verified — Act One's truth is reachable from the body, the stump, or the employer's paperwork independently.
• Every hole found in this pass is a refusal or an unasked question, not a roll. Two GM-chair passes found roll-shaped defects; the player chair found the shape of the ones that were left, which is the argument for running all three.