clay-borg/evidence/CB-EV-0027-the-tokens-on-the-table.md

CB-EV-0027 — the tokens on the table

CB-WP-0029 T04. Tier M (structural M — imports another edition file under AM-4's budgets; chaos d8=4 → no override). Declaration 12 of chaos window 2 — the last. Closed 2026-08-07.

Delivered: ADR-0016, Tokens.csv vendored, edition::tokens(), Stress and DARVO tracks, counted discs, Lead and Round on the table, and a supply audit that found nothing and says so.

---

1. The supply, measured

750 games, 2–6 seats, greedy and random policies:

token	supply	max per seat	max on table	exceeded?
Protection	6	1	2	no
Denied	5	—	3	no
Relation link	12	—	12	at the limit, never over
Focus / Blame	6	—	0 conflicts	no

The link row is the finding, and it is a negative one worth stating. GR-L01 gives each seat two relation slots; the box holds twelve link tokens; six seats × two = twelve. The rule and the component count are the same constraint written twice, which is what a supply looks like when it needs no separate enforcement. Nothing was designed to make that true — it fell out.

Protection is the one that could have gone wrong and did not. saturating_add(1) at lib.rs:986 has no upper bound, so the type permits a seventh token. Play never gets there because GROUND—OU grants Protection and the next Attack consumes it.

2. The check that was wrong, and what it cost to catch

The Focus/Blame check reported 2 conflicts. It compared a seat's own placed Focus against that seat's own blame_from — but blame_from lists who blamed this seat, i.e. other players' discs. Different tokens. Corrected: 0.

Fifth instance of the family CB-EV-0019 §1 named — a number computed correctly about the wrong subject:

#	pass	the wrong thing
1	CB-WP-0021	csv cost against a budget that never sees the code
2	to ground-game	"12 in the file" — a sum with no deal table
3	CB-WP-0018	SOLVE's inertness attributed to the wrong condition
4	CB-WP-0025	a timer bracketing whole games; a win rate attributed to the game
5	here	a token's own disc compared against other players' discs

And this is the first one caught before it left the repo — by me, in the same session, rather than by an adversarial reviewer or by the maintainer. Four of the previous five reached a document or another repository.

That is one data point, not a trend. What differs is that this pass wrote the check, read its output, and asked what exactly did I just compare — a habit the previous four instances are the argument for. The family still has no control; facts-check catches copies that disagree and nothing catches a correct computation over the wrong subject.

3. The decision not to enforce the supply

quantity does not bind, and the reason is not that nothing exceeded it.

No numbered rule in GroundRules.md mentions a token supply. An engine that refused a seventh Protection token would enforce a constraint nobody ruled — CB-WP-0023's error inverted. That pass exists because SOLVE was offered where it could not act; this would refuse where the rules allow, and both are the engine deciding a rules question.

So the check ships as a standing control. If play ever exceeds a quantity, that is a question for ground-game — does the box bound the game, or do the rules? — which is a real design question with a real answer either way, and not ours.

Registered as F22, withdrawn: a stated negative. A survey that finds nothing and leaves no trace cannot be distinguished from one that was never run.

4. What the tracks are for

stress 4 is a fact you read. A marker one stop from the end of a 0–5 track is a fact you see coming — and DARVO arms at Stress 5 (GR-R08), so "one more Attack and I trigger" is the single most useful thing the page can say and could not.

Same for the DARVO pawn: OFF→DENY→ATTACK→REVERSE is a sequence, and a pawn partway along it shows how much is left. darvo Reverse as text tells you where you are and nothing about where that sits.

Unverified. Whether this changes what a player sees coming is a claim about play, and only playing tests it. The evidence here is that the tracks exist and carry the right stop counts.

5. Two fixture defects, and a limit corrected

seat_centres matched every <circle>. Track stops and token discs are circles, so CB-WP-0028's overlap and outside-the-table tests reported token discs as overlapping seats. Seats now carry class="seat" and the helper keys on that. Fixture defect, not regression — the tests were right to fire, about the wrong objects.

The table height limit went 460 → 500. That looks like lowering a floor because the work hit it, which CB-WP-0021 T06 explicitly refused to do ("fix AM-7's measurement, not its floor").

It is the opposite, and the distinction is the point. 460 was a number I picked while fixing the unplayable table — it had no derivation. 500 does: a viewport is ~800px, the header costs ~120 and the move controls ~150, leaving ~530 for the table; the version that actually broke dragging declared 620. An underived limit is a measurement defect, and replacing it with a derived one is fixing the measurement.

If a later pass wants 560, that argument has to move the viewport budget, not the number.

6. Chaos window 2 — closed, and its verdict is due

CORRECTED 2026-08-07 — this section is wrong. Window 2 produced one override: declaration 3 (CB-WP-0020) rolled d8 = 8, drew S against a structural S, and changed nothing — which CB-WP-0020 recorded correctly at the time as "One." Eleven rolls at d8 (declaration 1 opened the window at d4) against an expectation of 1.375 eights. The rate was working; I asserted "zero" without counting, and each later evidence file cited the one before it. See ADR-0017 and F23.

This is declaration 12 of 12. The window opened 2026-08-03 at d8.

Twelve declarations, zero 8s, zero overrides. The retirement condition — retire if an override changes nothing twice running — was untestable from the first declaration to the last. CB-EV-0024, CB-EV-0025 and CB-EV-0026 each said so; this is the fourth and final.

window	rate	declarations	overrides	changed the outcome
1	d4	12	2	2
2	d8	12	0	—

The verdict this supports: d8 bought rarity by spending evidence. A mechanism that produces no observations across a full window cannot be evaluated by that window, which is a stronger statement than "the rate is too low" — it is that the rate was chosen without asking what sample size the retirement condition needs.

Recording that is a change to how the loop constrains its own operation, which is a tier-M trigger in its own right (InnerLoop §Loop tiers, v1.6). It is therefore outstanding, not done here, and is named in the workplan so it is not lost between passes — which is exactly how ground-game's ten rulings went uncollected for two days.

7. Cost

CB-WP-0028's cost, by re-running the instrument: make cost. Not inlined (§Single source of fact). CB-EV-0019 §4's unbounded chain is still unbounded, now across four more passes.

Open after this pass

• The window-2 verdict (§6) — the next thing this loop owes itself.
• The wrong-denominator family still has no control (§2), at five instances.
• Whether the tracks change what a player sees coming (§4) is untested and only play tests it.
• Relations, DARVO, Player_Mats and Glossary remain unvendored. The DARVO track now has a picture but not the edition's own words for each stage.
• Three of four scenarios still never dealt, carried from ADR-0015 D5.