ground-game/reports/260808-clay-borg-boards-and-modes.md

---
id: GROUND-RPT-0006
type: report
title: "clay-borg: all four Scenarios and all three Modes measured — modes change who wins, not whether the group survives"
domain: consumer
repo: ground-game
status: informational
owner: bernd
topic_slug: whynot
created: "2026-08-08"
extends: GROUND-WP-0003
---

# Boards and Modes measurement receipt (CB-EV-0033)

Upstream: clay-borg `evidence/CB-EV-0033-boards-and-modes.md`.

**Context:** clay-borg dealt `SCN_01` and only `SCN_01` for its whole
existence — `deal()` took a scenario id and the one caller passed a
literal, so **15 of the 20 Problem cards had never been dealt by
anything**. Fixed on our side; all four boards and all three modes now
run. 4 × 3 × 3 cells, 100 games each, two bots. Everything below is bots,
no felt play.

---

## 1. SCN_01 and SCN_02 are the same board — is that intended?

Identical suit **and** value at every priority. Every measured cell
matches exactly, at every seat band, under both bots.

A reskin is a legitimate choice and we are not calling this a defect. But
it means **"four scenarios" is three boards**, and any study treating them
as four independent samples double-counts one. We have pinned the current
state with a test so a future divergence is a decision rather than a
drift.

**Ask:** intended reskin, or did one deck get copied and not re-tuned?

---

## 2. SCN_04 is materially harder at 2 players

| board | 2p group success (/100) |
|---|---|
| SCN_01 / SCN_02 | 67 |
| SCN_03 | 73 |
| **SCN_04** | **52** |

**Your own data holds everything else fixed.** All four 2p deals are
Surface + priorities 1–2, all four give 6 available points against a
threshold of 5, all four start at Stress 2 over five rounds. The only
difference is the suit multiset:

| board | 2p suits |
|---|---|
| SCN_01 / SCN_02 | Repair, Clarify, Boundary |
| SCN_03 | Boundary, Clarify, Repair |
| **SCN_04** | **Repair, Clarify, Repair** |

SCN_04 is the only 2p deal needing **two of one suit**.

**We are not claiming causation** — nothing here demonstrates the second
Repair is what costs the 15 points. But the seat-band thresholds (5 / 7 /
9) are uniform across boards, and at 2p the boards are not.

**Ask:** is a ~15-point spread at 2p acceptable board variety, or should
the 2p threshold or SCN_04's priority-2 suit move?

---

## 3. Every board is a formality at 6 players

100/100 group success in **all twelve** 6p cells, both bots, every mode.
Same shape as F17. Reported, not acted on.

---

## 4. The three Modes change **who wins**, not **whether the group
survives**

We previously reported that the modes produced identical play. That was
**our instrument, not your game** — the bot never read the Mode card. We
built one that plays its seat's own objective and re-ran everything.

**Group success: unchanged in 34 of 36 cells.** (SCN_03 at 4p moves 99 →
100 in all three modes — that is a bot refinement, not a mode effect; it
moves under SHARED GROUND too.)

**Winning seats per game, BONDED COALITIONS:**

| board | 4p, mode-blind bot | 4p, mode-aware bot |
|---|---|---|
| SCN_01 / SCN_02 | 2.04 | **2.98** |
| SCN_03 | 2.12 | **3.29** |
| SCN_04 | 2.05 | **3.01** |

A seat whose score is its coalition's sum bonds harder, and coalitions
come out about half again as large. COMMON PROBLEM moves at 6p (1.10 →
1.17). Nothing else moves.

**Control:** under SHARED GROUND the two bots agree at all but ≤2
decision points across twelve boards — so the moving columns are
mode-awareness, not "a different bot".

**Design reading, offered not asserted:** the Mode card currently decides
the *distribution of the win*, and the threshold decides survival
independently of it. If a Mode is meant to make the group's task feel
different — not just the podium — nothing we can measure says it
currently does.

### The seat-band pattern is the interesting part

2p: nothing moves in any mode. 4p: the largest effect. 6p: **nothing**
under BONDED COALITIONS.

Our candidate explanation is that **two relation slots per seat cap
network growth**, so at 6p the incentive exists and cannot be acted on.
**This is untested.** It is cheap for us to test if you want it: run with
a third relation slot and see whether 6p coalition size then moves the
way 4p's does.

**Ask:** is the two-slot limit meant to bind this hard at 5–6 players?

---

## 5. Two things needing your ruling

**(a) SHARED GROUND's mastery rating counts cards where the shared score
counts points.** The Mode card reads *"All claimed Problem cards form one
shared score. … For a mastery rating, subtract 1 for each Blame token
still in play and 1 for each Denied Problem."* The shared score is
claimed **value** (what the threshold is compared against); our engine
subtracts those penalties from the claimed **count**. Both readings fit
the sentence and they differ on every game where a 3-point Problem is
claimed. We have not changed it — scoring is yours.

**(b) A package that adds a *file* is invisible to a consumer.**
`h2-scoped-problem-stress` ships `Rules_Text.csv` — 22 passages of
player-facing rules, including the one that says what a `stress_scope`
does — and clay-borg had **never read the file**. Our freshness check
verifies files it knows about; a file no reader mentions is not stale, it
is unseen. We have since read the passage we needed; the other 21 are
still unread.

**Ask:** could a package manifest name its own files, so that a consumer
reading none of them is a detectable state rather than a silence?

---

## Standing limits on all of the above

- **No bot models a rival playing their objective.** A competitive mode
  in which nobody anticipates an opponent is a weak test of that mode.
- **Bond-scoped joint SOLVE remains untested** (CB-EV-0032 criterion 4,
  unchanged) — these bots read the Mode, not the module.
- **No felt play.** All bots, every number.

---

## Ground-game disposition (2026-08-08 review)

| # | Finding | Disposition |
|---|---------|-------------|
| 1 | SCN_01 ≡ SCN_02 mechanically | **Not treated as intentional.** Fiction reskin only; three distinct boards, not four. Residual: re-tune SCN_02 suit and/or values (content workplan). |
| 2 | SCN_04 harder at 2p (~52 vs ~67–73) | **Acceptable variety for now** as a harder board; monitor. If table feel is too swingy, first lever is priority-2 suit (drop double Repair), not global thresholds. |
| 3 | 6p 100% group success | **Known pattern** (F17 / difficulty line). Not fixed here; deal/pressure modules and WP-0005 disposition (Standard thresholds stand). |
| 4 | Modes change *who* wins, not *whether* group succeeds | **Accepted as current design property**, not a defect. Threshold = survival; Mode = distribution. Mode-aware bot was clay-borg instrument fix. |
| 5 | 2 relation slots mute 6p coalition effect | **Intentional cap** (complexity/time). Documented; third slot stays extension, not core r0. Optional clay-borg sensitivity run welcome. |
| 6 | Mastery: points vs card count | **RULED: points.** Shared score and mastery penalties use **point values**, not card count. Modes.csv clarified. |
| 7 | Package file manifest | **Accepted process ask.** Catalog/modules should declare `consumed_files` / overlays so unread files are detectable. Tracked under WP-0008 residual. |