clay-borg/workplans/CB-WP-0047-all-four-boards-and-all-three-modes.md

---
id: CB-WP-0047
kind: product
title: "All four boards and all three modes"
status: done
state_hub_workstream_id: "772b2646-1258-409b-8480-6dedd6c5c894"
---

# Purpose

```
structural tier  M   (changes the state and therefore the recording, and
                      moves a reported number -- the threshold -- from the
                      engine to the edition)
declared tier    M
```

> *"lets have a look at the cards and the rules to those cards again and
> implement all the scenarios and modes"*

## What was actually missing

The two halves turned out to be very different sizes.

**The modes were already implemented.** `ScoringMode`'s three arms all
score, `--mode` reaches them, F14 fixed the driver. What was missing was
that **nothing had ever compared them**, and the page never named them —
see below.

**The scenarios were not implemented at all.** `edition::deal` has taken a
`scenario_id` since the day it was written, and the only caller passed the
literal `"SCN_01"`:

```rust
let dealt = crate::edition::deal("SCN_01", seats)?;
```

**The seam was the whole mechanism and it sat unused.** 15 of the 20
Problem cards had never been dealt by anything — not a gate, not a panel,
not a player. Nothing was red, because nothing asked.

## Task: deal every board, name every mode

```task
id: CB-WP-0047-T01
status: done
priority: high
state_hub_task_id: "d6298fa3-8c3d-4a47-ae52-945549b9ba95"
```

### The scenario is state

`GroundState.scenario`, `#[serde(default = "default_scenario")]` → `SCN_01`.
In the state for the same reason `variant` is (CB-WP-0038): the board and
the threshold follow from it, so a recording that did not carry it could
not be replayed.

**`standard-Np` keeps meaning `SCN_01`.** Twenty-six recorded scenarios
name that preset; a grammar that redefined it would have moved every one
of their boards while their hashes still claimed to pin them. `scn-03-4p`
names the rest. `--scenario` accepts `3`, `scn-03`, `SCN_03` or
`confidence` — ids are the edition's vocabulary, *"Broken Confidence"* is
the player's — and is validated against the edition, never a pattern, so
`SCN_09` is refused by name.

### The threshold now comes off the Scenario card

`threshold()` was `match seats { 0..=2 => 5, 3..=4 => 7, _ => 9 }` — the
engine's own copy of a number four cards already print, which is what F25
was raised about.

**The first version of this control was worthless and mutation said so.**
Reverting to the hardcoded bands left every test green, because all four
scenarios print 5/7/9 and the two paths are observationally identical on
every input the edition can supply. A control that cannot separate the
thing it is about from its fallback is not a control.

Fixed by splitting `threshold_from(list, scenario, seats)`, which can be
handed a card that **disagrees** (4/6/8). That test goes red under the
mutation; the loop over real scenarios never could.

### The page names the mode, the board, and what winning means

The header read `scoring CommonProblem` — the Rust variant's name, where
the Mode card is titled *"COMMON PROBLEM, PERSONAL EDGE"*. **CB-WP-0034
deleted this exact defect from the move buttons** and it was still
standing on the one line that says what winning means.

**And the coverage probe was holding it in place.** `("mode", "scoring
BondedCoalitions")` — a probe matching `Debug` output, which went red the
moment the defect was fixed. That is the **third** time this shape has
been found (CB-WP-0024, CB-WP-0034, here), and it is now specific enough
to state as a rule: *a probe that names a rendering pins that rendering;
probes name facts.*

The page now carries the Scenario's premise (*what happened*) and the Mode
card's rules text (*how this game is won*) — including the **tiebreak**,
which is a rule a player can play for: "lower Stress, then more Bonds"
changes what a losing seat should do in round five, and it lived in a
column nothing read.

### Mutations

| mutation | what went red |
|---|---|
| `setup` deals `"SCN_01"` regardless of preset | *"the four scenarios yield 1 distinct board"* |
| `threshold_from` ignores the card | *"the threshold did not come off the Scenario card"* |
| header back to `{mode:?}` | *"the page shows the Rust variant's name"* |

**Done 2026-08-08.** `make all` green. Verified live on
`make ground SCENARIO=4 MODE=coalitions`.

## What the panel found

`make panels` gained `scenario-panel`: 4 scenarios × 3 modes × 3 seat
bands, greedy throughout, 100 games per cell.

**1. Two of the four scenarios are the same board.** SCN_01 and SCN_02
have identical suits and values at every priority; every cell matches
exactly. Not a defect — a designed reskin is a legitimate choice — but
"four scenarios" buys **three** boards, and a panel that measured them as
four independent boards would be measuring one of them twice. Pinned by a
characterisation test so a future divergence is a decision, not a drift.

**2. SCN_04 is the hard board at 2 players** — 52% group success against
67% and 73%. It is the only deck that needs **two Repair** solutions in
the 2p deal, and the 2p Solution draw cannot always supply them. The seat
band the thresholds treat as uniform is not uniform across boards.

**3. The three modes produce identical play.** Group success is *exactly*
equal across all three modes in every cell. That is correct arithmetic and
a real finding: `GreedyPolicy` maximises the group outcome and never
consults `state.mode`, so **the two competitive modes are scoring lenses
over cooperative play**. `win/g` differs only by how the winner set is
carved from the same games. Whether COMMON PROBLEM and BONDED COALITIONS
change how the game is *played* is untested and cannot be tested by this
panel — it needs a policy that plays for personal score. Filed as F27.

**4. Every board is a formality at 6 players** — 100% group success in all
twelve cells. Consistent with F17's shape and with H2's motivation; noted,
not acted on.

## Not done here

- **No policy plays for personal score**, so F27 stands open and the two
  competitive modes remain unexercised as *incentives*. This is the
  largest remaining gap and it is the natural next pass.
- **SHARED GROUND's mastery rating counts CLAIMED CARDS, not points.**
  The Mode card says *"All claimed Problem cards form one shared score …
  subtract 1 for each Blame token still in play and 1 for each Denied
  Problem"* — and the shared score is the claimed **value**, while
  `mastery` subtracts from the claimed **count**. Underdetermined rather
  than plainly wrong; raised for `ground-game` rather than changed here,
  because scoring is theirs to rule on.
- **The trial log stamps the variant (CB-WP-0046) but not the scenario or
  the mode.** A note from SCN_04 coalitions is indistinguishable from a
  baseline SHARED GROUND note, which is the same defect one pass later in
  two new axes. The marker already carries attributes, so this is small.
- **`design.py` reported an existing reproduction as absent.** It stats
  `row["repro"]` directly, and the register writes **code spans** — so a
  backticked file path was `os.path.exists("`games/…/x.rs`")` = False.
  Named tests survived because `::` short-circuits before the stat, which
  is why it had never shown. F27's panel existed and counted as debt
  against a target of zero. ADR-0018's shape again: right computation,
  wrong string. Fixed with three checks, including that stripping the
  span must not turn the check off.

- **`catalog.yaml` and both `VARIANT.md`s were re-vendored in this pass**: `ground-game` moved H2
  from `experimental` to `measured`, `decision: keep-as-experiment`,
  citing our CB-EV-0032. That is their ruling on the previous pass, not
  work done here.