---
id: CB-WP-0023
kind: product
title: "SOLVE is offered only where it can do something"
status: done
state_hub_workstream_id: "ae3a1491-4417-4cb2-b490-b694463cc171"
---

# Purpose

```
structural tier  S   (implements a ruling inside an existing capability —
                      no new port, no canonical interface, no dependency)
chaos            d8 = 6  → no override
declared tier    S
```

Declaration 6 of chaos window 2.

## A ruling, not a judgment call

`ground-game` ruled SOLVE's legality on 2026-08-03 (GROUND-WP-0002 T02),
after clay-borg raised it from play:

| sub | ruling |
|---|---|
| **(a)** face-down Problem | **Not offered — illegal target.** *"only face-up, non-Denied. Browser no-ops were a filter bug, not a bluff mechanic."* |
| **(b)** no matching Solution in hand | **Not offered.** *"the engine only offers SOLVE when a matching suit is held."* |
| **(c)** already-claimed Problem | **Same-round race legal; prior-round claim not offered.** |

**The bluff reading is dead.** CB-WP-0018 raised (a) as *possibly* an
intended bluff in a commit/reveal game and deferred to ground-game rather
than deciding. The answer is that it was a filter bug, and the engine has
been offering an inert move since the day `legal_commands` was written.

`legal_commands` offers Investigate and SOLVE on **every** problem key,
with no reference to `face_up`, `denied`, `claimed_by` or the hand.

**Only SOLVE is ruled on. Investigate is not touched** — implementing more
than was ruled would be inventing rules, which is what this whole exchange
exists to stop.

## Task: filter SOLVE to what the ruling allows

```task
id: CB-WP-0023-T01
status: done
priority: high
state_hub_task_id: "52637f6f-314f-4715-9d81-5e9bea8ced97"
```

SOLVE is offered on problem `p` for seat `s` **iff** `p.face_up`,
`!p.denied`, `p.claimed_by.is_none()`, and `s`'s hand holds a card of
`p.suit`.

On (c): at Select time every `claimed_by` is from a prior round — claims
land at Resolve — so *"prior-round claim not offered"* is exactly
`claimed_by.is_none()`, and the same-round race is already legal because
both seats select before either resolves. **Say this in the code**, or the
next reader will add a round comparison that does nothing.

**Controls:**
- each of the four conditions must be independently mutation-provable —
  drop one filter, and a test naming *that* condition goes red;
- the maintainer's reported case must be reproduced and then fixed: SOLVE
  on a face-down Problem is offered before, not offered after;
- **`make sim` must still pass**, or the ruling has broken a scenario that
  encoded the old behaviour — in which case the scenario was encoding a
  bug and must be updated with a note, not quietly edited.

**Done 2026-08-04.** `make all` exits 0, 26 scenarios, rule coverage
**59/59**. **No scenario encoded the bug** — all 25 passed unchanged.

**The rule ended up somewhere other than where I put it, and a gate moved
it.** It went into `legal_commands` first. The AM-1 coverage gate then
demanded a scenario for the new GR-P05 — and scenarios drive `validate`,
not the offer layer. That is what showed the rule belonged in `validate`:
**a rule enforced only by the offer is enforced only for clients that ask
what is legal.** The browser would have been filtered and a scenario file
would have walked straight past it.

Once GR-P05 moved into `validate`, every condition in `legal_commands`
was dead code, and my own layering test said so in those words
(*"validate now rejects claimed — drop the filter"*). The offer layer is
back to one arm for Investigate and SOLVE together.

**And the reported case was not the one I reported.** CB-WP-0018 and the
message to ground-game described SOLVE offered on a **face-down** Problem.
Measured: `validate` already rejected face-down, so it never was offered.
Problem 1 is the Surface Problem and is face-up from the deal — the
maintainer's three inert SOLVEs were the **hand** case, holding no Clarify
for a Clarify Problem. The ruling covers both, so nothing is invalidated,
but the record was wrong and is corrected here.

Four conditions, each asserted separately, because one "SOLVE is filtered"
test would pass with three of four implemented. Mutations: dropping the
claimed filter and dropping the hand filter each turn their own test red;
inverting the layering assertion turns `validate_enforces_all_four` red.

## Task: retire what the ruling settles

```task
id: CB-WP-0023-T02
status: done
priority: medium
state_hub_task_id: "4a436635-2eff-42b1-957c-fdb8b310f910"
```

Update `specs/GroundRules.md` §Underdetermined for the SOLVE items, and
retire the provisional flags the ruling closes. **A ruling that lands in
code but not in the spec leaves the next reader with two sources.**

Report to `ground-game` that the ruling is implemented — closing the loop
is the part that has failed twice.

**Done 2026-08-04.** `specs/GroundRules.md` gains **GR-P05** stating the
rule and recording that the bluff reading is dead, and
`scenarios/ground/gr-p05-solve-legality.yaml` covers it — P1 refused for
holding no Clarify, P2 admitted and claiming. The prior-round-claim half
is asserted in the unit test instead, because reaching a second round
costs a dozen commands to test one rejection, and that is said in the
scenario rather than left as a gap.

## Task: evidence

```task
id: CB-WP-0023-T03
status: done
priority: high
state_hub_task_id: "3a05165c-df33-4668-8ca2-2340789b3531"
```

`evidence/CB-EV-0020-*.md`. Short.

- **How long the inert move survived**, and what it cost: a maintainer
  played SOLVE three rounds running with no effect and no explanation.
- **Whether any scenario encoded the bug.**
- **Quote CB-WP-0022's cost by re-running the instrument.**