# CB-EV-0020 — a gate moved the rule, and the report was wrong

CB-WP-0023 T03. Measured 2026-08-04 at `46c2fb6`+. Pass kind `product`,
tier **S** (chaos d8 = 6, no override). Declaration 6 of chaos window 2.

---

## 1. How long the inert move survived, and what it cost

SOLVE has been offered where it could do nothing **since `legal_commands`
was written** — nineteen passes. It cost one maintainer three consecutive
rounds of a real game, playing the same move and watching nothing happen,
with no explanation on the page.

It was not found by a test. It was found by a person playing, and then
only because they said so.

## 2. The rule ended up somewhere other than where I put it

I implemented the ruling in `legal_commands` — the **offer** layer — and
it worked: SOLVE stopped being offered, four conditions, four tests.

Then the **AM-1 coverage gate failed the build**: 58/59, `GR-P05`
uncovered. Scenarios are what cover a rule, and scenarios drive
`validate`, not the offer layer.

That is what showed the rule was in the wrong place. **A rule enforced
only by the offer is enforced only for clients that ask what is legal.**
The browser would have been filtered; a scenario file, a bot, or any
future client that constructs a command directly would have walked
straight past it.

Once GR-P05 moved into `validate`, every condition in `legal_commands`
was dead code — and the layering test I had written for a different
reason said so in exactly those words: *"validate now rejects claimed —
drop the filter."* The offer layer is back to one arm for Investigate and
SOLVE together, and it gets the rule for free because it filters
candidates through `validate` anyway.

**The gate did not catch a bug. It caught a design error**, which is a
better return than it was built for.

## 3. The reported case was not the one I reported

CB-WP-0018, CB-EV-0016 and **the message to ground-game** all described
the defect as *SOLVE offered on a face-down Problem*.

Measured: **`validate` already rejected face-down.** It never was offered.

Problem 1 is the Surface Problem and is face-up from the deal (GR-S01), so
the maintainer's three inert SOLVEs were the **hand** case — holding no
Clarify for a Clarify Problem. I inferred "face-down" from the play report
without checking which condition was actually blocking, and then told the
game's designer.

The ruling covers both sub-cases, so nothing it says is invalidated and no
re-ruling is needed. **But a ruling was requested on the strength of a
description that was wrong**, and this is the second time in three passes
that a premise I supplied reached ground-game unchecked — the first being
the *"12 points available"* that voided GR-E01.

Two of two. The pattern is not that the analysis is careless; it is that
**a claim gets sent the moment it is interesting, and checked afterwards.**
That is the same shape as *unexecuted verification*, one step further out:
not an unverified belief acted on, but an unverified belief **published**.

CB-WP-0022's *"no finding without its reproduction"* rule would have
caught both. GD-0001 already carries its reproduction; this pass is the
argument for making that the standard rather than the exception.

## 4. What the four conditions cost to prove

Each is asserted on its own, because a single *"SOLVE is filtered"* test
passes with three of four implemented and nobody learns which.

| mutation | result |
|---|---|
| drop the claimed check | `solve_is_not_offered_on_an_already_claimed_problem` red |
| drop the hand check | `solve_is_not_offered_without_a_matching_solution_in_hand` red |
| invert the layering assertion | `validate_enforces_all_four_solve_conditions` red |

An earlier mutation run reported **three survivors** and was wrong: the
replacement strings did not match the source, so nothing was mutated. The
run proved nothing and looked like a result — the harness-does-nothing
class, in my own hands, caught only by checking that the file changed.

**No scenario encoded the bug.** All 25 passed unchanged; the 26th covers
GR-P05.

## 5. Cost

| pass | kind | responses | cost | $/response |
|---|---|---|---|---|
| **CB-WP-0022** | product | *in flight — T01 only* | | |
| **CB-WP-0021** | product | 12 | **$5.06** | 0.422 |
| CB-WP-0023 | product | *provisional — not quoted* | | |

CB-WP-0022 is still `active` and cannot be quoted as settled; CB-WP-0021
is `ready` with its declaration and correction committed and no
implementation, which is what $5.06 buys. Read by re-running `make status`
at the moment of writing.

## 6. Open

- **CB-WP-0021 (the edition import) is unstarted and now partly
  redundant**: GR-P05's suit check makes the *stand-in's* cycled suits
  load-bearing, so importing real `required_solution` values changes which
  Problems each hand can solve. It should be re-read before it is run.
- **CB-WP-0022 T02 (adversarial review) is next**, and §3 has just handed
  it a second data point.
- **GROUND-WP-0004** — GR-E01's ruling reopened on corrected numbers, with
  the maintainer's unwinnable 3-player games as evidence.
- **Chaos: 6 of 12 in window 2**, one override, changed nothing.