The Verdict freezes the winner cell's hash; the template names the family

A registered Hypothesis declares a sweep, and the engine measures one spec per cell (M3.7). Every cell has its own hash — a different stop is a different Strategy under D4 — while the Strategy that entered MEASURED carried the template's hash, the one spec the sweep was built from. Two hashes, one Verdict. Which does it freeze?

The winner cell's. D2 says the spec that measured is the spec that trades, and deployment asserts hash equality against the Verdict (ADR-0002). The only spec whose measured numbers a deployed Strategy can inherit is the winner's; freezing the template's hash would let a Strategy trade a stop that was never the one measured to clear the floor.

The template's hash is recorded beside it as the family. It names what was searched — the sweep and its size are the search-correction's whole input (R4.4) — and it is what the overlap gate and the Register's "what did we search" answer point at (N6).

Consequences

• Verdict.spec_hash is the winner cell's hash; Verdict.family_hash is the template's. HypothesisResolved carries both.
• The template measures; the winner trades. The Strategy built from the template runs SPECIFIED -> COMPILED -> MEASURED and stops there. On a supported Verdict a Strategy is specified from the winner cell's spec, compiled, given the same Measurement, and is the one that enters FORWARD. Its hash equals the Verdict's by construction, which is what the FORWARD -> APPROVED guard asserts (D2).
• The reserve permits one look per spec hash (ADR-0006). The look is charged to the winner's hash. A second sweep that happens to contain the same winner is a second look at the same spec and is refused — which is correct: the reserve was looked at for that spec.
• A null Verdict freezes the winner's hash too. Nothing trades on it; the hash records which cell came closest, so a later supersession can say what it changed.

Considered options

• Freeze the template's hash. Simplest and it is what the M2 controls did implicitly. Rejected: it breaks D2 — the deployed spec would not be the measured one — and it would let a one-cell "sweep" and a nine-cell sweep with the same template look identical in the Register.
• Freeze every cell's hash. Most complete. Rejected: the Verdict is one resolution against one floor; the losing cells are search, and the search is recorded as its size and its refusals, not as verdicts.
• Make the sweep part of the spec's identity. Then the template's hash would be the winner's. Rejected: D4 says the hash is what the Strategy is, and a Strategy is one stop, not a set of candidate stops.