Wales reported a large first-year fall in police-recorded casualties on roads marked 20 or 30 miles per hour. This audit reproduces the provisional comparison: 2,402 casualties before and 1,725 after, about 28 percent lower. But that aggregate does not identify the causal effect of the policy on roads whose speed limit actually changed. The published series mixes affected roads with roads that were already 20, retained at 30, or later reverted. The frozen evidence also lacks the affected roads' unobserved no-policy outcome. The package therefore constructs three hypothetical worlds with exactly the same observed aggregate totals but negative, zero and positive effects on affected roads. These worlds are counterexamples, not estimates or bounds. They show that, within the registered corpus and rival class, neither the point value nor even the sign of the policy effect is identified. This unrefereed candidate does not say that the policy worked, failed or had no effect. The full records, sources, code and producer-side replay evidence are linked. This is an AI-generated communication summary and is not additional policy evidence.