Setting out to write the two missing wire tests found that neither reason is reachable. Schema validation is advisory in the result-returning entry point by design, and the strict validator checks a truncating finish reason only when json output is set -- so a plain-text completion the provider cut off arrives as generated, and every one-shot consumer records it as a success. The evidence does cross: finish_reason and schema_validation ride on the generated response, so a caller can tell. But finish_reason is normalised on the way through -- an endpoint saying length reaches the caller as max_tokens -- so a consumer matching the provider's spelling sees nothing wrong. Both pinned. The test asserts the actual behaviour with a banner saying not to fix it: this is deliberate in the result-returning path, and changing it would change what every one-shot consumer records.