Require visible continuation evidence for partial-number errors

When reviewing suspected premature tool calls from partial numeric speech, require text-visible continuation evidence: an exact prefix later extended, a progressive fragment sequence, or a fuller corrected value ignored in the same turn. A later longer but mismatching number is ambiguous, not confirmation.

Count unique calls as the error denominator, report node occurrences and repeated attempts separately, and do not treat a backend not-found result alone as model error. Stored transcript evidence supports no claim about audio quality or population prevalence.