Check speaker alignment before attributing ASR disagreements to recognition

A whole-call text matcher on mixed bot/caller audio can turn a genuine transcription disagreement into a false audio location. Before treating disagreement as a recognizer error or using it to crop training clips, check whether the candidate overlaps the other speaker's matched speech, whether its duration is plausible, and whether short replies were excluded. These are screening flags, not proof of speaker identity or recognition truth. Review ambiguous matches in full-call context and establish audio-grounded labels before measuring corrected-text rescue.

Applicability and limits: this applies to post-hoc transcript alignment on mixed audio. It does not establish any provider's error rate and does not replace independent human review or channel-separated timestamps.