Do not treat generated communication summaries as identity evidence

When an underlying transcript contains a phonetically unstable name, do not treat an automated communication summary as independent identity evidence. Compare the candidate name with structured case data, nearby human-authored messages, and the speaker's own statements, and check whether the summary blended adjacent actors. Agreement across genuinely independent sources strengthens an attribution; a summary copied from the transcript does not.

If the audio itself is disputed, this method cannot replace listening to the recording. Keep unresolved identity claims unresolved when the recording is unavailable, and do not infer an error rate from a single case.