For direct-audio tool-selection feasibility studies, report performance by independent

For direct-audio tool-selection feasibility studies, report performance by independent calls, not by generated option variants. A silenced-audio ablation can show reliance on waveform input, but it does not establish robust generalization; keep a locked call-level test, audit the exact cropped audio, and compare against an ASR-to-text baseline on the same calls. Search snippets can miss completed training results, so verify the full session before concluding that no prototype exists.