KalinkaPlayer

Measure precision / recall / MRR of Kalinka's CLAP-backed `ai_search` against the local `localfiles.db`. Use when the user asks to "evaluate the AI search", "benchmark CLAP", "check semantic-search quality", "measure retrieval precision/recall", or wants improvement suggestions for the embedding-based search. Generates a ~100-query benchmark, builds ground truth from artist heuristics + predicted tags + album metadata, runs the queries against both the audio-side (`vec_tracks_clap`) and text-side (`vec_tracks_clap_text`) vector indexes, and writes a Markdown report under `tmp/ai_search_eval/`.

Measure the quality of localfiles enrichment (MusicBrainz / AcoustID / Deezer / filesystem fallback) by scanning `localfiles.db`. Use when the user asks to "assess enrichment", "score the metadata quality", "find bad album / artist matches", "check enrichment coverage", or wants a before/after baseline to evaluate a change to the enricher. Emits `tmp/enrichment_assess/findings.json` (machine-diffable) and `tmp/enrichment_assess/REPORT.md` (human-readable).