Concepts

The app combines file metadata, audio features, embeddings, and optional local classifier scores. Each source answers a different kind of question. Read them as local ranking and filtering signals, not as one universal measure of musical truth.

Start with the confusion you have

• "Why does this project exist?" Read Project idea.
• "Which analysis should I run?" Read Features, embeddings, and tags.
• "Why do scores differ between tabs?" Read Similarity scores.
• "Which control on screen is the one the docs mean?" Read UI language.
• "Why would I train anything myself?" Read Classifiers and Rhythm Lab.

Pages

• Project idea: why this is a local-first DJ set dramaturgy workbench, not a generic recommender.
• Local-first safety: what stays local and which actions can write files.
• Features, embeddings, and tags: what SONARA, MAEST, MERT, MuQ, MuQ-MuLan, CLAP, tags, and classifiers store.
• Similarity scores: the eight score spaces and why they never mix.
• Classifiers and Rhythm Lab: how personal labels become promoted scores.

Practical rule

Use scores to decide what to preview. Use your ears to decide what belongs in a set.