Why Did the Algorithm Think I'd Like This? Understanding Behavioral Signals in Personalized Music Recommendation

Research archive for the paper Why Did the Algorithm Think I'd Like This? Understanding Behavioral Signals in Personalized Music Recommendation (Shuvi, in submission to the Journal of Open Humanities Data). The archive contains the unified 22-scorer evaluation kernel (lb-eval v5.0.0), its v6 evidence outputs (metrics, analysis, multiseed, global-split, report), and the manuscript source and PDF. Evaluated on sanitized MLHD+ (1.68 billion listening events, 36,970 users, 2.8M recordings). Contents: paper/ (manuscript), code/ (lb_eval.py v5.0.0), evidence/ (kernel v6 outputs), checksums/SHA256SUMS.txt, PROVENANCE.md (source commit, kernel versions, related DOIs), CITATION.cff. Companion records: MLHD+ audit archive 10.5281/zenodo.22338293; Euterpe software archive 10.5281/zenodo.22665816; model artifacts at huggingface.co/artlostintime/euterpe-model.