Companion data and code for: Evidence-grounded feedback improves singer-disjoint reference-action prediction

Single companion archive (version 4, 2026-11-03) for the manuscript under review. Extracts to one vocaguide_companion/ directory holding the aggregate tables, manifests, reference-action rules, constrained realizer, aggregation and audit code, configuration, figure generation, and dependency-light tests, with inventory.json pinning every other file by byte count and SHA-256. Combines the former data and code records into one self-verifying archive. Original VocalSet audio is not redistributed. Supersedes version 3.