What problem does it solve? Textbook music arrangement rules are often repeated without evidence. This Skill answers whether real recordings actually follow those rules by cross-checking each rule against a corpus of 63 measured ARR-SPEC entries derived from Cambridge-MT multitracks, producing hit rates and tiered evidence levels (strong rule / tendency / technique / suspended). ## Core Features & Use Cases - Corroboration Table: Maps rules against the corpus to compute hit rates and assign evidence tiers, with two honesty constraints preventing circular validation. - Reverse ARR-SPEC Corpus: Uses structure-and-parameter-only entries (no audio, scores, or lyrics) with honest TODO markers for fields machines cannot measure. - Conflict Resolution: Provides a procedure for when rules and data disagree—suspect the criterion before the rule, illustrated by real cases like the hardcoded 12 dB threshold that produced a false 16% hit rate. - Use Case: When adding a new arrangement rule to the library, run it through the corroboration workflow to check whether it qualifies as a strong rule eligible for the 20-item self-check lint table. ## Quick Start Ask the agent to check whether a specific arrangement rule holds in real music and what evidence tier it belongs to.
