How the rubric scores songs you already know.

Curation: Each entry is a historically-significant song. Selection draws from canonical lists (RIAA Diamond, Grammy SOTY, Billboard #1s across decades, songwriting-craft canon — Cohen, Dylan, McCartney, Joni Mitchell, Springsteen, Sondheim, etc). Curation is operator-led; the goal is craft diversity, not chart-rank ranking.

Scoring: Each song's lyric is run through the same 8-voice Crucible the public uses at /crucible. The result is frozen here at first publication — no re-scoring based on later rubric edits. If the rubric drifts, the calibration corpus drift becomes visible (which is the point).

What we publish: Title, artist, year, chart context, our score, our verdict, voice counts, and a 1-2 sentence critique summary that reflects the panel's actual reading. We never publish the lyric itself. Standard music-criticism practice.

Why no popularity bias: The rubric grades craft, not chart success. If a song that charted #1 scores 52 here because the lyric is craft-thin, we publish 52. If a song that never charted scores 92 because the craft is masterful, we publish 92. Auto-inflating known hits would break the entire claim the rubric stands on.

When entries change: Never. Once published, the row is frozen at frozen_at. Re-running the rubric on the same lyric should produce the same score (within ±2-3 points of natural Sonnet variance, per the consensus eval discipline). Significant drift is itself a published finding — the corpus's most useful property is its stability.