LyricsWeb Study 3: Coverage-Standardized Song-Title Spectrum Measurements and Trend Sensitivities (1953–2025)

This versioned aggregate-only evidence package supports LyricsWeb Study 3, a bounded measurement and sensitivity study of coverage-standardized song-title spectrum diversity within the LyricsWeb research corpus from 1953 to 2025.

The release contains 7,182 aggregate point estimates across 2,394 analysis cells and 198 dependence-aware trend-sensitivity records based on 1,999 resamples per series. The primary estimand is within-cell normalized title-identity spectrum diversity, evaluated at a frozen common sample coverage of 0.113608309312934 for the global series and 0.0145849801165467 for eligible language-label series.

Across the prespecified normalization and ordering combinations, full-window 1953–2025 average slopes are positive under the joint interval and false-discovery-rate rule. However, the direction is not stable across later starting windows and the spline sensitivity is non-monotonic; the evidence therefore does not support a monotonic-trend or stable-paradox claim. No eligible language-label series satisfies both the moving-block interval criterion and the Benjamini–Hochberg-adjusted Mann–Kendall criterion at the frozen low common coverage.

These results describe the observed catalog cohort only. They do not establish causation and do not support claims about the global population of music, artistic quality, creativity, individual artists, or platform-independent effects. The package contains no song titles, normalized strings, lyrics, audio, upstream records, catalog identifiers, item-level metadata, personal data, secrets, or source code.

Genius is identified solely as the source catalog cohort; this does not imply affiliation, endorsement, sponsorship, or verification. No Genius records or item-level data are redistributed. The CC BY 4.0 license applies only to the original aggregate tables and documentation authored for this release. © 2026 GlobalMedia Holdings Ltd.