Chart Methodology

Each eligible review contributes normalized dimension scores on a 0–1 scale. The overall review score is the weighted arithmetic mean of four required dimensions: songwriting, production, performance, and originality (each weighted 0.25).

For each candidate track, the Bayesian score is:

bayesian_score = (v / (v + m)) * R + (m / (v + m)) * C

• R = candidate's raw mean (weighted average of normalized dimension scores)
• v = eligible review count for the candidate
• C = prior mean, calculated once per run from all eligible scores in scope
• m = prior strength (launch value: 10)

This approach shrinks scores toward the community average when a track has few reviews, preventing tracks with a single 5-star review from dominating. As review count grows, the score converges to the track's true average.

At cutoff, a candidate track requires:

• Track and artist publicly visible, no trust/safety or legal exclusion
• At least one available allowlisted source
• Minimum eligible reviews: 5 for weekly, 10 for 28-day, 20 for all-time
• At least 3 distinct reviewers who do not manage the track's artist
• No single reviewer contributes more than one accepted review
• No unresolved aggregate integrity hold

Absence from a chart does not imply poor quality — it may mean insufficient reviews or reviewer diversity.

Ties are resolved in this exact order:

1. Unrounded Bayesian score (descending)
2. Eligible review count (descending)
3. Unrounded raw mean (descending)
4. Most recent eligible review published_at (ascending — longer sustained result wins)
5. Track UUID (ascending — deterministic final tie-break)

Equal displayed scores may have different positions. Public copy says "ties resolved by published chart rules."

• Calendar week: Previous completed ISO week (Monday 00:00 to next Monday 00:00 UTC)
• Rolling 28 days: Exact 28 days ending at scheduled cutoff
• All-time: From platform epoch to cutoff

A review belongs to a window by its published_at time, not creation time. Windows are half-open [starts_at, ends_at) in UTC.

When a review is hidden, a track is removed, or a moderation/legal action affects chart-eligible content:

• Emergency suppression filters the entry at the read layer within 5 minutes
• A replacement run is generated using the corrected history, typically within 4 hours for current runs
• Historical archives are annotated or replaced within 24 hours
• Prior runs are marked superseded — entries are never mutated in place
• All changes are audited and reproducible from retained input facts

• No paid placement or editorial ordering disguised as a score
• No provider stream counts or sales data
• No battle seeding or chart-to-battle promotion
• No personalized or predictive charts
• No cash prizes or official awards
• No ranking of artists — only individual tracks