Methodology

How New Music Roundup collects, matches, and scores albums from 13+ sources each week.

Step 1 — Album Matching

Each source returns its own list of new releases. Before scores can be aggregated, releases from different sources that represent the same album must be unified. This uses a two-stage approach.

Stage 1 — MusicBrainz ID Lookup

For each release, the artist and album title are normalised (lowercased, diacritics stripped, articles like "The" removed, punctuation collapsed), then queried against the MusicBrainz release-group API. If a match is returned with sufficient confidence, the canonical MusicBrainz Release Group ID is assigned. This is the preferred method — a shared MBID guarantees two records refer to the same album regardless of how each source spells the title or artist name.

Stage 2 — Fuzzy Matching Fallback

When no MusicBrainz match is found (e.g. for very new releases not yet in the database), fuzzy string matching is used instead. The normalised album title and artist name from two candidates are compared using the token_sort_ratio algorithm from the fuzzball library. A combined confidence score is calculated:

confidence = (title_similarity × 0.6) + (artist_similarity × 0.4)

Title is weighted more heavily (60%) as it is the more distinctive identifier; artist name variations like "The Beatles" vs "Beatles, The" are common and handled by token_sort_ratio's word-order independence.

Confidence Thresholds

Confidence	Action	Description
≥ 90%	Auto-match	High confidence — albums are treated as the same release
80 – 89%	Pending review	Flagged for manual confirmation via the admin interface
< 80%	Skip	Treated as distinct albums; not merged

Step 2 — Score Normalisation

Different sources use different rating scales. All scores are converted to a common 0–10 scale before aggregation.

Scale	Original Range	Conversion	Example
100-point	0 – 100	÷ 10	85 → 8.5
5-star	0 – 5	× 2	4.5 → 9.0
10-point	0 – 10	No change	8.3 → 8.3
Letter grade	A – F	Fixed mapping	A → 10.0, B → 7.5, C → 5.0, D → 2.5, F → 0

Some sources (Bandcamp Daily, Invisible Oranges) publish curated lists without numeric scores. These are assigned a fixed representative score reflecting that inclusion is itself an editorial recommendation.

Step 3 — Weighted Aggregation

Once all scores are on the 0–10 scale, a weighted average is computed. Each source has a base credibility weight. Specialist sources receive a 1.2× genre bonus when reviewing albums in their area of expertise — e.g. Metal Archives gets a higher effective weight for a death metal album than for a jazz record.

Aggregate score formula

aggregate = Σ(normalised_score × base_weight × genre_bonus) / Σ(base_weight × genre_bonus)

Source Weights

Source	Base Weight	Scale	Fixed Score	Genre Bonus (1.2×)
Metacritic	1.0	100-point	—	—
AllMusic	0.9	100-point	—	—
Metal Archives	0.85	100-point	—	Metal genres
All About Jazz	0.85	100-point	—	Jazz genres
ProgArchives	0.8	5-star	—	Prog genres
AnyDecentMusic	0.75	10-point	—	—
Bandcamp Daily	0.75	—	7.5	—
Saving Country Music	0.7	10-point	—	Country / folk genres
Album of the Year	0.7	100-point	—	—
Invisible Oranges	0.65	—	7.0	Metal genres
RateYourMusic	0.6	5-star	—	—
Spotify	0.5	100-point	—	—

Worked Example

A progressive metal album reviewed by three sources:

Source	Raw Score	Normalised	Base Weight	Genre Bonus	Final Weight	Weighted Score
Metacritic	85 / 100	8.5	1.0	1.0	1.00	8.50
ProgArchives	4.2 / 5	8.4	0.8	1.2	0.96	8.06
Metal Archives	82 / 100	8.2	0.85	1.2	1.02	8.36
Totals					2.98	24.93

Aggregate = 24.93 / 2.98 = 8.36

Data & Attribution

Album metadata and canonical identifiers are provided by MusicBrainz (CC BY-NC-SA 4.0). Cover art is sourced from the Cover Art Archive. The pipeline runs every Friday at midnight UTC via GitLab CI/CD. See the About page for the full source list.