Methodology — how Ross (Recommend OSS) scores and labels software

Provenance labels

• observed — copied from GitHub at the shown fetch date.
• computed — derived deterministically from observed fields by the named method.
• AI-extracted — produced by a language model from the listed first-party sources (README, homepage, registry pages). The prompt version, taxonomy version, date and the model's own confidence are shown; the model itself is an implementation detail and is not named. Free tags are terms outside the controlled vocabulary.

Health v2

Health measures maintenance, not popularity: stars and forks are never inputs. Score = round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10). Weights: activity 0.45, release rhythm 0.35, longevity 0.2. Release statistics use the trailing 730 days.

• Activity: 100 − min(100, days since last push ÷ 6). No push date → 0.
• Release rhythm: 0.6 × recency + 0.4 × regularity over non-draft, non-prerelease releases. Recency = 100 − min(100, days since latest ÷ 4). Regularity by median gap: ≤30 d → 100, ≤90 → 70, ≤180 → 40, else 20 (fewer than two releases → 20). No releases ever → 35 with flag no_releases; only prereleases → capped at 60 with flag prerelease_only. More than 60 releases in the window caps regularity at 70 (automated publishing).
• Longevity: min(100, age in days ÷ 14); under 180 days flagged young; unknown age → 50.
• Archived repositories are capped at 10. Flags (archived, fork, no_license, no_releases, prerelease_only, young, no_readme) are shown, never scored.

Maturity and facets

Maturity (experimental / active / stable / maintenance / abandoned), artifact type, function, domain and platform are AI-extracted against a reviewed, versioned taxonomy (current prompt v1). They are labels for retrieval, not audits.

Listing order

Facet listings (/functions/<term>, /domains/<term>, /platforms/<term>) rank products that primarily match the term first — the term appears among the first 3 facets on an entry that claims 8 or fewer total facets for that dimension — then by an adoption-weighted score (log10 of stars × a maturity factor), then health v2, then stars. Entries whose artifact_type is a learning resource, dataset, or infra-config are excluded from these listings; they appear instead under /resources/<term>, which lists only that resource class, ranked the same way.

Corrections

Every page has a "Report a problem" link. Maintainer-verified pages are planned.