rising-reference-v3
How this ranking is calculated
Seven-day acceleration among tracked early-stage projects, checked against engineering practices.
Experimental v3 profile in preparation. Earlier editions keep their original method.
Experimental profile: weights are being calibrated.
Family weights
- Accélération GitHub : 85 %
- Pratiques d'ingénierie : 15 %
Budgets are fixed for all candidates. Within each family, a source's relative weight combines relevance, rigor, reproducibility and resistance to manipulation. These editorial choices must be calibrated against real editions.
The score is the weighted mean of source families, minus 20% of their weighted mean absolute deviation. The same measurements republished by another site do not count as a second vote.
Ranking sources
GitHub (new tab) : 85.0 %
Activity and growth measure developer attention. Stars can be manipulated and are not a universal measure of quality.
Pertinence 1, rigueur 0.8, reproductibilité 0.8, résistance à la manipulation 0.6. Ancienneté maximale : 0 jours. Échantillon minimal : 0.
OpenSSF Scorecard (new tab) : 15.0 %
Nine practice checks, with at least six populated. This signal is not a code security audit.
Pertinence 1, rigueur 0.85, reproductibilité 1, résistance à la manipulation 0.8. Ancienneté maximale : 14 jours. Échantillon minimal : 0.
Missing data and coverage
New GitHub and Skills candidates first build a seven-day baseline; Rising candidates need two periods. Intraday surges can flag them before formal admission. A missing baseline after that wait may block publication.
A missing, stale or incomplete source does not count as zero. Every source in the profile is required to assign a rank. Percentiles are calculated within each source's observed pool before candidates are combined. Incomplete candidates retain an upper bound: if it could reach the Top, publication waits.
Le profil exige au moins 20 candidats complets. Le score dépend du pool suivi et ne prouve pas une supériorité universelle. Les installations et les stars peuvent être manipulées; les benchmarks ont des biais de tâches et de langues.
Discovery covers multiple domains and languages through partitioned searches. Ranking requires an open-source license recognized by our published list. GitHub Trending is used only to discover candidates. Intraday surges are separate signals based on our own observations.
Verify and reproduce
Each v3 edition preserves its profile, normalized observations and links, contributions, coverage, exclusions and a SHA-256 fingerprint. Sensitivity ranks come from 20% weight variations and removing each source or family. They describe scenarios, not statistical probabilities.
The evidence will be available after the first v3 edition is published.
Team assessments remain separate from scores. Method changes receive a new version; published editions stay immutable.