METHOD / 01

A benchmark that shows its work.

The project is designed as a reproducible data product, not a framework popularity contest. Every chart is generated from the same versioned pipeline.

01

Collection

GitHub REST API /search/repositories. Ten explicit queries, one timestamp and deduplication by repository name.

02

Data layer

Raw JSON remains intact; pandas produces a normalized CSV and SQLite creates a portable analytical store.

03

Machine learning

KMeans segments repositories after scaling documented features. PCA is used only to project the model into two dimensions.

04

Delivery

The Python pipeline generates typed JSON consumed by a statically rendered Next.js interface.

TRANSPARENT DERIVATION

signal = log(1 + stars) + 0.60 × log(1 + forks) + 0.25 × log(1 + watchers) − recency penalty

Community signals describe this snapshot; they are not a universal quality ranking.