Documentation
How OSS Risk Radar scores maintenance risk
Built for developers: how the system is put together and how to call its API, where the data comes from, how to train the model yourself, what the features are and how they performed, and exactly what happens when a repository is scored. For the scientific evaluation of the model, see Model evaluation.
End to end
How it's set up
The four services behind the product, the offline/online split that keeps scores fast and honest, and how to run the whole stack locally.
API reference
Every endpoint the Go API exposes, the async create → poll flow, and the JSON conventions — the web app is just a client of this.
Where data comes from
The public, point-in-time sources behind every signal — GH Archive history, GitHub metadata, and package registries.
Building the dataset
How raw history becomes a leakage-controlled, labeled training table — the offline pipeline and the 12-month label.
Train it yourself
A runbook: prerequisites, the commands to build the dataset, train the artifacts, and promote a new model.
Feature reference
All 43 signals grouped by what they measure, each with an exact definition and the reason it exists.
How scoring a repo works
Step by step, what happens between hitting analyze and seeing a probability, evidence support, and explanations.
Training results explained
How well the model works in plain terms, where it is more or less reliable, and what that means for a score.