Earlier experiments.
NoHalu started with agent challenges, recorded API calls, and a scoring engine. That work informs the questions we ask today.
Hosted runs are paused.
The public challenge runner, run receipts, shared results, and API-backed catalogs are currently unavailable. Old links lead here so you can see their status. Their source remains in the project.
The engine is still available as source.
The open-source halu-core repository contains the evaluation engine. Its documentation describes how to run it yourself. The current NoHalu reading experience does not need that engine.
Some material has a new home.
The Verifier persona is now an annotated resource. A historical browser-agent install note is available with its missing provenance clearly explained.