This repo

Benches we can show

The catalog is the benches whose tasks and result files ship with getEvals. Today that is the browser-agent harness. The scores are baselines from those files.

Agents

Open sourceBaselines

Browser agent eval

11 deterministic tasks on local fixture sites — extraction, navigation, reasoning, forms, and a multi-step login — scored by checker functions. The published rows are baselines.

Harness · rendered from results JSON

1. heuristic-scripted7/11 · Baseline2. naive-first-match1/11 · BaselineLLM adapterNo published run

A language-model run appears here when a results file is added and the site is rebuilt. This page does not republish leaderboards from other labs.