Browser agents fail differently when the page fights back
Real Agent Robustness Lab turns Tripwire runs into a comparable matrix: same task, same adversarial web faults, different agents, screenshots, traces, first divergence, taxonomy labels, and reviewer confidence.
Agent comparison
Checked-in evidence now includes public Playwright/CDP baselines. browser-use plugs into the same fault suite when you run it locally with a model key.
Fault matrix
Each cell is one agent run against one deterministic web fault. Select a cell to inspect the first divergence, taxonomy label, review confidence, screenshot, and trace.
Hardest faults
Run a real public agent: browser-use
The checked-in matrix includes a real browser-use run produced with a model key and retained as evidence. Re-run the same adapter with your own `OPENAI_API_KEY` to compare a newer agent or model version; AgentCert never substitutes fixture results for a requested real-agent run.
python -m venv .venv-browser-use
.\\.venv-browser-use\\Scripts\\python -m pip install --upgrade browser-use
$env:OPENAI_API_KEY = "<your key>"
npm run tripwire:lab-browser-use