v0.20.0Iris says what it checked. A verdict tells a pass from a check that never ran, names who recorded the evidence it judged, and cannot be replaced by the agent it judges→

Live Playground

Paste any output. See the real rules score it.

Runs the same Iris rule library that ships in the npm package — server-side, no install, instant. 26 rules across safety, relevance, completeness, and cost. 10 of them read an agent’s tool calls to judge what it did rather than what it wrote, and those skip here, because this page takes text only — install Iris to run them.

What you paste is sent to iris-eval.com, scored and returned. The text is not stored and not logged: the server records only a salted hash of your IP address (for the rate limit), the category, a length bucket, whether it passed and how long it took. Do not paste real personal data or live credentials.

Try a preset

Results

Paste output + click “Run evaluation” to see rule results here.