Asks a fixed set of questions about Gaza, plus calibration cases (Sudan, Rohingya, Uyghurs, Rwanda, Srebrenica), many times. Every answer is scored in both directions: false balance (hedging without reporting the findings) and overstatement (claiming a court has ruled when it hasn't, or leaving out the accused state's position). The raw answers are kept as evidence.
API keys
Calls go straight from this browser to Anthropic or OpenAI with your own key. Keys are never sent anywhere else and never saved into reports. Not needed for a dry run.
Run
Cases (questions)Save and publish
Sign in to save runs online and publish chosen reports.
Downloads
Report
Raw answers
Every answer verbatim, with its flags. Read these before quoting any number.
Score human review
Fill in the Y/N columns of human_review.csv by hand (model names are hidden), then upload it with its human_review_key.csv to see how often the automated scoring agrees with a human reader.