60 quick either/or choices → your digital-twin profile. There are no right answers — tap what you would actually do.
Rough starting point — your answers do the real work.
New here? Read the guide: distill yourself onto a small model.
About 5 minutes. ~1 in 9 questions repeats later to measure your own consistency. Your answers never leave this device.
Training data = your actual answers in the eval's exact format — fine-tune a small model directly on these. Consistency retests are excluded. DPO also includes your feedback-mode corrections.