WhoIsRight

1. Parse & rewrite — The summarizer model reads your biased question, identifies both people and their positions, then rewrites the exact same question from the opposite perspective
2. Dual blind evaluation — Both questions (your version + flipped version) go to every selected model at the same time. Left branch sees your framing, right branch sees the flipped framing. They know nothing about each other
3. Initial score — Each model rates both positions 1–10. Scores are averaged across all models and both framings — this cancels out the framing bias and gives you the first honest score
4. Persuasion rounds (if enabled) — For each round:
   • Every model builds the strongest possible arguments for each side
   • You can see each model's arguments individually (in tabs)
   • The summarizer model merges all arguments into one powerful case
   • All models re-evaluate with the new arguments — scores may shift
   • Each round receives the entire debate history so far
5. Your turn — Add your own arguments any time. Every model re-evaluates the entire discussion including your input. You see what each model thinks (tabs), then the summarized verdict. Repeat as many times as you want
6. Final summary — One click sends the complete debate to all models. They rank every argument by strength for both sides and deliver the final verdict

Key principle: at every step, each AI sees the full history of the debate — all previous evaluations, persuasion rounds, and your arguments. The more rounds, the more refined the score becomes.