Limits
Where a person's judgment is unreliable
- Rare events and very long horizons.
- Large numbers and base rates the person has never seen.
- Questions about themselves or people they love, where wanting and expecting blur. We record the want gap for this reason.
- After a person has seen someone else's answer.
How many checked forecasts before calibration means anything
Very little before about 30 checked forecasts. Under that, a few lucky or unlucky outcomes dominate.
Why the machine answer is hidden until the person commits
A number seen first becomes an anchor. If the person sees 0.55 they drift toward it, and you get the machine's answer twice instead of a second opinion. So we never show it first.