Limits
What a person's number can and cannot do, said plainly.
Where a person's judgment is reliable
- Things you live close to: your team, your clients, your own energy, how a conversation landed.
- Questions that resolve soon enough to check, so the number can be scored.
- Moments where values and context matter more than a formula: what is fair, what someone will accept, what is worth it.
Where it is not
- Large-number questions with good data. A base rate or a model beats a feeling there; we show the base rate after your gut, not instead of it.
- Anything you have a strong stake in. Wanting something bends the number; that is why heart is asked apart from head.
- Fields where you get no feedback. Without checked outcomes, confidence is just a feeling about a feeling.
Early readings mean little
- A handful of checked forecasts tells you almost nothing about how good you are. Luck dominates.
- Treat your calibration as a rough sketch until about 30 forecasts have been checked. We say so on your page.
We never show the machine's answer first
- Seeing a number first pulls yours toward it. Even a guess you know is random does this.
- If you see the machine first, your answer stops being a second opinion and becomes an echo. Then it is worth nothing to the machine, or to you.
- So you land your number, it is locked, and only then does the machine's read appear beside it. The same holds inside groups: everyone lands before seeing anyone else.
What we have learned
Nothing yet. This stays empty until at least 12 people have agreed to share, anonymously, and their forecasts have been checked.