Being right is half of it. The other half is knowing when you are right: if your 90% predictions come true 60% of the time, your intuition is overconfident in exactly the places that matter, and the fix is to slow down there, not everywhere. This page collects every prediction you locked in the units and every audit drill you checked, and compares how sure you said you were with how often you were right. Read the calibration table only once a level has at least five revealed predictions; below that, chance dominates.
The ledger needs JavaScript and browser storage. In a private window it works but forgets when the window closes.
The same table works for anyone whose predictions you track. When you direct an agent (unit 0, §5), ask it to state a result and a confidence before each experiment, and log both next to yours. After twenty experiments you have two calibration tables, and you know in which kinds of question to trust its 90%, and yours.