You log the call while it's still uncertain: the prediction, a line of reasoning, and one number for how sure you are.

As things change, a thirty-second check-in nudges that confidence up or down, and every check-in stays on the record.

When it's over you mark what actually happened and how satisfied you are with the call, your original prediction sitting right there so hindsight can't quietly rewrite it.

Do that a few dozen times and the reliability diagram has enough to work with. It plots what you said against what happened, so you can see where your confidence runs hot or cold.