The gentle art of the postmortem.
Review a completed forecast without giving hindsight permission to rewrite what you knew at the time.
An outcome is the beginning of a review
The final result arrives and the forecast can be scored. It is tempting to make the review equally simple: a good outcome means the reasoning was good, while an unwelcome outcome means the reasoning was poor. That story is satisfying because it closes the loop quickly. It also discards much of what makes a probabilistic forecast worth keeping.
A completed forecast gives you two different objects to inspect. One is the outcome under the original rule. The other is the reasoning you recorded before the answer was available. A postmortem becomes more useful when you let each object answer its own question rather than asking the result to explain every earlier decision.
Reconstruct the information boundary
Begin by listing what you actually knew when you saved the forecast. Use the original evidence notes, exported copies and dated sources where available. Separate facts that were public at the time from facts discovered later. A later revelation may explain the outcome perfectly without being information you could reasonably have used in the earlier estimate.
This does not mean every earlier choice deserves a defense. You may find a source you overlooked, a definition you misunderstood or an assumption you failed to check. Those are useful discoveries. The distinction is between an avoidable omission in the original process and a development that was genuinely unavailable at the time. They suggest different lessons.
Inspect the question before judging the number
A forecast can fail as a learning tool because its wording is ambiguous. Perhaps the deadline was unclear, two branches overlapped or the accepted evidence source stopped publishing. Address those issues before treating the numerical score as a clean verdict. If another careful reader could resolve the question differently, the uncertainty belongs in the review.
Do not repair the original question quietly after observing the result. Keep a note of the defect and write a better rule for the next forecast. If you cannot responsibly choose a branch under the original wording, leaving the entry unresolved may be more informative than forcing a tidy score. A complete-looking record is not automatically a more honest record.
Look for a specific mechanism to improve
Once the definition and evidence boundary are clear, ask which part of the process most needs attention. Perhaps you began with an unusually memorable comparison, counted repeated commentary as separate evidence or overlooked a dependency. Choose a concrete explanation that connects the process to the estimate. A vague instruction to be less confident is difficult to apply to the next question.
Try writing one change in a form you can observe: before assigning probability, identify a relevant comparison group; before making an update, trace the claim to its original source; before saving, test whether every eligible result fits exactly one branch. These practices can be checked later. They turn reflection into something more useful than a mood about the past.
Let several forecasts carry the larger judgment
One realized outcome provides limited information about the quality of a probability estimate. A path assigned twenty percent can happen without proving that twenty percent was absurd. Across a larger collection of comparable forecasts, you can begin examining recurring patterns, the average score and whether your estimates tend to be too concentrated. Keep differences in question difficulty and timing in view.
End the review with a short, specific lesson and a new question worth asking. You do not need to transform every surprise into a grand theory about your skill. The journal is useful because it preserves many small encounters between expectation and evidence. Give those encounters time to accumulate, and let the next forecast show whether the lesson changed your process in a way you can actually see.
Try closing an entry with three separate lines: what the rule resolved, what the original reasoning missed, and what you will do differently next time. Keep a fourth line for things you still cannot explain. The separation prevents a polished story from absorbing every uncertainty. A review that admits an open question gives your next investigation a more honest place to begin.