A forecast is a claim the future can answer.
It is not a measure of how wise you are.

1. A claim the future can answer

ESTABLISHED FRAMEWORK

“The launch will go well” can never be wrong, so it can never teach you anything. “At least 8 of our 12 colleagues will open the tool weekly by 30 November” can be wrong, and that is its value. The words that make a claim answerable are its resolution criteria: who decides, by when, counted how. Write them before the outcome exists. Many arguments about who “called it” are really arguments about what the sentence meant.

2. Four things people blur

ESTABLISHED FRAMEWORK

Likelihood is how probable something is. Impact is how much it matters if it happens. Calibration is a property of many forecasts: do your 70% claims come true about 70% of the time? Hindsight is the feeling, afterwards, that you knew. A 5% chance of flooding the shop deserves a plan because of impact, not likelihood. A 90% forecast that fails once is not refuted; a single outcome speaks to calibration very quietly.

3. Start from the reference class

ILLUSTRATION ESTABLISHED FRAMEWORK

Hypothetical: you plan to finish a report in two days and your plan feels 80% safe. Ask a different question: of my last six reports this size, how many finished on time? Say two did. That gives a starting point near one in three, a base rate from a reference class. Then adjust for what is genuinely different this time (fewer meetings, a clearer brief), and nudge, don’t leap. Six reports is a tiny class, so treat the number as an anchor, not a verdict. The point is the order: outside view first, inside story second.

4. Why one outcome says so little

ESTABLISHED FRAMEWORK

Suppose ten forecasts are each exactly 70% likely. By ordinary binomial arithmetic, five or fewer coming true happens about 15% of the time, by chance alone. So a bad week of misses is weak evidence that your judgment is poor, and a good week is weak evidence that it is good. Ambiguity makes it worse: if resolution wasn’t written down, memory quietly picks the reading that flatters. Small samples invite stories; keep the stories loose.

5. Wanting, and hindsight

INTERPRETATION

When you want something to be true, the number drifts upward and the reasons arrive afterwards. When the outcome is known, “I always suspected” rewrites the record. A modest defence is to write the number and the reason down before. That doesn’t make you immune; it only makes the drift visible to the one person who can fix it, you.

6. What would change my mind

ILLUSTRATION

Claim (hypothetical): “A cousin’s bakery will still be open in two years.” Starting estimate: 70%. Before looking at any news, list what would move it, and which way:

• Sales fall three months in a row → lower, perhaps substantially.
• The landlord renews the lease early → raise it a little.
• A larger competitor opens next door → lower a little, unless regulars are loyal.
• Two days of rain → almost no movement. Noise should not move you.

These sizes are judgments, not calculations. Their purpose is to decide in calm what counts as evidence, so that later you update on the evidence, not on mood.