Decisions.

The framework

1. 01

   Write what would change your mind — before you gather anything.

   A belief that no possible observation could contradict is not a judgement, it is an attachment. So before you open a dashboard or ring a customer, finish this sentence in writing: I would abandon this if I saw {something specific}. The condition has to be observable, with a number and a date on it — "if fewer than three of the next ten qualified prospects raise this problem unprompted" beats "if the market didn't want it". Then watch for the move that ruins the whole exercise: when the evidence arrives and you find yourself amending the condition instead of the belief, you have shifted the goalposts, and the honest thing is to say so in the meeting rather than quietly redraft the line.

2. 02

   Go looking for the no.

   Wason's participants never lied and never ignored data; they simply proposed tests that could only come back yes, and so do you. Look at the shape of your own enquiries this week: what you typed into the search box, which three customers you rang, which expert you chose. If you asked why this pricing model works rather than who has tried it and lost money, you ran a confirming test and learned nothing. The correction is deliberate and mildly unpleasant. For every source you consult because it will probably support the decision, consult one chosen because it probably will not: the customer who churned, the competitor who tried this and retreated, the person on your team who argued against it and went quiet.

3. 03

   One standard for both sides: you audit disagreement harder.

   In 1979 Lee Ross, a social psychologist at Stanford, working with Charles Lord and Mark Lepper, gave students holding opposite views on capital punishment the same mixed body of research to read. Both groups finished more convinced of what they already believed: each accepted the studies that agreed with them and picked apart the methods of the ones that did not. You do this in every review. The confirming number gets a nod and a slide; the disconfirming number gets the sample is too small, that client was always difficult, the quarter was unusual. The test is procedural rather than moral. If the objection you just raised would not have occurred to you had the number landed the other way, strike the objection and keep the number.

4. 04

   Keep a page memory cannot rewrite.

   Willpower is a poor guard against a bias you cannot feel operating, so build the correction into how decisions get made rather than into how hard you try. Keep one page per significant decision: the belief, the case for, the strongest case against, and the observation that would reverse it — all dated and written before the outcome is known. That last condition is the entire value of the page. Memory rewrites what you believed to match what happened, so without a dated record you will recall having weighed the objection that turned out to be right, and learn nothing from having waved it through. The page is the only version of your reasoning nobody can edit afterwards.