Separate what you know

Take the live decision you named in The inspectable record and produce a written split: established fact with a source tag on every line, inference, and unknown split into decision-relevant and comfort. Expect about 45 minutes for the first one and fifteen for later ones.

Two things make this fail. The first is a fact column containing verbs of intention, which means the inferences have been written in the wrong place and the exercise has produced a tidier version of what you already believed. Read every line back and delete any word that describes what someone wanted, meant or decided. The second is an unknown column with nothing in it that could be resolved this week, which means the split has produced no action. Rewrite at least one unknown as an observation somebody could make in the next five days, and name who.

Choose a decision you took at least six months ago whose outcome you now know. Reconstruct it using only material that existed at the time: emails, documents, notes, calendar entries. Write what was established, what was inferred, and what was unknown, as of the date of the decision. Then write, separately, what you remember believing. Expect about 90 minutes, most of it in the archive.

The exercise produces two findings and the second is the point. The first is the gap between the reconstruction and the memory, which is usually larger than people expect and always in the direction of the outcome. The second is what you did not go looking for at the time, which is more useful than any bias you can now name, because it is specific, it is yours, and it will recur. It fails if you work from memory and consult the record only to confirm, which is the natural order and the wrong one. Open the archive first.