A personal decision journal: writing the entry before the outcome is known

Goal

Make it possible, months later, to learn from a decision without the distortion of knowing how it turned out. The journal records what was known, what was expected and how sure the author was, at the moment of deciding; the review then asks whether the decision process was sound, not only whether the result was good.

Prerequisites

An append-only place (a dated text file per year, a bound notebook), a rule for which decisions qualify (anything with consequences beyond a week, anything that cost more than an hour of deliberation, any decision the author notices being made on gut feeling), and calendar reminders for reviews.

Steps

1. Before acting, write the date, the decision in one sentence, and the situation as currently understood, including what is unknown.
2. List the options that were actually considered, including "do nothing", with one line each on why they were rejected.
3. State the expected outcome as something observable: what will be true by when, with a number or threshold where possible.
4. Write the confidence as a number between 0 and 1 that the expected outcome occurs, and the main reason it might not.
5. Note the emotional state and time pressure in one line; both are invisible in hindsight.
6. Set a review date and enter it in a calendar. Close the entry; it is not edited afterwards.
7. At the review, add a dated section: what happened, whether the expected outcome occurred, what information was available at decision time but ignored, and what was unknowable. Classify the pair as good decision/good outcome, good decision/bad outcome, bad decision/good outcome or bad decision/bad outcome.
8. Twice a year, read all reviewed entries in one sitting and note recurring patterns: options never considered, systematic over- or underconfidence, decision types that keep landing in "bad decision/good outcome".

Expected result

A record in which each decision can be judged by the information available at the time, a calibration picture (confidence stated against outcomes observed), and a list of patterns worth changing.

Limits and test basis

This is a proposed protocol; no result is claimed. Confidence numbers from a single person over a few dozen decisions are noisy, and the classification into decision and outcome quality is itself a judgement made in hindsight. The related hypothesis entry on this wiki states the calibration effect as a testable claim; this article only describes the record that such a test would need.