The Decision Log

One file. Plain text. Chronological. Written before the outcome is known and read by nobody but you.

Which decisions go in it

Not all of them. A log that attempts completeness dies within a month.

Two filters. Anything difficult to reverse, and anything where you notice that you feel certain. The second catches more than people expect, and it catches the entries that turn out to be worth having, because confidence is exactly the condition under which nobody checks.

Two to five entries a month is a working rate. Fifteen means you are logging operational calls and will stop by March.

The six fields

1. Date and decision. One sentence, in plain language. Deferred the Nampa roof replacement to FY28.

2. Mode and participants. Whether you commanded, consulted, or built consensus, and who was in it. This field costs five seconds and later tells you whether your worst calls cluster around the times you decided alone.

3. What actually drove it. Two or three considerations, not the full defensible list. The distinction matters: the defensible list is what you would say in a meeting, and the real drivers are frequently one item on it plus something you would not put in a memo. Write the real ones. This is the field that makes the log unshowable, and being unshowable is what makes it useful.

4. What you expect, with a number. Specific enough to be wrong. Not “this should hold up” but “no active leaks in the east wing before FY28, seventy percent.” A prediction that cannot fail will teach you nothing.

5. What would mean you were wrong. Named in advance, while you are still uncertain. This is the field people skip and the one that does the most work, because after the fact you will always find a reading of events under which you were basically right.

6. Review date. The date the prediction should have resolved. Put it in your calendar rather than trusting the log to surface itself.

Mechanics

Write it after deciding and before announcing. That window is short and it is the only point at which your reasoning is both settled and uncontaminated by anyone’s reaction to it.

Ten minutes is the target. A log that takes forty minutes an entry is a document being written for an imagined reader, which defeats it.

Keep it somewhere with no sharing feature and no collaborators. A text file. A paper notebook. Anywhere your assistant, your board, and your successor cannot open it.

Four ways it dies

Written afterward. A log filled in once the outcome is visible records your revised memory rather than your original expectation, and produces confidence in a method you never actually used.

Written for a reader. The moment you imagine somebody else seeing it, field three becomes the defensible list and the instrument is finished.

Adopted as a process. Rolled out to the leadership team with a template and a deadline, this becomes a document people optimise. Individuals can keep honest logs. Organizations cannot.

Vague predictions. “Expect improvement” is unscoreable, and a log full of unscoreable entries is a diary.

The test for an entry

Could a stranger reading it in two years, alongside a record of what happened, say plainly whether you were right?

If yes, the entry is sound. If it would require interpretation, rewrite field four before you close the file, because in two years you will be that stranger and you will interpret generously.