Part V · Chapter 12 · Noise · lesson 1

A good decision can have a bad outcome

11 min

Builds on Expected value, and its two failure modes

Someone drives home after a long evening, decides they are fine, and arrives without incident. Nothing happened. No police, no accident, no near miss.

Ask whether it was a good decision and almost everybody says no, immediately, without hesitating. Now change one detail: the same person makes the same decision and puts the car into a wall.

Nothing about the decision changed. Everything about how it gets judged did.

A decision is made with the information available before the outcome exists. An outcome is produced by the decision and by everything else — other people, the weather, timing, and a great deal of chance.

Judging a decision by its outcome is judging it by inputs the decider never had. It also teaches the wrong lesson in both directions: it rewards reckless choices that happened to land and punishes careful ones that happened not to.

The four boxes

Every decision, once the outcome is known, sits in one of four places.

Good decision, good outcome. The one everybody wants, and the one that teaches least, because it is indistinguishable from luck at the time.

Good decision, bad outcome. The most under-appreciated box. The correct call at the correct odds still loses a large share of the time, and if you review it as an error you will stop making it.

Bad decision, good outcome. The dangerous box. Something reckless was rewarded, and the reward is a training signal pointing directly at doing it again, harder.

Bad decision, bad outcome. The easy one. The lesson arrives with the consequences attached, which is why this is the only box that reliably produces learning.

Predict

Which of the four boxes causes the most durable damage over a working lifetime?

Good decision, bad outcome — because it discourages doing the right thingBad decision, good outcome — because the reward reinforces the behaviourBad decision, bad outcome — because it costs the most moneyGood decision, good outcome — because it breeds complacency

Why the mistake is so hard to stop making

The outcome is loud and the process is quiet.

An outcome is a fact, it has a number, it arrives on a specific day and other people can see it. A decision process is invisible, has no natural units, and is only assessable by reconstructing what was known at the time — which is exactly the thing that becomes impossible once you know how it turned out. The next lesson but one is about why.

So the outcome is the only variable that is easy to observe, and it gets used as the measure. Bonuses are paid on it, reputations are built on it, and sensible organisations full of intelligent people do this constantly.

Not the same thing

Decision quality

Whether the choice was sound given the information and options available at the time.

Measured in: Process — what was known, what was considered, how it was sized
For example: Sizing a position so that the plausible bad case is survivable, and it then not happening.
Chase it alone and: It is invisible and has no natural score, so almost nobody tracks it and it loses every argument with a number.

Outcome quality

Whether the result was good.

Measured in: The result, in the units the result comes in
For example: The position made money.
Chase it alone and: It contains everything the decider did not control, so on a single instance it is a very noisy reading of the thing you wanted to measure.

Which is which? Put each one on a side.

A business that failed after the founder checked demand, priced honestly and kept costs low
A single holding that tripled in a year
A written record of what someone expected before they acted, checked afterwards

What to do instead, given that outcomes are all you can see

The advice "judge the process, not the outcome" is correct and almost useless as stated, because the process is not observable from where you are standing. Three things are genuinely available.

Ask what else could have happened. Not as a consolation, as a description. If a decision had one plausible path to a good result and five to a bad one, landing on the good one does not make it a good decision — it makes it a lucky one, and the ratio is the thing to notice.

Look at the size before the direction. Sizing is the part of a decision that is almost entirely under the decider's control, and it is therefore the part where the outcome tells you the most. Someone who bet an amount that would have ended them was making a bad decision whether or not it landed.

Use samples, not instances. One outcome is noise. Twenty outcomes from similar decisions start to say something, which is why calibration needs a track record and why one spectacular result says almost nothing about whoever produced it.

Check

Two people each put a quarter of their net worth into one holding. One tripled, one lost most of it. What can you conclude about the quality of the two decisions?

The first was clearly better — it made moneyNothing about the two decisions, but something about the sizing, which was identicalBoth were bad decisions, because concentration is always a mistakeThe second was better, because the loss will teach them more

Why this is worse in fields with long feedback loopsOptional

Deciding well and finding out are separated by different amounts of time in different fields. A chess move gets assessed in minutes. A hiring decision takes a year. A decision about where to put money for a working lifetime takes thirty.

Where the gap is short, outcomes are a workable proxy for decisions, because the noise averages out over many observations. Where the gap is long, a person may make fewer than ten decisions of a given type in their whole life, and ten observations from a noisy process is not a sample — it is anecdotes.

That is the structural reason financial judgement is hard to acquire by experience alone, and the reason a written record beats a memory.

Worth remembering

A decision is made from what was known beforehand; an outcome includes everything the decider did not control.
Judging decisions by outcomes punishes sound calls that lost and rewards reckless calls that landed.
The most damaging box is a bad decision with a good outcome, because it reinforces the behaviour and the bets grow.
Sizing is the part of a decision most under the decider's control, so it is where an outcome is most informative.
Where feedback takes decades, there are too few observations for experience alone to calibrate judgement.