Alternative Histories: How to Judge Decisions Without Knowing the Outcome

Here's the problem. You make a decision. The decision produces an outcome. You observe the outcome and update your view of whether the decision was good.

This is how almost everyone evaluates decisions. And it's the wrong method.

Nassim Taleb introduces the concept of alternative histories as the correction. Any realized outcome — any career result, investment return, business outcome — is one path through a tree of histories that did not happen. To evaluate the decision correctly, you need to reconstruct the tree, not just the leaf you landed on.

The Russian Roulette Case

The logic is clearest at the extreme. Someone offers you $10 million to play one round of Russian roulette: one bullet, six chambers. You pull the trigger. Nothing happens. You collect the money.

By visible outcome, this looks like a success. Wealth acquired, risk survived. The retrospective narrative writes itself: "bold enough to take the opportunity when it arose."

But five out of six histories from that same decision end with your brain on the wall. The distribution of outcomes from playing is dominated by catastrophe. The decision was bad, even though the specific outcome was favorable. The outcome does not validate the decision because the outcome is one draw, not an evaluation of the distribution the draw came from.

The mistake is using the result to judge the process. The result is noise about the process in any situation where a random variable was involved.

The Drunk Driver Who Always Makes It Home

The extreme case is obvious. The same logic applies everywhere, and most people don't see it.

A man drives home drunk every Friday for a year. Every Friday, he makes it home safely. The track record is perfect. His confidence grows. His friends stop worrying. The safety record seems to validate the practice.

But a full tree of alternative Fridays — different pedestrians, different oncoming drivers, different reaction times, different red-light timing — contains dozens of outcomes in which he kills someone or himself. The one path that materialized is not representative of the distribution of paths. The record of safe arrivals is evidence of favorable draws, not evidence of a safe practice.

Almost every argument that begins "nothing bad has happened yet" is this argument. The bridge held. The levee held. The hedge fund didn't blow up last year. The strategy didn't fail for the past three years. These are statements about observed paths, not statements about the distribution.

The Successful Founder Problem

A founder drops out of college, concentrates their savings into one product, works 100-hour weeks, and exits ten years later for $400 million. The profile interview describes visionary commitment and precise market timing. The decision is praised retroactively as a masterpiece.

What the profile doesn't include: the counterfactual distribution of that decision. A thousand founders with identical profiles, identical conviction, and identical work ethic, split across a thousand similar product bets, produces a distribution dominated by failure. The profile writers only visit the fraction that survived. The decision that looked brilliant in the realized outcome was, in the counterfactual distribution, a lottery ticket with long odds.

This isn't a claim that the founder lacked skill. Skill matters. It shifts the distribution. But even skill-shifted distributions contain a lot of bad outcomes, and the one visible success tells you very little about where the distribution was centered.

How to Actually Grade a Decision

The Talebian approach: mentally reconstruct the full distribution at the moment the decision was made. Before the outcome was known.

Ask: - What were the realistic alternative histories from this decision? - How were they distributed? Symmetric? Fat-tailed? Heavily skewed? - If this decision were made a thousand times with identical inputs, what would the distribution of outcomes look like? - Is the observed outcome near the center of that distribution, or is it an outlier?

A decision that looks brilliant because it produced a great outcome, but whose distribution was dominated by bad outcomes, was a bad decision. It just happened to produce a good draw.

A decision that looks unremarkable because it produced a mediocre outcome, but whose distribution was concentrated around good outcomes, was a good decision. It just happened to produce an unlucky draw.

Evaluating decisions this way is uncomfortable because it refuses to let good results validate bad processes. The stock picker who got lucky doesn't get credit. The trader who executed well but happened to get an adverse outcome doesn't get blamed. The evaluation is of the process — the distribution it was drawing from — not the single result.

Why This Matters Practically

The reason this isn't just theoretical: if you evaluate decisions by their outcomes, you systematically reward lucky strategies and punish good ones that happened to draw a bad outcome.

An organization that promotes based on result — who made money last year, whose initiative landed — will eventually fill its leadership with people who got lucky in the conditions that obtained. When conditions change, those strategies underperform, but by then the feedback loop has created institutional momentum for the wrong process.

The person who evaluates their own decisions by result will steadily build confidence in strategies that happened to work in the specific environment they were applied in — which may be very different from the environment they'll operate in next.

The alternative histories lens interrupts this. Before you promote the result, ask: what distribution was this drawn from? The answer changes who you celebrate and what you replicate.

For the full framework, read Fooled by Randomness: How Luck Masquerades as Skill.