Good Decision ≠ Good Outcome

Here’s a question that quietly wrecks most people’s judgement: was that a good decision? Almost everyone answers it by looking at how things turned out. Won the bet? Good call. Lost it? Bad call. It feels airtight — and it’s wrong, because a single result is a terrible witness. In a world that runs on probabilities, a brilliant decision can lose and a reckless one can win, and if you grade by the scoreboard you’ll learn exactly the wrong lessons. This final teaching lesson is about prying those two things apart — the decision and the outcome — and then pulling the whole course together, because every model you’ve met so far is really a tool for making the decision good before the dice are thrown.

Resulting: judging the decision by the result

Picture running a red light at 3 a.m. on an empty road and gliding home without a scratch. Did the safe arrival make it a good decision? Obviously not — you got away with a bad one. Now flip it: you stop at every light, drive perfectly, and a drunk driver still T-bones you. The crash didn’t make your driving a bad decision. Outcomes and decisions are different animals, and your gut keeps confusing them.

The poker champion and decision scientist Annie Duke named this reflex resulting: evaluating the quality of a decision based purely on the quality of its outcome. Academics call the same bug outcome bias — we rate a choice as smart or dumb after the fact according to whether it worked, even when the information available at the time hasn’t changed at all. The tell is simple: if your verdict on a decision would flip just because the result flipped — same information, same odds, same everything else — you’re resulting.

Worked example. A surgeon recommends an operation with a well-documented 95% success rate for a dangerous condition. The patient is one of the unlucky 5% and dies on the table. Resulting says: bad call, the surgery killed them. But run it the honest way — given a 95% success rate against a life-threatening condition, recommending the surgery was an excellent decision. The tragic outcome doesn’t change that; it was simply the 1-in-20 landing. Grade the surgeon by the result and you’d punish the right call and, somewhere else, reward a reckless one that happened to survive.

The misconception is that good decisions are supposed to produce good outcomes reliably enough that the outcome can stand in for the decision. In a deterministic world it could. But you’ve spent this whole course learning that the world is probabilistic — outcomes are draws, not verdicts — so the outcome is a noisy, often misleading proxy for the quality of the choice that preceded it.

The 2×2 of decision and outcome

Because the decision can be good or bad independently of whether the outcome is good or bad, you actually have a two-by-two grid, not a single line. Think of it like a kitchen: a great recipe followed carefully can still flop because the oven died, and a sloppy one can still come out edible because you got lucky with the ingredients. Recipe quality and dinner quality are separate axes.

Lay decision quality (was it the smart bet given what you knew?) against outcome quality (did it turn out well?) and you get four cells, each with its own name and its own lesson.

The two diagonal cells are where everyone gets fooled. A bad beat is a good decision with a bad outcome — you did everything right and the low-probability disaster still landed; the temptation is to “learn” from it and never make that good bet again. Dumb luck is a bad decision with a good outcome — you got away with it; the temptation is to credit your genius and do it again, this time with the odds finally catching up to you. The off-diagonal isn’t rare, either: the more chance there is in a domain, the more of your results land in those two deceptive boxes.

Worked example. Two investors each put their savings into a single volatile meme stock. Investor A’s triples; Investor B’s goes to zero. Same reckless, undiversified decision — but A sits in dumb luck and B in deserved loss. If you interview them afterward, A will give you a confident-sounding theory of why it was brilliant. The grid says ignore the theory: it was one bad decision that produced two different draws.

Sort the bet, not the result

Time to make this a reflex. Below are scenarios where the outcome is described on purpose — and you should ignore it. Judge each decision on the bet itself: the odds and stakes as they were known before the result. A good decision can lose; a bad decision can win.

Luck vs skill: why outcomes need time to talk

Why can’t you just trust one outcome? Because an outcome is a single draw from a distribution, and a draw tells you almost nothing about the shape it came from. Flip a fair coin once and “heads” doesn’t prove the coin is rigged. Flip it ten thousand times and the pattern can’t hide. The same logic governs decisions: the more luck baked into a domain, the longer results take to reveal whether your decisions were any good.

A single poker hand is mostly luck — even a perfect player loses plenty of individual hands, and a fool wins some. Across a poker career of millions of hands, luck averages out and skill is laid bare. One business quarter can be a fluke in either direction; a ten-year track record is hard to fake. So the rule of thumb is: in high-luck domains, never read decision quality off a small sample of outcomes — wait for the sample to grow, or judge the process directly.

This is also where the whole course converges into a single definition. A good decision is the one that maximizes expected value given what you knew at the time. Unpack that and every model you’ve learned is inside it:

• It starts from the base rate — the realistic prior odds of the thing (lesson 02), so you’re not fooled by a vivid story into ignoring how common the outcome actually is.
• It weighs expected value — probability times payoff across all branches (lesson 03), so you take the bet with the best long-run average, not the flashiest single result.
• And it respects the range of outcomes — the full distribution, not a single point forecast (lesson 04), so you’ve sized the bet for the bad draws as well as the good ones.

A decision that nails all three is good, full stop — even if this particular draw turns out badly. That’s the payoff of thinking in probabilities: it lets you call a decision good before you know how it ends.

Judge the process, not the result

If the outcome is an unreliable witness, what do you grade instead? The process — the quality of the reasoning that went into the bet. A pilot who runs the full pre-flight checklist made a good decision even on the one flight where a freak failure occurs; a pilot who skips it made a bad one even on the thousand flights that land fine. You’re auditing the checklist, not the landing.

The practical tool is a decision journal: before the result is known, write down the decision, the odds you assigned, the expected value, and your reasoning. Later, you can grade the process against what you actually knew at the time — instead of letting the result rewrite history. The core question to log is: “Knowing only what I knew then, would I make this same bet again?” If yes, it was a good decision regardless of how it landed. If no, it was a bad one even if it won.

This directly defuses hindsight bias — the “I knew it all along” feeling, where after the fact every outcome looks obvious and inevitable. It wasn’t obvious; you just know the answer now. A decision journal freezes your real, pre-result uncertainty in writing so hindsight can’t quietly inflate it. It’s the single cleanest way to separate skill from variance: when you can see what you genuinely believed beforehand, you can tell the difference between a good call that lost and a bad call that you’re now pretending you saw coming.

Worked example. You write in your journal: “Launching the product now — I estimate 60% it succeeds, expected value strongly positive, and I can survive the 40% downside.” The launch flops. Resulting screams idiot. The journal answers calmly: you assigned it 40% downside and could absorb it; the bad draw landed, but the bet was sound. Would you make it again with the same information? Yes. Good decision, bad outcome — logged, and learned from honestly.

Where this lies to you

Here’s the failure mode hiding inside everything above — and it’s a sneaky one, because it weaponizes the very lesson you just learned. Once you know that “good decision, bad outcome” is a real category, it becomes the most comfortable excuse in the world: every time something blows up, you can shrug and say “great decision, just bad luck” — and never update, never learn, never improve. That’s resulting’s evil twin: using variance as a permanent alibi.

The hard truth is that sometimes a bad outcome really is evidence your model was wrong. If your “60% to succeed” launches keep failing 80% of the time, the universe isn’t unlucky — your probability estimates are miscalibrated. A string of bad beats is itself data: it should make you suspect your odds, your base rates, or your read of the range. The skill isn’t to ignore outcomes entirely; it’s to weigh them correctly. One bad result in a high-luck domain is mostly noise. A pattern of bad results is signal you can’t wave away with “variance.”

So the balance you’re after has two sides held in tension: accept variance on any single outcome (don’t abandon a good bet because it lost once), but update your model when the outcomes pile up against your predictions (don’t defend a losing process by calling it bad luck forever). Resulting overreacts to one result; the “bad luck” alibi underreacts to all of them. Good judgement lives between them — calm about the single draw, ruthless about the trend.

When to reach for it

Reach for the decision-vs-outcome distinction the instant anyone — a colleague, a boss, a pundit, you — praises or blames a choice purely because of how it turned out. The CEO hailed as a genius after one lucky acquisition; the manager fired over a well-reasoned bet that variance sank; the friend convinced they’re a brilliant investor after one stock mooned. Whenever the verdict rests entirely on the result, ask the magic question — given what was known beforehand, was the bet sound? — and then look at the process, the base rate, the expected value, and the range, not the scoreboard.

Course recap

Step back and look at the ladder you just climbed. You started by separating possibility from probability — refusing to treat “it could happen” as “it’s likely,” and forcing yourself to attach a number to the chance, not just acknowledge the chance exists. You learned to anchor that number on the base rate — the realistic prior frequency of a thing — so a vivid story or a frightening headline couldn’t yank your estimate away from how common the outcome actually is. With odds in hand, you learned expected value — probability times payoff, summed across every branch — so you could compare bets by their long-run average instead of their best-case fantasy or worst-case nightmare. Then you stopped pretending the future is a single number and started thinking in ranges — forecasting the whole distribution of outcomes, sizing for the tails, and respecting your own uncertainty. And now, finally, you’ve learned to separate the decision from the outcome — to grade the bet by its quality given what you knew, not by the single draw that happened to land.

Each rung needed the one below it: you can’t compute expected value without probabilities, can’t set sane probabilities without base rates, can’t think in ranges without admitting probabilities are spread out, and can’t judge a decision well without all of it. Put together, they’re a single skill — making good bets in an uncertain world, and being able to tell they were good before you know how they end.