You take five breakout trades. Two of them work. The third — the one you sized up on, because it “just felt right” — blows up. The other two go nowhere. You stare at the account, down for the week, and try to extract a lesson from the wreckage.
You used too much leverage.Leverage2. Trading with borrowed money so each price move is amplified — bigger wins, bigger losses, and you can lose more than you put in. Maybe. You should have waited for a cleaner setup. Maybe. You shouldn't have listened to that guy on Twitter. Almost definitely. But also, almost certainly: none of those lessons is actually the right one, because you can't extract a lesson from a sample of five trades, any more than you can tell whether a coin is biased after five flips.
You don't have a trading problem. You have a sample-size problem. The breakout setup behind those five trades wins 55% of the time at +1.5R over hundreds of instances — five trades simply can't show you that, so whatever lesson you just “learned” is noise dressed up as insight. Until you fix the sample size, every other “fix” you try will be chosen by variance — which means it'll work for a week, then unwork itself, and you'll be back at the same five-trade-lesson loop, learning a new wrong thing each time.
By the end of this essay you'll know why your last five trades are almost useless as feedback — and what to measure instead.
The casino is the right mental model.
A blackjack table has an edge of about half a percent. That sounds like nothing. A single hand of blackjack feels like a coin flip — because at the level of a single hand, it basically is. If you played 10 hands and walked away, you might be up, you might be down, and you would have no way to tell from the outcome whether the casino had an edge at all.
The casino doesn't care. The casino plays 50 million hands a year per major property.
Quantitative Momentum3. Gray and Vogel make this case at length: at the level of the firm, the casino's edge is statistical certainty. At the level of an individual table or shift, it's noise. The retail trader tends to see only the second view of their own trading. Across that volume, the half a percent stops looking like a coin flip and starts looking like a business. The math shows up. The variance averages out. The house wins. And notice what the casino is notdoing across those hands: it isn't playing them to find out whether it has an edge. It knows the half percent before the first card is dealt — the volume is there to harvest a known edge, not to discover one.
The retail trader has the opposite relationship to their edge. Whatever statistical advantage you have on the market — and you can have a real one — exists in the long run, not in any one trade. The one trade is the coin flip. The 500 trades are the business.
The casino doesn't care about any single hand. That is the entire reason the casino exists.
When you act like the gambler — sizing up on conviction, judging the system by the last five trades, quitting after a losing week — you are positioning yourself on the wrong side of the same math. The edge that should have been yours becomes variance, and variance, over a small sample, looks like ruin or a windfall depending on the week. Neither is real.
The math has a name: expectancy.Expectancy4. The average outcome of a trade if you ran the same setup many times. Formula: (win% × avg win) − (loss% × avg loss). A system that wins 55% of the time with +1.5R wins and −1R losses has expectancy of +0.375R per trade. Positive expectancy is what "edge" actually means — and it only shows up in your account at large N. Pick a domain below, dial up the number of trials, and watch what an edge — or no edge — actually produces as the sample size grows.
A real +0.38R edge. Five trades can still look like a losing week; five hundred make it a business. Same edge — only the sample size changed.
The resulting trap.
The poker pro Annie Duke calls it resulting: judging the quality of a decision by the outcome.
Thinking in Bets5. Duke's Thinking in Bets is the canonical text on this. Her premise: in any domain governed by partial information and randomness — poker, trading, weather forecasting — decision quality and outcome quality are different variables, and confusing them is the highest-frequency human error. A good poker player can shove all-in on the optimal equity calculation and lose. A bad poker player can shove on a hunch and double up. Resulting says the bad player made the right move and the good player made the wrong one. Resulting is wrong.
Trading is identical. A losing trade taken from a backtestedBacktest6. Running a strategy over historical data to see how it would have performed — before risking real money on it. 55% win-rateWin rate7. The share of trades that close in profit. A 55% win rate means 55 of every 100 trades are winners — which still leaves 45 losers.system isn't a bad trade — it's a perfectly good trade that landed on the wrong side of variance.Variance8. The random spread of outcomes around the long-run average. Even a +0.5R-per-trade edge produces wins and losses on any given trade — only across hundreds does the average show up. Small-sample variance is what makes a good system look bad after five trades and a bad system look brilliant after three.A winning trade taken from a tip with no stop and 4x normal size isn't a good trade — it's a bad trade that got lucky.
The reason this is the trap is that the feedback signal you get from the market is exclusively outcomes. The market never tells you whether your decision was good. It tells you whether you made or lost money. If you train yourself on the outcome — if you take “I made money” as evidence that you traded well — you are training yourself to reach for whatever decision produced the last good outcome, which is almost never the same decision that will produce the next one. You are running a learning loop with backwards rewards.
Now test yourself: can you judge a trade without looking at the P&L?
Friend tells you AAPL 'looks like a winner' at lunch.
→ You buy AAPL with no stop, no thesis, no size discipline.
5 trades. 5 calls. Was each one a good decision?
If you scored low on the quiz, you're in good company. Grading a trade by its result is built into the way humans learn. The fix is not to guess harder at decision quality — it is to separate the two variables in your journal. Was this trade taken because the rules said so? Was the size correct? Was the exit pre-committed? Those three questions are about decision quality, and they have nothing to do with the P&L for that trade.
The exit ramp.
The exit ramp out of small-sample thinking is not “trade more.” Trading more for its own sake is exactly what blows up the gambler. The exit is to stop abandoning an edge that history has already validated, just because a handful of live trades went the wrong way.
Large samples do two different jobs, and it's worth keeping them apart. You validate an edge in history — hundreds or thousands of instances, enough that the expectancy is real and not an accident of a good week. Then you realizethat edge in the only place it can pay you: live trades, taken the way the casino takes hands — mechanically, without sizing up on the ones that “feel right,” without skipping the ones that come after a loss. The discovery happens once, over history. The realization happens trade after trade. The edge shows up across the sample. Not in any one trade. Not in any one week.
One caveat the casino metaphor hides: the casino can be the house because its bankroll dwarfs any single bet, so no run of bad hands ends the game before the math arrives. You don't have that luxury. Being the house only survives if your position sizing keeps a losing streak from busting you before the sample gets large — the MAX DRAWDOWNfigure you watched move in the simulator above is exactly the thing sizing controls. Size each trade so the worst plausible drawdown is one you can sit through, and the edge has time to play out. Size for the one that “feels right,” and an ordinary losing streak takes you out before it can.
That is the casino. You can be the house, or you can be the gambler. It is a choice every trade.
Before you take a trade, can you name four numbers: entry condition, stop, average R you expect, and how many historical instances that R is based on? If you can't name all four, you don't have a trade. You have a tip. We pull the four numbers apart in Essay 2.