A trader runs four setups. The account is up. Everything looks fine. Then they tag each trade with which setup fired it and discover three are flat and one is carrying the whole book — while another has lost money every month for a year and nobody noticed, because the total was green.
That's attribution, and it's the difference between knowing you make money and knowing why.
Tag at entry, not from memory
The rule that makes this work: record why you took the trade at the moment you take it, mechanically, as a field. Not afterwards. Reconstructing intent from memory produces a record of the story you tell yourself, not what happened.
The tags should be whatever genuinely distinguishes your plays — the setup that fired, the level or condition involved, the market regime, the time of day. If you can't tag it, you can't measure it, and if you can't measure it you're managing it by vibes.
If you can't say which setup made the money, you can't repeat it — and you can't cut what's bleeding.
Measure each slice like a separate business
Once tagged, run expectancy per tag. Each setup gets its own number: trades, win rate, average win, average loss, expectancy after costs. Now the questions answer themselves. Which play pays? Which one is a hobby? Which one only works in a particular regime?
This is also how you catch the ugliest pattern in trading: a winning strategy funding a losing one. The aggregate hides it perfectly. Attribution makes it impossible to miss.
The trap: slicing too thin
The obvious next move is to slice further — by setup and hour and volatility — and it's where attribution turns into self-deception. Every extra dimension divides your sample. Four setups across four hours is sixteen buckets, and with 200 trades that's a dozen or so per bucket. Every one of those numbers is noise wearing a lab coat.
Slice on one dimension at a time, and only when each bucket still holds enough trades to mean something. Otherwise you're doing data mining and calling it analysis.
Tag the label, measure the mechanic
One subtlety worth the trouble. Tags that don't persist across days — a specific price, a particular strike — can't be ranked, because tomorrow's version isn't the same thing. What persists is the mechanic: what actually happened to trigger the entry. First touch or third? Reclaim after a break, or a failed break faded? Same setup name, completely different trades.
Rank the mechanic, not the label. That's usually where the real answer is hiding.
What it's for
Attribution isn't bookkeeping — it's the instrument that makes strategy decisions arguable from the record instead of from the last trade. Cut what has never paid. Size up what has. And when you're tempted to add a rule because of a bad day, look at the attribution first and find out whether that day was even the setup you're about to change.
See also: reading your equity curve and what is a trading edge.