A trading bot with a 90% win rate sounds unambiguously great. It can also be a strategy that will eventually blow up an account — if the 10% of trades that lose are, on average, ten times bigger than the 90% that win, the strategy is a net loser dressed up in an impressive-sounding headline number. Win rate alone can't tell you which situation you're looking at. Here's what can.

The core numbers, in order of how much they actually matter

Profit factor
Gross profit ÷ gross loss. A profit factor of 1.5 means the strategy made $1.50 for every $1.00 it lost, across every trade. Above 1.0 means the winners outweigh the losers in dollar terms — below 1.0 means they don't, no matter how high the win rate looks. This is the single most informative number on a track record, because it accounts for both how often a strategy wins and how big the wins and losses actually are.
Maximum drawdown
The largest peak-to-trough decline the account ever experienced, as a percentage. A strategy that returns 40% a year but has a real 35% drawdown along the way is a very different (and much harder to actually hold through) experience than one returning 15% with a 5% drawdown. Return without drawdown is an incomplete picture — nobody experiences a strategy's average, they experience its worst stretch.
Win rate
The percentage of trades that closed profitably. Genuinely useful, but only alongside profit factor — a low win rate can still be very profitable if wins are large relative to losses (classic trend-following), and a high win rate can still lose money if losses are disproportionately large (a common failure mode in strategies with no real stop-loss).
Sample size
Not a performance metric itself, but it determines how much to trust everything above. A profit factor of 2.0 over 8 trades is close to meaningless — it's well within the range of pure luck. The same number over 200+ trades, across multiple market conditions, is real evidence.

The trap: cherry-picked windows

Almost any real strategy has had at least one good month. Marketing material built around "look at this stretch" is technically true and still misleading, because it's not telling you how the strategy performed outside that window. The only honest version of a track record is the full one — every trade, from the strategy's actual start, not a hand-picked slice.

A quick test: ask whether the numbers you're looking at update automatically as new trades close, or whether they're a static screenshot. A live, automatically-computed track record can't quietly exclude a bad week after the fact. A screenshot can.

Reading an equity curve

A cumulative profit-and-loss equity curve is one of the most honest single charts a trading bot can show you, because it makes cherry-picking visually obvious — a steadily rising line with real dips is very different from a flat line that suddenly jumps once, which usually means one outsized trade is doing all the work. Look for a curve that trends up (or down) consistently across many trades, not one dominated by a single event.

Putting it together

No single number tells the whole story. A track record worth trusting has: a profit factor meaningfully above 1.0, a drawdown you could actually tolerate living through, a sample size large enough to rule out luck, and — critically — a way to verify all of it wasn't cherry-picked after the fact.

What our own performance page shows

We built our performance page around exactly these numbers: win rate, profit factor, and a live cumulative equity curve, computed automatically from every closed trade across the fleet — not a curated selection. It updates as new trades close, including losing ones, because a track record that can quietly leave out its worst stretches isn't a track record at all.