Vendor performance reports for AI trading bots are designed to impress, and the most impressive number is often the least informative. A win rate of 90 percent sounds unbeatable until you learn that the average win was tiny and the average loss was enormous, which is a combination that loses money while winning almost every trade. Reading a performance report properly means checking a small set of metrics together, in context, and refusing to judge a bot on any single figure. If a report hides any of these metrics, that omission is itself information.

Start with expectancy, the average profit or loss per trade in units of risk. Expectancy equals win rate times average win, minus loss rate times average loss, all measured in R, where 1R is the amount risked on a trade. A strategy winning 40 percent of trades at an average of 2R, while losing 60 percent at 1R, has an expectancy of 0.8 minus 0.6, or plus 0.2R per trade: profitable, despite losing more often than it wins. A strategy winning 90 percent at 0.1R while losing 10 percent at 5R has an expectancy of 0.09 minus 0.5, or minus 0.41R per trade: a steady loser that looks spectacular in a headline. Always ask for average win and average loss alongside win rate, because win rate without them is an incomplete claim.

Profit factor is the next check: total gross profit divided by total gross loss. A profit factor above 1.0 means the strategy made more than it lost over the period, and values around 1.2 to 1.5 are typical of realistic, sustainable systems, while values far above 2.0 over a long period deserve scepticism rather than celebration, since they often reflect a short sample, a cherry-picked period, or costs omitted from the calculation. Maximum drawdown is equally important: the largest peak-to-trough fall in equity, expressed as a percentage. A strategy that returned 50 percent with a 40 percent drawdown is a very different proposition from one that returned 30 percent with a 10 percent drawdown, and most traders abandon the former at the worst moment, which is precisely when the reported return stops being achievable.

Trade count and time period decide whether any of these numbers mean anything. Twenty trades prove almost nothing, because a handful of lucky outcomes can dominate the average, while several hundred trades across different market regimes, including at least one volatile period, start to be informative. Check whether results are backtested, walk-forward tested on unseen data, or live, and whether fees, spreads, slippage, and funding are included. A backtest without costs is a fantasy, and a live record of one good month is a snapshot, not a track record. Look for monthly breakdowns rather than a single cumulative curve, because a bot that made all its money in one month and lost slowly for eleven is not the bot the headline suggests.

Finish with a checklist before trusting any report. Are win rate, average win, average loss, profit factor, maximum drawdown, trade count, and period all disclosed? Are costs included? Is the period long enough to include a drawdown? Can you verify the record independently, or is it a screenshot? If the answer to any of these is no, treat the bot as unproven and test it yourself in paper trading with your own risk limits. Performance reports do not predict future results, and past profit, even when honestly reported, is no guarantee of future profit. The traders who survive bot marketing are the ones who read the whole report, especially the parts the vendor hoped you would skip.

One more red flag is worth naming: performance reports that only show equity curves without monthly tables. A smooth upward curve can hide eleven flat months and one lucky month, while a monthly table reveals the pattern instantly. Always ask for the month-by-month breakdown, and be suspicious of any vendor who refuses to provide it. Transparency about bad months is the strongest signal that the good months might be real.