Three weeks into running a new bot live, I was down 30% against what the backtest said I should have made. I almost switched it off.
I'm glad I didn't, and I'm also glad I never wrote "stop at 30% divergence" on a sticky note and left it at that. The rule is useful. It just hides a trap.
What normal variance looks like
Take a strategy that wins 40% of the time. Winners pay 2.5R, losers cost 1R, where R is whatever you risk per trade. Expectancy is 0.4R per trade, so a backtest says 100 trades should net about +40R.
Now the part people skip. Each trade's outcome swings a lot, with a standard deviation of roughly 1.7R. Over 100 trades that adds up to about 17R of normal wobble either way.
So a 30% shortfall, 12R, is only 0.7 standard deviations below plan. That's not a broken system. That's a Tuesday.
The same rule at different sample sizes
Stretch it to 400 trades and the expected result is +160R with about 34R of wobble. A 30% shortfall is 48R, or 1.4 standard deviations. Still inside what a healthy strategy produces fairly often.
At 1,000 trades the expected result is +400R with about 54R of wobble. A 30% shortfall is now 120R, over 2.2 standard deviations. That's the point where I start taking it seriously.
Same percentage, three completely different meanings. Most traders apply the number without the sample size, and that's where the mistakes come from.
Why most people get it wrong
There are two ways to misuse a threshold like this. The first is panic: you hit 30% down after a few weeks and kill a strategy that was fine. The second is denial, where you read "within variance" as permission to keep running something that's actually broken.
Both come from the same place. A percentage feels objective, but it doesn't know how many trades stand behind it. Sample size is what turns a number into evidence.
My first bot sat in the second camp for six weeks. I kept saying "it's just variance" without ever checking how much variance was plausible. That phrase is only an answer if you've done the arithmetic.
What a real gap looks like
Say you're at 600 trades and sitting 3 standard deviations under plan. That's roughly a one-in-700 event for a healthy strategy. Either you've hit extraordinary bad luck or the live system isn't the one you tested.
Before blaming the market, check the boring stuff. Are fills worse than the backtest assumed, did fees change, or did a parameter get edited at 2am? In my experience the cause is usually one of those, and it's fixable.
A version you can actually use
Before you go live, work out your own numbers. Take your backtest's average trade and the standard deviation of trade results, then multiply the deviation by the square root of your trade count to get the expected wobble.
Write down two lines. One is a warning, around 1.5 standard deviations below plan, where you review the setup and check fills and fees. The other is a stop, around 2.5 or 3, where you pause new entries until you understand the gap.
Use the percentage rule only as a rough guide for high trade counts. For anything under a few hundred trades, the standard deviation bands are the honest tool.
If you're thinking about running a strategy on rules like these, we publish our backtest and live results side by side at v33systematic.com, including the periods that run well below plan. Judging a divergence starts with seeing both lines on one chart.