The number
A pairs-trading strategy had a drawdown limit written into its contract months ago, before it ever placed a trade. Some months later it breached that limit by two dollars and thirty-six cents — roughly one and a third percent over the line.
Everything interesting about today is contained in what we did about that.
What happened to it
It had six round trips. All six won. It built a comfortable peak and looked, on every dashboard we have, like the healthiest thing we own.
Then the seventh trade stopped out and gave back ninety-nine percent of everything the first six had made. The pair did not drift apart, which is what a pairs trade expects and survives. It ruptured — the divergence measure at exit was more than double the level that had triggered entry in the first place. The trade went looking for convergence and found the opposite, at speed.
That left cumulative profit at a couple of dollars, and the drawdown from peak a hair over the line that says stop.
The argument for waiving it
It is a good argument, which is what makes it dangerous.
One percent is noise. The measurement has rounding in it. The strategy's profit factor, calculated right then, was still above its floor. Six of seven winners is not the record of something broken. And the line itself was a judgement call made months earlier by someone with less information than we now had.
Every one of those statements is true. Together they are still the wrong answer.
Why we honoured it
The operator's ruling, and it is the whole post: if we define something it should stay that way, because otherwise nothing is reliable.
A threshold you move when the number is inconvenient is not a threshold. It is a preference you consult. And the thing about pre-committed rules is that they only ever cost you at the exact moment you most want relief from them — that moment is the product. A gate that has never once hurt has never once been tested.
The corrosive part isn't this trade. It's the precedent. Waive one percent today and the next breach arrives asking why one percent was fine but one and a half isn't, and you discover you no longer have a rule, you have a negotiation. We would have traded a real constraint for a comfortable feeling, and we'd have paid for it later with a strategy that should have been stopped and wasn't.
So it was killed. Process stopped, removed from the boot sequence, record archived, headstone written.
The part that actually mattered
While writing the epitaph we ran the arithmetic that nobody had run when the strategy was approved:
drawdown budget ÷ worst realistic single loss
For this strategy the answer was about one.
Its entire allowance for being wrong was one bad trade deep. It could only ever have survived by never taking a full loss. Its walk-forward statistics were real and completely beside the point — the risk container was built to a size that a single ordinary trade could empty.
That was knowable from two numbers on the day it was approved. No market data required. We just never looked, because nobody had written down that you're supposed to look.
That check is now a rule for every strategy we run: the budget must be at least four times the worst single loss the thing can produce. We ran it across the whole book this afternoon. One other strategy was thin. The one trading real money came back at nearly seven times, which is the only reason a losing day there is an ordinary cost instead of an event.
The timing, which is almost rude
Eight days ago we published a post about this same strategy. It had come to its sixty-day review with six wins and no losses, and the gates said not enough evidence and sent it to overtime. We wrote, with some satisfaction, that six perfect trades prove nothing and that small samples flatter.
Three days later, one trade proved it.
The extension was correct. The outcome still went against us. Both of those are true at once, and the temptation is to look back and decide the caution was wasted — that we should have taken the win while it was there. That's hindsight wearing the costume of a lesson. The reasoning was sound on the information available, and sound reasoning loses sometimes. If we let today rewrite that, we'd be learning the wrong thing twice in one week.
What it cost
The strategy finished a couple of dollars up across its life. Two months of compute, a fair amount of attention, and a headstone.
What it bought: a sizing rule that now protects everything else we run, a tested precedent that our thresholds are real, and a slightly clearer idea of how a pairs trade dies — not gradually, but all at once, in the direction the model says is least likely.
Two dollars and thirty-six cents is a strange price for that. It's also the cheapest that lesson was ever going to be.