Most people set a daily loss limit the way they set an alarm clock. Some round number that felt responsible on the evening they configured the profile, never revisited, and never costed. It sits there doing something to your results and you have no idea what, which is an odd position to be in about a rule you chose.
The rule is not free. A stop at the day level does two things at once: it removes some of your worst days, which is the reason you wanted it, and it also ejects you from days that were bad in the morning and fine by the close, which is the part nobody budgets for. Whether the trade is worth making is an empirical question about your specific profile, and the backtest is the cheapest way to ask it.
What the replay will and will not do for you here
Be clear about the tool before you lean on it. The Backtest a Profile panel on Autopilot takes three inputs, a profile, a number of days, and a start equity, and it replays historical signals against that profile with no real orders. The panel states that it honors all filters, sizing, leverage caps, and risk guards, which is the sentence that makes this whole exercise possible: whatever guards are configured on the profile are applied during the replay, so changing a guard and re-running gives you a genuine before and after.
What I cannot tell you from the panel is whether a daily loss limit is one of the guards available on your profile, under that name or any other. Open the profile editor and look before you plan the experiment. If there is a day level loss control, this article is a recipe. If there is not, the decision does not disappear, it just moves to you: the cap becomes something you enforce by hand with the pause control, and the numbers below are still the ones that tell you where to set it.
Running the pair so the comparison means something
The whole value of this is in changing exactly one thing, and it is easy to accidentally change three.
Run both tests on the same day. The days field counts back from whenever you press the button, so a run this morning and a run next Tuesday cover different windows and the difference you measure will be partly calendar. Use the same day count and the same start equity in both, and use your real account size rather than a default, since a cap expressed in dollars behaves completely differently against four thousand than against forty.
Pick a window long enough to contain at least one genuinely ugly stretch. A cap is a tail control, and a period with no tail in it will report that your cap costs you money and protects you from nothing, which is true of that window and useless as guidance. If the longest window you can run is calm, say so in your notes and treat the result as provisional.
Then run it twice. Cap off, record. Cap on at your candidate number, record. Nothing else touched.
The four numbers to write down
Total return is the number everyone looks at and the least informative of the four. Take these instead.
- The total result of each run, so you know the headline cost or gain.
- The worst single day in each run, in dollars. This is the tail the cap is supposed to be cutting, and it is the reason the cap exists.
- The trade count in each run. A cap that halts the day removes the rest of that day's entries, and the drop in count tells you how much participation you sold.
- The number of days the cap fired. If it is zero, you have not tested anything, you have simply set the cap outside the range of your own history.
Suppose you run a twelve thousand dollar profile over one hundred and eighty days with a candidate cap of three hundred and sixty dollars, three percent of the account. The uncapped run does 214 trades with a worst day of nine hundred and forty dollars. The capped run does 187 trades, the worst day is three hundred and eighty, and the cap fired on nine days out of one hundred and eighty.
Now you have something to reason about. You gave up 27 trades, you cut the worst day by about sixty percent, and the control engaged on roughly one day in twenty. Whether the headline result went up or down, the shape of what you bought is clear, and it is a shape you can hold against your own tolerance rather than against a rule of thumb.
Reading the difference honestly
Two traps live in that comparison and both of them flatter the cap.
The first is that a daily loss cap is a fantastic thing to fit to history. Set it just above your second worst day and the backtest will show you removing the worst day at almost no cost, which is not a strategy result, it is the definition of a number chosen after seeing the answer. The defence is to pick the cap from a rule you can state before you look, such as a fixed percentage of equity, or a multiple of your average losing day, and then report where that lands rather than searching for the level that reads best.
The second is that nine firings across one hundred and eighty days is nine observations. It is not enough to distinguish a cap that reliably catches bad days from a cap that happened to catch nine. Before you treat the comparison as settled, look at what happened on those nine days after the halt. If the days that triggered the cap generally kept getting worse, the control is doing what you hoped. If several of them recovered into the close, you were stopped out of the day rather than saved from it, and the cost of the guardrail is higher than the trade count alone suggested.
When you keep the cap even though the number says it costs you
Here is the case for setting a cap that the backtest scores as expensive, and it is a serious one rather than a consolation.
The replay places no real orders, and the module says so directly on the same screen where it warns that with paper mode off it places real orders on your connected exchanges, that execution prices may differ from signal prices because of market conditions and latency, and that you are solely responsible for every trade. The recommended path is paper first, verify the behaviour, then live with small sizes. A simulated bad day costs nothing and produces no urge to intervene. A real one, watched in the afternoon on an account that matters to you, produces a very strong urge, and the way that urge usually resolves is that someone turns the profile off in the middle of the drawdown and turns it back on after the recovery. That behaviour is far more expensive than any cap.
So the cap has a second job that no backtest can score, which is keeping you from becoming the discretionary risk manager of a system you built specifically so you would not have to be. If a three hundred and sixty dollar day is the point at which you would start interfering, then that is the cap, and the trades it costs you are the price of the profile surviving contact with you.
The failure mode to plan for either way is the one the cap does not cover. A day level limit responds to losses as they accumulate during a session. It does nothing about a gap that arrives before the session opens and takes the position straight through its stop, and it does nothing about a feed that stops updating so the profile never registers the loss at all. Autopilot puts an emergency kill switch and a pause control at the top of the page for exactly the situations no configured guard anticipated. Know where they are, and check once a month that the profile you think is enabled actually is, since the header will tell you how many of your profiles are running and how many are in paper.