You run a profile through the backtest, it comes back with three hundred trades spread across forty tickers over six months, and the number lands as a kind of promise. Three hundred chances. What it actually is, most of the time, is a description of an account considerably larger and more attentive than the one you have. The gap between the two is not a rounding error. On a small account it is often most of the run.
This is worth being precise about, because the usual reaction is to distrust the whole thing and go back to trading by hand. That is an overcorrection. The run is not lying. It is answering a question about a particular pile of money, and you have to tell it which pile.
The start equity field is doing more work than you think
The Backtest a Profile panel on Autopilot takes three inputs and nothing else. A profile, a number of days, and a start equity in dollars. The panel describes what it does plainly: it replays historical signals against that profile with no real orders, and honors all filters, sizing, leverage caps, and risk guards. So whatever limits you already put on the profile are in the run.
Which means the one thing the replay cannot know is how much money you have, unless you type it in. Leave a round default in the start equity box and the run sizes every position off that number. If the default is ten thousand and your account is four thousand, you are reading the record of a different trader. Every entry that fitted at ten thousand and would not have fitted at four thousand is still in the trade count you are looking at.
Counting the trades your money could have been in
Take the three hundred trades over one hundred and eighty days. That is roughly 1.7 entries a day. If the average hold is four days, then on a typical day the profile is carrying somewhere near seven positions at once, and on the busy days it is carrying far more than that, because entries do not arrive evenly. They arrive in clusters, when the engine reads the same condition across many instruments at the same time.
Now put a four thousand dollar account underneath it. Seven simultaneous positions means about five hundred and seventy dollars a position if you are fully deployed with nothing in reserve, and no sane person runs fully deployed. Call it four hundred. The clustered days are worse: fifteen open at once means two hundred and sixty a position, which on most instruments is below the size where the trade is worth doing at all.
So the honest reading of a three hundred trade run on a four thousand dollar account is not three hundred trades. It is however many trades fit through the narrowest part of your capital, and you can estimate it in ten minutes. Decide the smallest position you are willing to take. Divide your usable capital by it to get your real concurrency ceiling. Then walk the entry dates in the run and count how many entries arrived while that ceiling was already full. Those are gone. On the small accounts I have done this for, a quarter to a half of the trade count disappears, and it disappears unevenly, concentrated in exactly the clustered periods.
The minimum ticket nobody puts in the model
There is a second subtraction underneath the first, and it is the one that decides whether the strategy is viable at your size rather than merely smaller.
A two hundred and sixty dollar position on a venue charging a flat commission per order is a fee problem before it is a strategy problem. Two orders to get in and out at a fixed cost each will eat a meaningful percentage of a position that small, and it does so on every trade regardless of whether the trade worked. Forty tickers makes it worse, because forty tickers is forty different spreads, and the thin ones cost you more on entry and exit than the headline commission does.
Work out one number before you go further. Take your intended average position size, apply your venue's actual commission on a round trip, add a realistic spread cost, and express the total as a percentage of the position. Then compare it to the average gain per trade in the run. If the run's average trade is worth 0.8 percent and your friction is 0.5 percent, the strategy you tested does not survive the trip to your account, and no amount of retuning the profile fixes that. It is an account size problem with two solutions: fewer and larger positions, or a venue with proportional rather than fixed costs.
Re-running it at a size you actually have
The fix is not clever. Put your real number in the start equity box and run it again.
Before you do, tighten the profile itself so the replay has something to honor. The panel applies the profile's filters and sizing, so a profile carrying forty tickers will keep carrying forty tickers unless you narrow the universe. On a small account the productive edit is usually to cut the instrument list hard, to the handful with the tightest spreads and the best behaviour in the run, and to accept that you are now testing a smaller strategy. That is the point. You want the record of the thing you can run.
Then compare the two runs on three things and only three. The trade count, which tells you how much participation you gave up. The average trade, which tells you whether friction has eaten the edge. And the worst stretch, because the smaller account version does not just earn less, it also concentrates: fewer positions means each one is a larger share of the book, and the drawdowns get lumpier even when the strategy is identical.
On the days field, pick a period long enough to contain a bad market rather than a number that flatters. Thirty days of a friendly tape tells you almost nothing. If you can only get a short window, treat the output as a plumbing check rather than evidence.
What the shrunken run still will not tell you
Even a correctly sized replay is a replay. The panel is explicit that no real orders are placed, and the module carries a warning on the same screen that is worth reading slowly: when paper mode is off, Autopilot places real orders on your connected exchanges, execution prices may differ from signal prices because of market conditions and latency, and you are solely responsible for the trades. The stated route is to start in paper mode, verify the behaviour, then go live with small sizes. That sequence is the guardrail. Skipping it because the backtest looked good is the most common way people lose money on a profile that was fundamentally sound.
The specific failure mode to plan for is the clustered day, because that is where the small account and the automation disagree most violently. The run assumed every entry got filled at the signal price. In a real cluster your capital runs out partway through, the fills you do get are the earliest ones rather than the best ones, and the trades that would have carried the period are the ones you were too full to take. Nothing in the backtest shows you that, and the platform's own dashboard will not either, since the tiles report what happened rather than what was declined. Keep your own note of the days the account was fully deployed. After a couple of months that note, not the trade count, is what tells you whether the profile is the right size for the money you have.