At 55% wins, a run of six losses over a few hundred trades is close to certain, and most people switch the EA off during it. Work out the run you should expect before it arrives.
Each trade is an independent loss with probability 1 − win rate. The figures above are exact: a distribution over the current run length is walked forward one trade at a time, which gives the probability of never having reached a run of k losses in n trades. Subtract from one and you have the chance of at least one such run. The expected longest run is the sum, over every k, of the probability of reaching k — that identity is worth remembering, because it is why a long enough sample always produces a run that feels impossible.
No simulation, so no error bars on these numbers. The assumption that trades are independent is the weak point, and it is a real one: a grid EA does not lose trades independently, it loses them in one event. For those the number above is optimistic, and the honest tool is the lot sequence rather than the win rate.
From the data. EA 0037 survived its audit with a positive result: 59 opens, 56 closes, +$5,175 on a $10,000 start, 15.3% drawdown. It also had eleven consecutive losses totalling −$1,716 at a fixed 0.1 lots. Nothing was wrong with the EA during those eleven trades. The win rate had not changed. That run is the shape of the win rate, and the only decision available was how much it cost. The full audit is here.
The useful move is to size so that the run you just computed is survivable, then stop looking at the streak. If the expected run is nine losses and you risk 2% a trade, that run costs 16.6% of the account — acceptable. At 5% a trade it costs 37%, and the recovery from there needs a 59% gain, which is the other tool on this page.
What does not work is treating the streak as information. Switching an EA off after six losses and on again after two wins is a strategy that only sells trades to your broker: you take the losses you were always going to take, and give up the wins that were also always going to come, and pay the spread on all of it.
There is one case where the streak is information: when the lot size rises during it. A martingale doubles into the run, so eleven losses at 1.4x per step is not eleven risks, it is one risk of about 50 to 1. That is the failure mode in lab note 004, and no streak calculator can price it, because the trades are not independent any more.
Related: drawdown recovery · position size · seven checks for an EA that stopped