The picture everyone has of a blown account is a margin call: one violent move, the balance goes to zero, the terminal prints a red line. Across 190 free EAs on the same 32 months of EURUSD, that happened 14 times. The other 41 accounts that ended below $1,000 were not stopped out. They kept trading, on purpose, all the way down.
One dot per file, placed by the balance it finished with. Log axis, because the difference between $6 and $114,134 does not fit on a linear one.
Red: finished under $1,000. Grey: finished between $1,000 and the starting balance. Green: finished above it. 73 of the 190 dots are in one of the two tails.
Treat the 190 accounts as one portfolio that started with $1,900,000 and ended with $1,449,309 — a net $450,691 gone, -23.7%. That is not because half of them exploded. It is because the median file lost a little and the bottom decile lost everything: p10 is $157.1.
Closing balance at each percentile across all 190 files that closed a trade. p50 = $8,067.
The ten worst closing balances. All ten traded hundreds or thousands of times, and every one of them finished its last trade and asked for another.
| # | File | Closed at | Return | Trades | Win rate | R:R |
|---|---|---|---|---|---|---|
| 257 | EMA 6.12 | $6 | -99.9% | 537 | 56.1 | 0.61 |
| 29 | FarhadCrab1 | $11 | -99.9% | 5,270 | 100.0 | — |
| 179 | open_close&stochastic_strategy | $12 | -99.9% | 559 | 26.8 | 2.14 |
| 50 | Trader v7.51 MACD | $13 | -99.9% | 3,725 | 56.2 | 0.36 |
| 239 | up3x1_premium_v2M | $14 | -99.9% | 1,972 | 32.5 | 1.53 |
| 112 | Open_close EA drawn on the chart | $18 | -99.8% | 497 | 24.1 | 1.32 |
| 51 | Trader v7.51 RIVOT | $19 | -99.8% | 3,159 | 60.0 | 0.19 |
| 259 | Two PerBar | $20 | -99.8% | 239 | 36.0 | 1.04 |
| 86 | Backbone | $20 | -99.8% | 1,117 | 8.7 | 3.86 |
| 160 | Trix EA | $21 | -99.8% | 6,176 | 25.4 | 1.54 |
Look at the win rates in that table: 24%, 26.8%, 33.7%, 56%, 100%. A file with a 100% win rate is in there, at $11. Win rate does not sort winners from losers — position size and what happens after the losing trade does. The last one in the list, Two PerBar, closed 6176 trades to lose $9,980 of $10,000. That is a strategy with a small negative edge and a lot of patience.
Grouped by number of closed trades over the 32 months. Same symbol, same period, same spread.
| Trades | Files | Median return | Median closing balance | Median max DD | Under $1,000 |
|---|---|---|---|---|---|
| under 50 | 39 | -3.2% | $9,677 | 16.2% | 3% |
| 50 - 150 | 26 | -3.7% | $9,629 | 15.2% | 12% |
| 150 - 400 | 35 | -14.1% | $8,592 | 39.1% | 23% |
| 400 - 1,000 | 46 | -18.8% | $8,124 | 20.9% | 35% |
| 1,000 + | 44 | -42.7% | $5,727 | 80.0% | 30% |
This is the finding I did not expect. The files that traded the least — under 50 trades in 32 months, which is barely a trade a month — came out at a median −3.2%, with 3% of them under $1,000. The files that traded more than 1,000 times came out at −42.7%, with a median drawdown of 80% and 30% of them under $1,000.
Of the 41 accounts that ended below $1,000, 37 had taken at least 200 trades. Of the 32 that finished above their starting balance, 18 had taken 100 or fewer. The correlation between log(trades) and return across all 190 is -0.148 — weak on its own, which is the honest way to report it. The relationship is not a straight line; it is a floor that falls away once an EA is trading constantly.
Two readings, and both are probably true. An EA that trades hundreds of times is usually a scalper paying the spread 537 times, and 17 points a round trip is a real cost. And an EA that never stops is an EA with no exit condition — it does not stop because it cannot stop, it can only run out.
Across all 190 files there were 14 stop-outs. Not fourteen accounts destroyed — fourteen events, and the median account that experienced one still had $1,185 in it afterwards, after a median of 164 trades.
Only 6 of the 41 accounts that ended below $1,000 had a stop-out at all. The other 35 simply kept opening positions with money they still had. A stop-out is a broker safety net, not the mechanism by which a bad EA loses money. Losing money is a decision the EA makes every hour, and the account balance is just the scoreboard catching up.
The 32 files that finished above $10,000 have a median of 65 closed trades and a median maximum drawdown of 16.1%. The 41 that finished under $1,000 have a median of 537 trades and a median drawdown of 98.7%. One group trades rarely and keeps its equity steady; the other trades constantly and does not.
That is also the trap in reading this page as a recommendation. 65 trades is not much evidence. Several of the survivors are up because of one position that went their way, and a file that trades 65 times in 32 months can look excellent for two years on nothing but luck. Lab note 005 filters for the ones that survive that test as well — positive, under 20% drawdown, at least 30 trades — and 190 files produce 11 of them, best case +0.48% a month.
Every file was compiled in MT4 build 1492 and run on EURUSD H1, February 2024 to October 2026, $10,000, 17-point spread, open prices only. Closing equity is read from the final account statement, not calculated. The CSV behind this page lists opens, closes, return, maximum drawdown, win rate, R:R, closing equity and stop-out count for all 372 files: batch-b.csv.
What it does not prove: that these EAs are bad in general. It proves they are bad on this symbol, in this period, at this spread, with default settings. Change any of those and the numbers change — that is the subject of another note. It also does not prove a file is malicious or broken. Most of the 41 are simply strategies with a small negative expectancy, run for two and a half years, which is enough time for a small negative expectancy to become a small balance.
The aggregate is the argument. 190 files, one harness, $1,900,000 in, $1,449,309 out. If free EAs were roughly break-even with a wide spread, the total would land near the starting capital. It landed $450,691 short.