372 free downloads, 190 that actually traded, 32 that finished positive. Ask for a drawdown a person could live with and at least thirty closed trades and 13 rows survive, which is 11 distinct EAs. Three of them went onto a $10,000 demo today, on their own machine, with nothing edited.
Positive over 32 months, maximum drawdown under 20%, at least 30 closed trades. Duplicated files with identical results counted once. The shaded rows are the ones now trading.
| # | File | Closed | Return | Max DD | Chart |
|---|---|---|---|---|---|
| 310 | Daydream01 | 43 | +16.71% | 15.87% | EURUSD H4 |
| 145 | The new moving-average trading system | 64 | +12.88% | 17.71% | — |
| 137 | Virtual trailing stop | 103 | +5.41% | 10.28% | — |
| 252 | Currencyprofits_01.1 | 44 | +4.82% | 5.31% | — |
| 48 | MACD EA | 46 | +4.69% | 6.74% | — |
| 169 | Eugene | 32 | +4.69% | 16.23% | — |
| 61 | х1 | 76 | +3.61% | 6.02% | — |
| 118 | MA2CCI | 261 | +3.31% | 6.42% | — |
| 316 | SMC MaxMin at 1200 | 983 | +2.59% | 5.69% | GBPUSD H4 |
| 4 | VR Locker Lite - positive-lock strategy | 145 | +2.56% | 3.36% | USDCHF H4 |
| 274 | Channels | 50 | +0.71% | 10.45% | — |
Source: batch-b.csv, columns ret_pct, dd_pct, closes. Anything on this list that is not shaded is not running — not because it failed, but because three is what one machine can record cleanly.
Sorting by return would put the same kind of EA on all three charts: slow, low-drawdown, a handful of entries. That measures one hypothesis three times. The filter's whole purpose is to find files that behave 11 different ways, so the three were picked for trade frequency instead:
| Chart | File | Backtest | Max DD | Closed | What it tests |
|---|---|---|---|---|---|
| EURUSD H4 | Daydream01 | +16.71% | 15.87% | 43 | Rare, discretionary-looking entries — can a good-looking equity curve survive when it trades once a month? |
| GBPUSD H4 | SMC MaxMin at 1200 | +2.59% | 5.69% | 983 | High frequency, thin edge — the case where spread and execution decide the result, not the signal. |
| USDCHF H4 | VR Locker Lite | +2.56% | 3.36% | 145 | Locked pairs, small target — the strategy most sensitive to swap and to a broker that widens spreads on one side only. |
Every file in the batch was tested on $10,000 with its own default settings. A forward test on a different number is not a forward test of the same thing, and on this shortlist the difference is not academic: the default lot on Daydream01 is 1.00. One lot of EURUSD on a $1,000 account is 100:1 exposure on the first position, and a 15% drawdown measured on $10,000 becomes a margin call on $1,000 before the strategy has done anything wrong.
So the demo was funded to match the backtest, not the other way round. That also means the returns on this page are directly comparable to the ones on note 005 — same starting capital, same lot, same symbols.
Two data sources, and only one of them is trusted for money:
Closed trades come from the terminal log, which writes to disk on every fill: ticket, timestamp, side, lots, symbol, price. That record is complete and it is what the curve will be built from.
The equity readout is not trusted. MQL4's AccountBalance() through this Wine build returns a value that does not survive a unit check — the terminal shows $10,000.00 and the same call returns a number three orders of magnitude away. Until that reconciles, the equity curve is reconstructed from closed trades plus the price history, and floating P/L is labelled as an estimate. A sampler EA is running on a spare chart as a cross-check; it does not trade.
That is the unglamorous half of forward testing. Nobody's screenshots of a rising curve show you which number they read.
Every order the three EAs have opened on the demo, taken from the terminal log rather than from the account readout. 2 so far.
| Ticket | Time (UTC) | Side | Lots | Symbol | Price |
|---|---|---|---|---|---|
| 379792486 | 2026.10.29 06:40:56 | buy | 0.01 | USDCHF | 0.82885 |
| 379792487 | 2026.10.29 06:40:56 | sell | 0.01 | USDCHF | 0.82876 |
Two tickets one second apart on the same symbol is the lock: VR Locker opens both sides and waits for the spread to widen on one of them. It is a real mechanism, not a bug — and it is also why a forward test of this file needs months before the trade count means anything.
A forward test is only worth publishing if the ways it can fail are written down before the result is known:
The single machine dies. Three EAs share one $5/month VPS. If the host reboots or the terminal restarts, the EAs reload from their last saved state and any open position stays with the broker — the curve would then contain a gap I cannot reconstruct. The sampler exists to make a gap visible rather than comfortable.
Too few trades. Daydream01 closed 43 trades in 32 months, which is one every three weeks. Four weeks of forward testing would tell us almost nothing about it, and saying otherwise would be the same mistake the whole site exists to point at. The honest timeline is quarters, and the page will say how many trades it is standing on every time it is updated.
One symbol, one broker, one spread. All three run on the same demo feed. If SMC MaxMin fails here, that is evidence about this spread, not about the file.
The backtest was lucky. Each of these was selected because it passed a filter on the same 32 months everyone else failed. Some of the eleven are selected for luck by construction; that is exactly the hypothesis a forward test can kill.
What gets published: trade count, realised P/L and drawdown per EA, updated here, win or lose. If all three go negative the page stays up and says so — a free EA that failed forward is more useful to a reader than the nine hundredth screenshot of a curve with no trade list under it.