2026-10-28
The 77-file batch is up, including the 34 that never traded
Published the first full batch today. 77 unique files, 32 months of EURUSD H1, 17-point spread. The headline number is not the profit — it is that 34 of them opened zero positions and another 22 opened positions and closed none.
That means more than two thirds of this sample produced no trade record at all. Not a bad record. No record. And every one of them installs cleanly and shows a smiley face in the corner of the chart, which is the entire test most buyers run.
Full table is in the database. Nothing is filtered out of it.
2026-10-27
Why the tail cut is 5%, and why it flips so many results
The test deletes the best 5% of closed trades by profit and re-sums the rest. It is crude — there are better robust statistics — but it answers one question fast: does this thing earn from its logic, or from a handful of lucky closes?
Five of the EAs in this batch finish positive. Remove the top 5% and three of them go negative. The worst case swings by $8,766 on a $10,000 account.
A system that depends on 5% of its trades is not a system. It is a bet that those trades recur.
2026-10-26
An EA called 'One Trade At A Time' opened 2,954 positions at 200 lots
The file is 8_一次一单EA.ex4 — literally "one trade at a time". In the tester it opened 2,954 positions at a fixed 200.00 lots. On a $10,000 account. It never closed any of them, so the tester reports zero P/L and no error.
This is the clearest example in the batch of why the file name, the marketing copy, and the tester report are three different documents. Read the third one.