There is no “best EAs” list here, and no Scalper / Grid / Martingale tabs. What is below is one measured dimension — closed trades per month — because that is the one thing the harness observed for all 372 files rather than for the ones with a guessable name. Across the four buckets that traded, median return falls from the slowest to the fastest and median drawdown rises — the grouping is the finding, not just a label on it.
Every site in this niche sorts EAs into Scalper, Grid, Martingale, Trend and News, and every one of them is guessing from the file name. We have the source for all 372; we tried to sort them that way. A keyword pass over the 375 .mq4 files matched grid parameter names in two files and martingale parameter names in two — in a batch of 372 free EAs, where grids are not rare. Either this pile contains almost no grids, or the parameters are named nothing a keyword can find. It is the second one: a behaviour test for the same thing — reward-to-risk under 0.4 with a win rate above 70%, the shape of a strategy that loses big exactly once — finds 11 of the 190 files that traded. So the real grid population is at least five times what the source text says, and a label generated from a title would be wrong for most of it.
A taxonomy where most entries read “unknown” is a worse lie than no taxonomy. So the buckets are behaviour, the cut points are round numbers, and every count below can be re-derived from the public CSV in one line of arithmetic. When a source-parsed strategy type is ready it will be added as a second layer with the confidence stated per file — not as a folder name invented from a title.
One cross-cutting fact that does not get a page: 12 of the 190 files that traded did it without a stop loss, and every one of them shows a win rate above 85%. That is not a strategy category, it is a defect, and it is listed on the shortlist page where the three worst examples sit.