Calculation and raw data

The scale

Internally we compute on a scale from 1 to 5 with one decimal. We publish the value times two, so 0 to 10. That is why grades move in steps of 0.2: 7.4 and 7.6 exist, 7.5 does not. The overall grade is the weighted sum of the six criteria, rounded to one decimal.

Six criteria and their thresholds

The formula applies to every EA without a written review. 31 EAs with a full written review keep their hand-set sub-grades; the rules further down apply to them just the same.

Rules that apply to every grade

1. Live accounts published by the vendor or AlgoTradingSpace are chosen by the seller. They earn at most 8.0 for evidence. Only our own account in our EA park earns more.
2. Martingale without a proven hard stop: risk control at most 5.0 and overall at most 6.0, however good the rest is. We call that the cap. The EA’s page shows it next to the grade and says where the grade would stand without it.
3. An AI agent from a different model family grades every EA blind on the same six criteria, without seeing our grade. Where a criterion differs by more than 1.0, the mean of both reviews counts until a human re-checks; caps from rules 1 and 2 still apply. Reviewed so far: 131 EAs, 408 deviations, in 398 of them the second reviewer was stricter. It sees less than we do, though: no installation, no setfiles. Taking its value alone would turn missing information into a penalty.
4. We do not grade the operator’s own products. They are therefore missing from the medians and the raw data.

Medians per group

The groups differ mainly in whether a live account is on record. The last column computes the grade without evidence, the other five criteria re-weighted to 100%. Where the groups sit close together there, the gap comes from the evidence, not from the seller.

131 graded EAs, every value on the 0 to 10 scale. “Review” means hand-set sub-grades from a written review, “formula” means computed as above. No live figures.

* Cap applied (rule 2).