Method
How the AT Score works
One score, the same for every account, with the formula in the open. Version 3.4, 20 October 2026.
The five parts
| Part | Weight | What it measures |
|---|---|---|
| Skill | 20% | Daily Sharpe ratio, a year's worth: 1 gives 39, 2 gives 63, 3 gives 78. |
| Return per risk | 20% | Yearly return divided by max drawdown (Calmar). |
| Drawdown | 25% | Worst of closed, floating and equity drawdown. 10% or less = 100, 60% or more = 0. |
| Worst day | 15% | Biggest one-day loss in % of the account. 5% or less = 100, 50% or more = 0. |
| Consistency | 20% | Share of green trading days, and the last 60 days against the whole record. |
Open positions count: floating profit or loss is part of the return and the drawdown.
Adjustments
- Survival: if the account was ever 50% or more under water, the score is cut (down to half at 100%).
- Experience: new records are multiplied by 0.8, rising to 1.0 at 3 years.
How much to trust a score
Score = trust × result + (1 − trust) × 40. Trust grows with trading days: days ÷ (days + 40).
| Trading days | Trust | How we read it |
|---|---|---|
| Under 5 | none | No score yet. The board median is shown, marked “projected”. |
| 10 | 20% | Mostly the starting value of 40. |
| 40 | 50% | Half result, half starting value. |
| 120 (about 6 months) | 75% | The score starts to mean something. |
| 250 (about 1 year) | 86% | Mostly the account's own result. |
The score counts trading days, not the number of trades. An account with 5,000 trades in 10 days is still a 10-day record.
Not in the score yet
Martingale and grid detection, stress tests on today's open lots, return against holding gold, and luck correction for many robots. Each new version is logged here.