How the Grade Works
The methodology, published in full — because a trust product with a secret methodology isn't one.
The five gates
Every agent is backtested on historical data with a strict train/test split, then judged on the out-of-sample portion only. Five hard checks decide whether it passes the gate at all:
The agent must make money on data it never saw during construction. In-sample results are ignored — they only prove the strategy memorized the past.
Net expectancy must beat 1.5× the full round-trip cost (fees + slippage). A strategy that only wins before costs is a donation to the exchange.
The edge must hold on a meaningful share of the universe it trades (≥60%), not be carried by one lucky symbol.
Simulated at responsible position sizing, the worst drawdown must stay under 35%. An edge you can't hold through its losing streak is not an edge you have.
Performance can't collapse between the training and test periods — a large in-sample/out-of-sample gap means the parameters were curve-fit, not discovered.
The letter
Agents that pass all five gates are scored on Edge, Robustness, Risk, and Execution, and the weighted result maps to a letter. The scale is deliberately harsh:
- A / B — strong, robust edge. Rare. Expect to almost never see an A.
- C — a real but modest edge; tradeable with disciplined sizing.
- D — passes the gate with a thin edge; high drawdown or fragile.
- F — fails the gate. Most first drafts land here. That is the product working, not failing.
The headline metric is never win rate
Tessen headlines net expectancy per trade (in basis points, after costs), sample size, and risk:reward. Win rate is shown only as context — a 78% win rate with 1:0.3 risk:reward is a losing strategy wearing a good number, and we have the six-year backtest to prove it.
We graded ourselves first
The first strategy ever graded by this engine was our own. It passed all five gates and earned a D — a real but thin edge with a ~33% simulated drawdown. We published that instead of hiding it, because the alternative — a platform that grades its own strategy an A — is exactly the kind of product Tessen exists to replace.