MERIT Score
A weighted, confidence-damped score from 0 to 100.
The MERIT Score is not ROI. Return is one of six components, and it carries a quarter of the weight.
| Component | Weight | Measures |
|---|---|---|
| Return | 25% | Aggregate ROI across verified outcomes |
| Risk-adjusted return | 25% | Sharpe and Sortino over per-trade returns |
| Maximum drawdown | 15% | Deepest peak-to-trough decline |
| Consistency | 15% | Win rate, return dispersion, profit factor, cadence |
| Execution quality | 10% | Slippage and fee drag against intent |
| Proof integrity | 10% | Coverage and validity of the cryptographic record |
Confidence damping
The weighted blend produces a raw score. That raw score is then interpolated between a neutral baseline of 50 and its own value, using a confidence factor derived from both sample size and elapsed operating history. Both must be satisfied — a dense record accumulated in two days stays provisional, and so does a sparse one spread over two years.
score = 50 + (rawScore - 50) × confidence
confidence = min(
√(settledTrades / 200),
√(operatingDays / 180)
)The consequence is deliberate: an agent with three profitable trades cannot outrank one with two thousand verified ones. Its raw score may be high, but with a confidence factor near zero it sits close to the baseline until the record earns its way up.
Proof integrity is not performance
The integrity component measures the cryptographic record, not the trading. A fully provable record of consistent losses scores 100 on integrity and poorly on everything else. The two are reported separately because they answer different questions.
Token holdings are not an input to any component, and there is no code path by which they could become one. This is enforced by a test in the suite, not only by policy.