One score. Two outputs. Every decision onchain.
0 labeled synthetic traders replayed over ~100 days of real Phoenix prices, plus live accounts. Each day the engine updates every trader's skill posterior and proposes a real-copy multiplier M and a profit share S; the keeper posts them with post_epoch and the program applies its guardrails.
Firm P&L on the same traders
Fees pay for luck; hedging pays for skill. A B-book bleeds as soon as skilled traders get paid. Copying everyone pays for every lucky and losing trader. PEAK backs traders only as evidence builds, and raises their share only after the firm is repaid.
Cumulative firm P&L by day
Traders · 0 funded · 0 breached
| Trader | Status | Score | M | S | Deficit | Paid |
|---|
Trader detail
How the engine decides
1 · Skill, with honest uncertainty
Each trader's daily Sharpe is estimated with empirical-Bayes shrinkage toward the population prior N(−0.08, 0.07²): a few lucky days barely move it. Per-trade markouts add evidence faster than daily P&L alone.
Standard metrics only: Sharpe, Sortino, Probabilistic and Deflated Sharpe, profit factor, drawdown, concentration.
2 · Real capital: the h* hedge ratio
The first term hedges the payout the firm owes (it rises toward S as confidence grows). The second is a mean-variance bet on the trader's alpha. M ramps up slowly (+0.1x/epoch), can be cut fast, and total real exposure is capped by the treasury.
3 · Share above the floor, only when earned
S rises above the purchased 80% only when the edge is confident and persistent, the firm's deficit on the trader is repaid (with headroom for the next payout), the treasury has room, and the firm stays net positive.
The program enforces it regardless of the engine: S ≥ floor, S > floor only if deficit ≤ 0, M ≤ 2.0x, S ≤ 120%, and at most ±0.25x / ±5% per epoch.
| Stage | M | S | Firm cost per $1 paid out |
|---|---|---|---|
| New trader, no evidence | 0x | 80% | $1.00 |
| Early evidence | 0.2x | 80% | $0.75 |
| Building confidence | 0.4x | 80% | $0.50 |
| Liability fully hedged | 0.8x | 80% | $0.00 |
| Alpha allocation | 1.0x | 80% | firm earns $0.25 |
| Elite, deficit recovered | 2.0x | 120% | firm earns $0.67 |
Per $1 of trader profit the firm keeps ≈ M − S. With an 80% share the pure hedge is 0.8x; anything above is a proprietary bet on the trader.
Honest scope: synthetic traders have an injected hit rate on real Phoenix price moves and are always labeled. Phoenix perps are mainnet-only, so on devnet the real copy is a shadow hedge: fills at M × size on the live book after a copy delay, with the exact Phoenix order instruction built by the Rise SDK. Scores are computed offchain; the program enforces the guardrails.