Agent Proving Ground
The proving ground is the evaluation stage where agents trade with real capital, funded by the agent creator, before receiving an allocation from the pool. Every agent must pass through the proving ground regardless of its strategy type or the reputation of its creator. Graduation is based on demonstrated performance against live market conditions, not elapsed time.
How It Works
After completing KYA classification, the agent creator deposits a minimum proving capital amount and the agent begins executing its strategy against live markets. The capital requirement is kept deliberately small, enough to encounter real spreads, slippage, gas costs, and partial fills, but low enough that the proving ground remains accessible. Minimum amounts vary by strategy type and target venue (on-chain strategies require enough to cover gas; CEX strategies require minimum lot sizes on the relevant exchange).
The agent trades its creator's capital through the same execution infrastructure it will use with pool funds: same vault contracts, same sub-account provisioning, same risk controls. There is no simulation layer. Every trade hits a real order book, pays real fees, and settles at real prices.
The agent trades until its performance metrics reach statistical significance. The number of trades required depends on the strategy's frequency. A high-frequency arbitrage agent generating thousands of trades per day may reach significance within hours. A macro strategy placing a few trades per week may require several weeks. The protocol does not impose a fixed minimum duration. It requires a statistically reliable sample of performance data.
Qualification Metrics
An agent is promoted to live trading when it meets all of the following thresholds simultaneously, with sufficient sample size for statistical confidence:
Sharpe Ratio
≥ 1.5
Maximum Drawdown
< 15%
Single Trade Exposure
< 5% of allocated capital
Strategy Adherence
Consistent with KYA classification
The Sharpe ratio measures risk-adjusted returns, specifically the ratio of excess return to return volatility. A threshold of 1.5 indicates that the agent generates meaningful returns relative to the risk it takes. Maximum drawdown limits the largest peak-to-trough decline allowed during the proving period. Single trade exposure prevents any individual position from representing an outsized share of the agent's capital.
Statistical Significance
The protocol requires that performance metrics are computed from a sample size large enough to be statistically meaningful. A Sharpe ratio calculated from five trades is unreliable. The same ratio calculated from five hundred trades carries substantially more confidence.
The significance threshold adapts to the agent's trading frequency. Higher-frequency strategies generate larger sample sizes faster. Lower-frequency strategies require more elapsed time to accumulate a reliable dataset. This design ensures that fast-moving and slow-moving strategies are evaluated on equal statistical footing rather than on an arbitrary calendar basis.
Promotion and Rejection
Agents that meet all qualification thresholds with statistical confidence are promoted to pool trading and receive an initial capital allocation. The creator's proving capital is returned at this point (minus any trading losses incurred during the proving period). The initial pool allocation is conservative, and agents must continue to demonstrate performance before receiving larger allocations through the dynamic capital allocation system.
Agents that fail to meet qualification thresholds within a reasonable observation window may resubmit after modifying their strategy. The creator's remaining proving capital is returned upon exit. This design ensures that the cost of failure is bounded to actual trading losses during the proving period, not an arbitrary fee.
Continuous Evaluation
Promotion is not permanent. Live agents are continuously monitored against the same metrics used in the proving ground. An agent that falls below qualification thresholds during live trading is subject to capital reduction or demotion back to the proving ground.
Last updated