Desk cockpit · backtest / test
Backtest
PAPER · advisory · no real capitalNot a final APY number — the FULL history: the equity curve with in-sample and out-of-sample explicitly separated, the drawdowns, and the gate behaviour (where kills fired, what was refused). The counterfactual shows what the deterministic gates SAVED. A degenerate or mock backtest is flagged, never laundered into a clean number. Paper / advisory, live from api.earn-defi.com, fail-closed. Not investment advice.
1 · Configuration
What was actually tested — the period, the assets & venues, the fee model, the AUM tier, and which kill-conditions were enabled. A backtest with the gates OFF is not the same experiment.
2 · Equity — in-sample vs out-of-sample
The in-sample span (where the strategy was fitted) is solid; the OUT-OF-SAMPLE span (held-out, the honest test) is distinct — dashed and shaded. The RWA-floor benchmark is the reference line. The out-of-sample boundary is never blurred: only the held-out part is a real test.
3 · Drawdown
The underwater curve — peak-to-trough depth over the run. Shown for the same window as the equity curve.
4 · Summary
5 · Monte-Carlo distribution
Block-bootstrap of the real daily yield series → the distribution of annualized return and worst drawdown across resampled paths. p5 / p50 / p95 are the confidence band. A single point estimate hides the tail; the distribution does not.
6 · What the gates saved
The counterfactual: the same book WITHOUT the kill-conditions vs WITH them. This is the desk’s identity made visible — the deterministic gates are not overhead, they are the edge. If the backend exposes no counterfactual pair, we say so (no fabricated saving).