$1,000 mainnet experiment
3× leverage · total exposure up to 3× current equity · variable sizing · funding excluded
No policy passed the evidence check
The sizing and exit engine is implemented. It currently proposes no new positions. No exchange orders have been submitted by this engine.
What determines a position
The forecast is adjusted for historical forecast errors and uncertainty. Larger downside reduces the allocation. Simultaneous losses across coins consume the same risk budget. Total positions can reach $3,000 initially, but the engine can leave cash unallocated.
Stop and target distances are learned from earlier price paths and scaled to a coin’s current volatility. The holding period is selected from measured horizons, rather than automatically set to 24 hours. These are experimental rules, not proven optimal exits.
What the study found
529 recorded signals yielded 496 complete eligible paths. The split used 198 for training, 123 for calibration and 135 for the final audit; 40 paths were excluded at boundaries to prevent overlapping outcomes leaking forward.
| Horizon | Stop / volatility | Target / volatility | Mean / trade | Mean / period | Lower estimate |
|---|---|---|---|---|---|
| 1h | 0.235 | 0.974 | −0.27% | −0.31% | −0.50% |
| 4h | 0.423 | 2.130 | −0.07% | −0.14% | −0.53% |
| 24h | 1.208 | 3.485 | −0.60% | −0.53% | −1.38% |
| 48h | 1.961 | 7.369 | +0.64% | −0.44% | −2.34% |
A period is a 48-hour block. Giving periods equal weight reduces the influence of busy, correlated bursts of signals. The uncertainty estimate also accounts for comparing four horizons. No lower estimate was positive, so no candidate was admitted for trading.
The final concurrent audit book stayed at $1,000 with zero trades. This is abstention, not proof of profitability. The history was already studied before, so these results are retrospective diagnostics, not fresh prospective evidence.
What remains before live trading
A qualifying strategy still needs prospective measurement and a broker that reconciles fills, persists position rules and verifies exchange-side exits. Hourly candles do not validate liquidations, outages or actual execution quality. The current study uses estimated costs of 0.1% per side; real costs can differ.
Funding is excluded from the new policy and its reported performance. Hyperliquid still charges or credits funding, so actual account equity must be read from the exchange.