JARVIS
Methodology

Engineered for survival, not narrative.

Jarvis was built by an operator who refused to deploy capital against models he could not stress-test, regimes he could not classify, or fills he could not audit. The platform is the codification of that discipline.

Walk-forward validation

Every parameter set is back-tested across rolling out-of-sample windows. The production V38 configuration was validated continuously against the May 2025 – May 2026 window with no peek-ahead, no parameter tuning on the test set, and a final 30-day forward dry-run.

Regime-aware capital allocation

Markets are not stationary. Trend-following and mean-reversion are mutually destructive when run together. Jarvis classifies the prevailing regime monthly and switches between two purpose-built sub-strategies, each tuned for the conditions it dominates.

Defense first

Hard 1.5× ATR stops, 24-bar time stops, 60% max gross exposure, per-asset leverage caps mirroring exchange limits. The system is designed to be small in the worst month, not large in the best.

Validation period
12 months
Out-of-sample method
Walk-forward
Universe
9 perpetuals
Bar size
4 hour
Round-trip cost assumption
8 bps
Maximum gross exposure
60%
Stop methodology
1.5× ATR
Max hold (CHOP)
24 bars