GAIA ORGANISM
An evidence-gated machine learning organism for market microstructure -- filtering high-frequency noise through CUSUM event sampling to emit directional outcome probabilities locked to a 30-minute lead-time law.
Conformal prediction threshold holding tight bounds
Signals valid before market moves occur
ROCm memory-pinned high throughput inference
Core Architectural Principles
Replaces standard time-series bars with volatility-driven event sampling to eliminate non-informative tick noise.
Eliminates overlap leakage across sequential labels to guarantee reliable out-of-sample performance validation.
Capital sizing is capped at Quarter-Kelly fractional allocation with automatic regime kill-switches.
Model publishes predictions only when uncertainty bounds are strictly met -- remaining idle during ambiguous regimes.