Probability fields derived from WebGL and mathematics.

EDGX GPU Field Engine runs a real-time WebGL probability field renderer alongside a BiRNN temporal network, 20-asset AgentNet, GJR-GARCH volatility engine, ROCS calibration system, and APEX 5-algorithm ensemble — all from a single HTML file with zero external dependencies — covering 26 assets across four time horizons.

Edgecore.Ltd designs and architects complex financial applications from the ground up — transforming granulated market data into structured intelligence pipelines that operate at institutional precision and scale. Every system is grounded in custom-engineered mathematical and scientific frameworks: from proprietary volatility engines and Bayesian inference models to multi-regime detection algorithms that classify market states with quantified, reproducible certainty. Deep learning modules — including bidirectional recurrent networks, ensemble neural architectures, and adaptive retraining schedulers — are built without black-box dependencies, ensuring every weight, gradient, and decision boundary remains mathematically traceable from raw feed to final signal. Whether the scope demands probability field renderers, prediction modelling systems, or calibration layers derived from GJR-GARCH dynamics and Ornstein-Uhlenbeck mean-reversion theory, Edgecore.Ltd translates the frontier of quantitative finance into deterministic, production-grade software.

Every module derives from live Binance data

Eighteen engines, zero synthetic data, zero Math.random() in any prediction or learning path. Every constant is a named value. Every weight is initialised from a deterministic xorshift32 PRNG seeded from the asset symbol hash.

BiRNN + AgentNet → Bayesian direction vote

Two independent neural networks feed into a Bayesian weighted majority vote gated by three conditions: ATR noise floor, factor consensus threshold, and HMM regime suppression.

Live direction predictions across all 26 tracked assets

buildAnalyticalPredictions() runs in the background for all 26 MA_ASSETS. Each asset gets BiRNN delta estimates (T1–T3 enhancements) plus computeDirectionVote() gating and PARCE recalibration across 2m / 5m / 10m / 30m horizons.

GJR-GARCH + Kalman + Fisher scoring

ROCS tracks per-factor directional accuracy via exponential moving average across 28 named factors, fitting GJR-GARCH volatility and calibrating predictions with Fisher scoring, Hedge bound (LOG_N_FACTORS=ln(28)) and stationarity guards on every kline.

GPU engine component map

All components exist as named functions and constants in a single-file HTML/JS build. No external ML libraries for agent training — pure-JS backpropagation with AdaGrad per-weight learning rates and deterministic PRNG seeding.

On-chain signal minting and monitoring

The Signal Forge converts prediction events into on-chain NFT signals via Polygon / ERC-721. sfScoreActive() combines PredCache, APEXCache, and AgreementState into a unified score ∈ [0,1]. Auto-monitoring runs across all 26 MA_ASSETS. IPFS metadata includes timestamp, asset, direction, confidence, GARCH σ, and HMM posterior.

Live accuracy scoring across all horizons

Every prediction is logged with refPrice, predPrice, horizon, and dirCorrect flag. Settlements score against actualPrice when the horizon elapses — directional accuracy tracked separately from magnitude error.

4-trigger precision retraining — never every minute

The scheduler evaluates four independent conditions on every minute tick. Any single trigger fires a retrain — subject to an 8-minute minimum gap to prevent thrashing on the same data.

Request more information or explore partnership opportunities

Whether evaluating the platform, exploring a commercial arrangement, or interested in a research or technology partnership, complete the form below and all enquiries will be reviewed directly.

One file. Full engine.

The entire EDGX GPU Field Engine Forge runs as a single self-contained HTML file — WebGL shaders, BiRNN (T1–T3 enhancements), AgentNet, ROCS, APEX, HMM, GJR-GARCH, Haar Wavelet, VPIN, PARCE, Agreement Engine, Signal Forge, paper trading modal, and 26-asset analytics dashboard.