Arbitrage systems rarely fail at calculation. They fail at execution.
In crypto arbitrage and algorithmic trading infrastructure, execution risk — including liquidity gaps, exchange latency, and unstable orderbooks — is often the primary source of losses. Faster computation does not eliminate execution instability.
In HETHA.IO, AI-based execution forecasting is implemented through the AI Module — a neural network layer integrated directly into the arbitrage chain validation pipeline.
Importantly, the AI Module is not responsible for finding arbitrage opportunities. It does not search markets and does not generate trading signals. It operates strictly after an arbitrage chain has already been constructed. Its role is to evaluate whether that chain should be executed under current market conditions.
Why Execution Forecasting Matters in Arbitrage
Arbitrage profitability depends not only on price spreads, but on whether a chain can be closed within an acceptable time window, without slippage beyond model assumptions, and without breaking due to liquidity gaps.
Even a mathematically profitable chain can fail at execution. This is where execution forecasting becomes necessary.
Base Model: Execution Feasibility Filter
The Base neural network evaluates an execution coefficient reflecting the expected behavior of chain execution under current conditions. It takes into account market conditions, exchange responsiveness, internal system load, and historical execution behavior.
Its output is applied via the base_threshold parameter. If the predicted execution coefficient exceeds the defined threshold, the chain is blocked before being sent to the trading layer.
The Base model functions as a structural safety filter inside the execution layer, reducing the likelihood of incomplete chains, unstable execution, and exposure to adverse microstructure conditions.
Pro Model: Chain Closing Time Forecast
The Pro neural network focuses on execution duration. It forecasts how long a chain is expected to remain open before closure under current market conditions.
According to system documentation, the model uses 12 market indicators across multiple timeframes (1m, 15m, and 1h) to estimate closing behavior.
Its output is controlled through the pro_threshold parameter. Chains predicted to exceed the acceptable execution window can be excluded from execution.
This helps reduce prolonged capital lock, execution timing mismatches, and instability during volatile market phases.
Training and Data
Both neural networks were trained on thousands of previously executed orders within HETHA.IO. Training is continuous, allowing the models to adapt to evolving exchange behavior and liquidity patterns.
The AI Module does not guarantee profit and does not modify arbitrage logic. It operates as a risk-reduction layer designed to improve execution stability in live market environments.
AI Module Settings in HETHA.IO
AI Module configuration: Base and Pro execution forecasting parameters.
AI in Arbitrage Is About Stability, Not Prediction
In HETHA.IO, neural networks are not used to predict price direction. They are used to predict execution behavior.
This distinction is critical.
The AI Module does not increase theoretical profitability. It improves execution reliability — which, in real arbitrage systems, is often the primary source of performance variance.