How our AI works

McCortex OI is built around a simple premise: markets generate more information than a person can track in real time, but a person should still decide how much of that information gets turned into risk. This page explains how our AI trading actually works, from the data that feeds the machine learning models to the moment an order reaches the market, and where the limits of the approach sit. Nothing here is a promise about what the models will find. It is a description of the pipeline itself.

What data goes in

Every model in McCortex OI starts from the same raw material: market data. That means price and volume series across spot and futures markets, order book depth where it is available, and the timing relationships between assets that tend to move together or diverge. The workspace pulls this data continuously rather than in batches, because a model trained on stale inputs is only useful in hindsight.

Data quality matters more than data volume. Before anything reaches a model, feeds are checked for gaps, duplicated ticks and obvious feed errors, since a single bad print can distort a signal for hours if it is not caught early. We also normalise data across the markets we cover, spot and futures, long and short setups alike, so that a model trained on one instrument behaves consistently when it is pointed at another with similar characteristics.

What we do not do is feed models on sentiment scraped from social platforms or unverified news wires. That kind of input is noisy and hard to audit, and McCortex OI is built around inputs that can be checked and reproduced. The result is a narrower but more dependable base: price, volume and order flow, observed carefully rather than guessed at from adjacent signals.

What the models are actually doing

At the core of signal generation are machine learning models trained to recognise recurring structure in price and volume behaviour, the kind of patterns that show up before certain moves, without any claim that they show up every time. A model does not know the future; it estimates how similar conditions have tended to resolve in the data it has seen, and it updates that estimate as new data arrives.

We deliberately avoid describing model architecture in detail on this page, not because it is a secret, but because architecture details age quickly and rarely help a trader decide anything useful. What matters operationally is that models are retrained on a schedule, tested against data they were not trained on, and monitored for drift, meaning a model that starts behaving oddly against fresh data gets flagged before it can influence live signal generation.

Backtesting is part of this cycle, but it plays a narrower role than marketing language sometimes suggests. It is a check on whether a model’s logic holds together against historical conditions, not a preview of future returns. We do not publish backtest returns or accuracy figures on this site, because a backtest result is a description of the past, and treating it as a forecast is exactly the kind of shortcut this page is trying to avoid.

From signal to order

A signal on its own is just a statement of probability, so McCortex OI translates that statement into a concrete order through a defined execution logic layer. This layer decides direction, long or short, sizing relative to the account’s configured limits, and the type of order to place, before anything reaches the market. Execution happens on our servers rather than on the trader’s device, which is why a strategy keeps running through a dropped connection or a closed laptop.

Execution logic also handles timing. A model might generate a signal the instant conditions shift, but placing an order at that exact instant is not always sensible if liquidity is thin or the market is moving erratically. The execution layer applies basic checks around slippage and order type so that a signal is expressed as an order in a way that is workable, not just theoretically correct.

Throughout this process, the trader retains control over the strategy library itself: which tested strategies are deployed, on which markets, and with what boundaries. The models generate and refine signals; execution logic turns accepted signals into orders. Neither step removes the trader’s earlier decision about which strategies to run in the first place.

The risk gate before execution

This is the step we consider most important, and it sits before the order goes out, not after. Every order generated by a strategy passes through a risk gate that checks it against the limits the trader has set: maximum position size, exposure per strategy, and boundaries on how much of the account a single trade or a single strategy can touch. If an order would breach one of those limits, it is stopped or resized before it reaches the market, not flagged for review afterward.

This ordering matters. A risk system that only reports on breaches after they happen is a record-keeping tool. A risk system that intervenes before an order is placed is a control. McCortex OI is built on the second kind, because the point of automation is not just speed, it is consistency in applying limits that a person might otherwise skip in a fast-moving market.

Risk limits are configurable per strategy rather than fixed platform-wide, which means a trader running several strategies at once can size each one differently based on how much of the account they want it to touch. Alerts on limit breaches and fills are configurable as well, so the trader stays informed about what the risk gate is doing without needing to watch the screen continuously.

What this approach cannot do

It is worth being direct about the boundary of all this. Machine learning models describe patterns in data that already happened. They do not predict the future, and no combination of signal generation, execution logic and risk limits changes that basic fact. A model can identify that certain conditions have often preceded certain outcomes; it cannot guarantee that the next occurrence will resolve the same way, and McCortex OI does not represent it as capable of that.

Markets also change in ways that historical data cannot fully anticipate. New liquidity conditions, structural shifts in an asset, or simply periods that look unlike anything in the training data can all produce results a model was not built to handle well. That is precisely why risk limits sit ahead of execution rather than being treated as an afterthought, and why we retrain and monitor models on an ongoing basis instead of treating any version as finished.

Trading digital assets carries substantial risk, including the total loss of capital, and that risk does not disappear because a decision was made by a model rather than a person. Past performance and any illustrative figures do not guarantee future results, and nothing on this page, or anywhere on this website, should be read as investment advice. The purpose of McCortex OI is to apply discipline and consistency to a process that markets do not slow down for, not to remove the underlying uncertainty that trading always carries.