How our AI works

SynqoraX is built around a simple idea: markets generate more data than any single trader can watch, and machine learning is well suited to finding structure in that noise. This page explains, in plain terms, what happens between raw market data arriving on our servers and an order reaching the market. It does not promise outcomes. It describes a process, and every process has boundaries worth understanding before you rely on it.

What data goes in

Every strategy running on SynqoraX starts from market data: price series across the timeframes a strategy is configured for, order book activity, and volume patterns across the spot and futures markets we support. This data is collected continuously so that models are always working from a current picture rather than a stale snapshot. We also track basic account-level context, such as open positions and the risk limits a trader has set, because a signal only matters in relation to what is already happening in an account.

We do not treat all data as equally useful. Noisy, low-liquidity conditions are handled differently from deep, active markets, and a strategy’s inputs are scoped to the instruments and directions, long or short, that it was designed for. The aim is not to feed models everything available, but to feed them what is relevant to the decision at hand. This keeps the pipeline manageable and keeps each model’s job narrow enough to reason about, rather than asking one system to understand every market condition at once.

Data quality is treated as an ongoing task rather than a one-time setup. Feeds are monitored for gaps and irregularities, because a model built on a strategy’s tested logic is only as reliable as the numbers it receives. None of this guarantees a particular result; it simply means the inputs are handled with the same discipline as the logic that acts on them.

What the models are actually doing

At a working level, the models used across SynqoraX strategies are pattern-recognition systems. They are trained to notice recurring relationships in market behavior, such as how price tends to move after certain volume or volatility conditions, and to translate those relationships into a structured read of current conditions. We do not publish internal accuracy figures or claim a specific success rate, because any such number, taken out of context, tends to mislead more than it informs. Market conditions change, and a figure from one period says little about the next.

What matters more than a headline number is that each model has a defined scope. A model built for one strategy is not assumed to generalize to another, and strategies are tested across varied historical conditions before they are made available in the library, not to prove a fixed win rate but to check that the logic behaves sensibly across different regimes. This is closer to engineering discipline than to forecasting. The models are not trying to predict a single future price; they are trying to describe the present state of a market in a way a strategy can act on consistently.

This is also why SynqoraX offers a library of tested strategies rather than one universal model. Different strategies suit different conditions, and a trader selecting a strategy is really selecting which patterns they want their capital exposed to.

From signal to order

When a model’s read of the market meets the conditions a strategy is looking for, it produces a signal, an internal instruction that a particular action fits the strategy’s logic at that moment. A signal on its own does not move capital. It is checked against the strategy’s rules, including position sizing and the direction, long or short, the strategy is configured to take, before it is allowed to become an order.

Because execution happens on our servers rather than on the trader’s device, this sequence continues whether or not the trader is actively watching the app. A strategy deployed and running keeps evaluating signals and, where conditions are met, keeps placing orders, which is the same behavior whether the trader’s phone is open, locked, or turned off. This server-side design is also what allows position and performance reporting to stay current in one place, since the record of what a strategy has done lives alongside the logic that did it, rather than depending on a device staying connected.

Not every signal results in an order. Many are generated and discarded because they fail a later check, which is by design rather than a fault in the system.

The risk gate before execution

Every signal that survives strategy logic still passes through a risk gate before an order is placed, not after. This is a deliberate ordering. The gate checks the signal against the limits a trader has configured, position size relative to account balance, exposure per instrument, and any stop conditions attached to the strategy, and it can block or resize an action that would breach those limits.

This matters because a model can be right about a pattern and still produce an outcome a trader would not want if sizing is left unchecked. The risk gate exists to keep execution inside boundaries the trader has explicitly set, rather than leaving that judgment entirely to a signal generated moments earlier. Limits are configurable per strategy, which means a trader running several strategies at once can size each one differently based on how much of their balance they are comfortable exposing to it.

The risk gate does not eliminate risk. It enforces a boundary. Trading digital assets carries substantial risk, including the total loss of capital, and no configuration of limits changes that basic fact. What the gate does is make sure that risk stays within parameters a trader chose deliberately, rather than being decided implicitly by whatever a model happened to signal.

What this approach cannot do

It is worth being direct about the limits of this system. Machine-learning models describe patterns in data that already exists. They do not predict the future, and no combination of data, models, and risk limits changes that. Markets can and do move in ways that have no clean precedent in historical data, and a model trained on past conditions has no special ability to anticipate a genuinely new one.

We do not publish accuracy percentages, win rates, or backtest returns on this site, and we would treat any such figure with the same caution we ask of our traders: past performance and any illustrative figures do not guarantee future results. A strategy that has behaved consistently across tested conditions can still produce a losing stretch, because that is the nature of markets, not a flaw specific to any one model.

Nothing on this website is investment advice. SynqoraX gives traders tools, tested strategy logic, execution infrastructure, and configurable risk limits, but the decision to deploy capital, and how much of it, remains the trader’s own. Understanding what the models are doing, and what they are not, is part of using them responsibly.