How explainable trading decisions work

A trading signal should be inspectable, not treated as a mysterious answer. CryptoMetric AI starts with recent market candles and applies the template selected by the user to one symbol and timeframe. The templates contain fixed indicator rules; the user chooses which template, symbol, timeframe and risk settings to run. They are not tailored recommendations for the user's finances, objectives or ability to bear loss.

From market data to a signal

On each cycle the software fetches recent open, high, low, close and volume data. The selected rule-based strategy returns BUY, SELL or HOLD with an internal confidence score. A machine-learning classifier evaluates the same historical observations as a second input, while a deterministic detector describes the current market as trending, ranging or volatile.

The software combines the strategy and model inputs, adjusts them for regime fit and directional agreement, and can reduce a directional result to HOLD when the score is too low. It stores the contributing directions and adjustment factors with the final signal. This record explains what the code considered; it does not establish that the inputs were correct.

Confidence is not a probability of profit

The displayed confidence is an internal score produced by the decision pipeline. It is not a forecast of return, a probability that a trade will win, or evidence of a demonstrated edge. Historical relationships can break, market data can be incomplete or delayed, and a model can be wrong even when its score is high.

The output is therefore information for the user to assess. A BUY or SELL label does not create an obligation to act. Users remain responsible for deciding whether the template, configuration, market and level of risk are appropriate for them, and for stopping automation when they are not.

Risk checks are a separate stage

A directional signal does not automatically mean an order is suitable. Before an entry, the risk manager independently checks matters such as the configured confidence threshold, stop distance, open-position limits, daily realised loss, consecutive losses, volatility and exposure. A failed check rejects the entry and records a reason. Passing the checks only means the request met those programmed limits at that moment; it is not an endorsement or prediction.

Keep the human in control

• Start in paper mode and inspect several complete decision records.
• Understand the selected template instead of choosing it from a name or score alone.
• Check the exchange account independently and keep a manual way to manage positions.
• Treat every signal as fallible information, including when the strategy and model agree.

Transparency helps a user question a decision. It cannot turn uncertain market data into a guaranteed outcome, and it does not shift the user's decision or account control to the platform.