Prediction Events

A Prediction Event represents a set of predictions that were generated at a particular point in time, referred to as the event_time. This is the time at which our models started to generate a prediction, which is an indication of what data may have been available for the model as feature input.

An example of a single event is shown below:

{
    "event_time": "2025-01-01T00:10:00+00:00",
    "predictions": [
        {
            "value": {
                "20": 0.39441445179536083,
                "50": 0.05682774614447045,
                "-20": 0.06880328920223622,
                "-50": 0.017616259919968107,
                "100": 0.021361007443595967,
                "-100": 0.006250882128355161
            },
            "prediction_for": "2025-01-01T02:00:00+01:00"
        },
        ...
    ],
    "created_at": "2025-03-04T07:32:01.117632+00:00",
    "is_simulated": true
}

The created_at time represents when the event was ingested into our API. You'll notice on the example above there is a large offset between the event_time and the created_at timestamps.

This is because the event shown above is a simulated event, as indicated by the is_simulated field. For non simulated (real) predictions, the created_at time will be shortly after the event_time.

Tip

To understand more about simulated predictions, and why they exist, see Prediction Versioning.

Each event contains a set of predictions (usually up to 72hrs ahead of the event_time). Each prediction is made for a specific point in time, prediction_for, and is valid for the resolution of the Prediction Series.

Important

We don't mix resolutions in a series. If the market moves from 1hr to 15min, we create a new series with the appropriate resolution.

The value of a prediction depends on the properties of the series, you can read more about the properties of a series on the Prediction Series page.