Five properties of the data, and how it behaves over time.
Measured values with timestamps.
Readings, prices, loads or flows. Data that is mostly identifiers, such as order, part or supplier numbers, is not what the AI learns from.
Hourly cadence or finer.
Each observation is a chance to adapt, so the cadence sets how quickly an instance keeps up.
A few hundred data points of history.
That is what a cold start needs.
Conditions that move.
Values that drift over time, for whatever reason: a machine wearing, a patient changing, the seasons turning or a grid taking on new load. That movement is what the CLS keeps up with.
No new sensors.
The CLS works on data that is already being collected.