Core concepts

A handful of terms carry most of the platform. They compose cleanly: sensors feed experiments, experiments produce runs, and process-class templates make runs comparable across the fleet.

Sensor arrays

A sensor array is a set of inputs to a system — built from physical hardware, replay of recorded data, or model-backed virtual sensors. The abstraction is the same across all three, so an experiment does not care whether a channel is a real thermocouple or a simulated one.

Experiments and runs

An experiment is the definition of something to exercise — the sensors, the models in the loop, and the conditions to apply. A run is one execution of an experiment on the edge, in real, simulated, or hybrid mode. Runs are deterministic and replayable: given a seed and a fixed timestep, a simulated run reproduces.

experiment (definition)  ──run──▶  run #1  (edge A, real)
                          ──run──▶  run #2  (edge A, simulated)
                          ──run──▶  run #3  (edge B, hybrid)

Process-class templates

A process-class template describes a kind of process, independent of any single edge. When runs across the fleet instantiate the same template, they become comparable — that shared identity is the join key the cloud aggregates on.

This is what makes fleet-level learning possible: without a common template identity, runs on different edges are just unrelated time series. With it, they are a comparable set the cloud can aggregate and, in the model-loop phase, learn from.

How they fit together

Sensors define what you measure; experiments define what you do with them; runs are the record of doing it; and templates are what let many runs — across many edges — line up into something a fleet can improve against.