What is Fleet Learning?
Fleet Learning is the compounding improvement in thermal control performance that arises when a distributed fleet of instrumented pods, operating across different climates, tariffs, grids and workloads, trains a shared control model. Diversity of operating condition, not volume of records, is what makes a control model generalise, which is why many small sites can be worth more than one large hall.
Definition
Fleet Learning is the compounding improvement in thermal control performance that arises when a distributed fleet of instrumented pods, operating across different climates, tariffs, grids and workloads, trains a shared control model.
Why it matters
A single large data hall generates a great deal of data about one thermal environment. A distributed fleet of small pods generates less data about many thermal environments, and for training a control policy the second is worth more. Diversity of operating condition, not volume of records, is what makes a control model generalise.
Why it cannot be bought
The dataset cannot be purchased, because no one is selling it. It cannot be retrofitted into an already-deployed fleet, because the instrumentation has to be present from commissioning. And it improves with every unit shipped, which means the gap widens rather than closes.
Coral's position
Fleet Learning is the only asset in Coral's business that compounds. Every other component of a Coral pod is purchasable by a competitor with sufficient capital.
Take the vocabulary into a real site conversation.
An architecture review applies these terms to a specific building: its power envelope, its thermal capacity, its density target and its jurisdiction.