The Arcflux thesis
Every environment is fragmented. People hold it together.
Operational truth is scattered across protocols, controllers, vendor platforms, configuration files, drawings, and the heads of the people who run the place. Each source holds part of the picture. None holds all of it.
Engineers and operators bridge the gaps by hand — reading points, naming things, documenting relationships, keeping the model current. Humans are still the integration layer.
And the model decays. Equipment is replaced, sensors move, configurations drift, and the documented picture quietly stops matching the plant. AI inherits exactly this fragmentation. Without a current, trustworthy model of the environment, it cannot safely move from observation to reasoning — let alone action.
A discovery engine that turns fragments into a working model — and shows its work.
The engine ingests what the environment already emits. It forms hypotheses about what exists and how it connects, weighs the evidence for and against, asks a person when it isn't sure, and notices when reality stops matching what it believes.
The result is the Operational Genome™ — a living model of what exists, how it relates, what it can do, and what is happening. An ontology no one has to write.
Every belief carries its receipts.
Arcflux does not assume its inferences are true. Every conclusion in the model carries where it came from — observed directly, imported from an existing source, inferred from evidence, or confirmed by a person — along with its confidence, and anything that contradicts it.
Confirmation is not overhead to be minimised. A person agreeing with the model is a different kind of knowledge than a model being sure of itself, and the system keeps the two distinct — permanently.
Understanding comes before operation.
Arcflux does not automate anyone's building. It enters an unfamiliar environment and constructs a useful operational model with minimal upfront configuration — then keeps it true. It discovers what exists. Resolves identities across systems. Infers relationships. Exposes uncertainty instead of hiding it. Asks for confirmation where it matters. Detects when the model no longer matches reality.
The method and the numbers get published, not asserted.
This idea exists now.
If you're building AI for physical environments, we want to compare notes.
If you own or operate a building or a plant and would host an install, we especially want to hear from you.