Units: ensemble models
The smallest reasoning components: models that disagree on purpose so the blend can see where confidence is real.
P1-Chaos is the predictive-ML system on the Intelligence OS stack. Ensemble models extract signal; blending produces a calibrated probabilistic output. It is operable as a standalone reasoning system, and composable as a calibration Unit within Tempo. The result is a distribution that is always the current one, corrected against the observations present at the moment it is issued.
A forecast traditionally hands you one number and treats it as equally good at every horizon. Real outcomes fall on either side of that number, and downstream systems that planned against the average begin to miss. Chaos plans from the ensemble rather than around it.
Like Terra on the farm and Via on the route, Chaos treats each issuance as the input to its next cycle. The output underneath a Chaos decision is never the raw ensemble; it is the ensemble as corrected against the most recent observations.
The smallest reasoning components: models that disagree on purpose so the blend can see where confidence is real.
Declared relations between models. Signal is extracted, not averaged away. The blend is a Function, not a spreadsheet.
A standalone reasoning system, composable as a calibration Unit within Tempo. Operable at either level of granularity.
Chaos is a domain-specific instantiation of the Intelligence OS architecture applied to predictive ML. It is a Function within Tempo and a governed input to every application that needs a calibrated distribution rather than a point estimate. Ensemble precision determines the quality of the field Tempo and Maritimo optimise over, which determines the decisions those systems issue. Outputs are traceable to the ensemble, the blend, and the cycle that produced them.
Tell us which signals you already collect. We will come back with the ensemble the stack can consume.