Twin Ops connects the two engines. A compiled, certified model runs frozen next to your live adaptive monitor — and the distance between them is the product. Part of the Analytics layer: artifact hosting plus your channels as traffic — no flat fee.
Pricing: inside Analytics — hosting plus channel traffic; recompiles spend plan credits. Full pricing →
The idea
A compiled artifact from the Dynamics Compiler is deployed into a live Analytics session as a frozen reference predictor, running in parallel with the live adaptive filter. The live filter adapts; the twin does not. Their divergence is the signal.
Every tick produces a twin prediction, a twin residual and a support status — the certified model measured against reality, live.
TwinDivergence when the system has steadily left the certified model while the live filter is calm; SupportExit when the system leaves the model's certified support.
The twin is authoritative inside its support; the live model outside it. Divergence is an event, not a debate — and the twin is never silently tuned, only updated by an explicit, reported recompile.
Graph-and-context platforms — Azure Digital Twins, AWS IoT TwinMaker — model what is connected to what; the predictive model itself is something you bring. Twin Ops supplies exactly that piece: a compiled, certified predictor whose divergence from the live system is a first-class signal. Infrastructure we sit beside, not a rival.
The loop
A live session watches the system, segments it by regime, and keeps its data quality high.
The accumulated, QA'd, segmented data becomes a compile request — controls from the connection graph, lags from live structure.
The certified artifact returns to the session as a frozen twin under continuous supervision.
Smooth coefficient drift triggers a fixed-structure update; structural divergence triggers a recompile on broader data.
End-to-end
Cycling runs under Analytics (cell health, sensor freeze, charge/discharge regimes as event labels). After a month, compile-from-session yields a degradation/dynamics artifact with a Passport. A SupportExit on new chemistry or temperature is the automatic signal: broaden the data and recompile.
Historical logs give a first compile. The twin is deployed under the monitor: TwinDivergence catches a heat-exchanger degrading weeks before a threshold alarm, while periodic updates keep coefficients fresh without re-opening the structure.
A team arrives for anomaly detection only and lives on Analytics. A month in, the console shows regime changes, stable connections and nonlinearity verdicts — and offers to compile a passported twin for that system, on data that's already clean.
Status
Twin Ops is where both engines meet, through a shared artifact contract rather than a shared core. The data bridges — compile-from-session, twin-residual monitoring and trigger-based updates — are in active development, validated on design-partner data before general availability.