The twin is the reference
Alerts are measured against expected physics, not fixed thresholds, so fewer of them are noise.
Elidyne is the control pane where every loop is watched, tested, approved and traced. Operations is its live desk for systems in service: signals, the twin, open incidents and the control plan share one screen, and every change to the real system goes through a standing grant with bounds, a fallback and a named approver.
Station operations for the same tsunami buoy, in service: live signals against the anchor twin, an open incident, and every controllable point with its grant.
Each stream is compared with what the twin expects, so a reading shows as inside or outside its band, not just as a number.
An event opens an incident that gathers the signals, the twin re-run and every decision taken, in order.
Every setpoint or mode Elidyne may touch is listed with its bounds, its fallback and who approved it.
Each monitor states what it may read, what it may raise and whether it may act, at a stated autonomy level.
Watching is cheap. Acting on a live system is not. Operations keeps the two clearly apart.
Instruments, historians and telemetry arrive through read-only connectors. Freshness is shown for every stream, and the twin says what each reading should be.
An alert arrives with the reason it fired, the twin re-run behind it and what Elidyne proposes next. The incident thread keeps all of it for the event report.
Each controllable point has its own grant: the bounds, the rate, the watchdog, the fallback and the approver. The controller on the asset keeps its own limits too.
| Point | Level | Bounds |
|---|---|---|
| Sampling mode | L2 | Event mode on trigger |
| Telemetry cadence | L1 | 1–15 min · with approval |
| Clock re-sync | L1 | Once per 24 h |
| Hold the line now | Operator | Never automated · gated |
A pilot battery coating line runs on the same Operations desk as a deep-water buoy. The twin here is an oven thermal model with an adhesion model from an earlier investigation, and control stays advisory until the line owner signs the grant.
A standing grant is how your engineers say yes in advance, precisely. Anything outside it waits for a person.
Alerts are measured against expected physics, not fixed thresholds, so fewer of them are noise.
Every write goes through a single governed gateway that the operations desk can pause at any time.
Incidents, decisions and control changes stay in one log that the next shift and the next report can read.
Watching is cheap. Acting on a live system is not. So every action carries its bounds, its fallback and its approver.
Six products, one control pane. Operations mainly carries Measure and Act; the rest of the loop is already connected when you need it.
An incident that needs real analysis opens an investigation in Workbench, with the signals attached.
Open Workbench →Watch and steer live loops, inside limits
You are hereEvent and shift reports are drafted from the operations log and incident threads.
Open Reports →04 · Under the hoodElidyne explains each alert against the twin and drafts the control plan.
Open Agents →05 · Under the hoodTwin re-runs and forecasts are placed on compute close to the data.
Open Compute →06 · Under the hoodEvery write to a live system passes through a standing grant with bounds and a fallback.
Open Governance →Measure → Twin → Test → Limits → Act → Trace. Every product serves the same six steps, in every loop.
Tell us about the system you run. We will map the signals, the twin and the first control points worth putting inside a governed loop.