Skip to content
An engineer reading a reasoning trace

Glossary

Every term we use, in one sentence a non‑engineer understands.

The same list our writing rules and our claim checker use. If a word on this site isn’t here, tell us and we’ll add it or stop using it.

31 terms

A to Z

Air‑gapped
Running with no network connection to the outside world at all; updates arrive as signed bundles your own team applies.
Anomaly
A pattern in the sensor data that does not match how the machine normally behaves.
Calibration
Whether the confidence a prediction states matches reality — a 90 % range should contain the true answer about 90 % of the time.
Causal chain
The sequence of causes behind a symptom (e.g. feed impurity → catalyst fouling → hot spot), not just the symptom itself.
Confidence interval
The range within which the true result most likely lies — a measure of how sure the statistics are.
Continual learning
Learning new patterns without retraining from scratch and without forgetting old ones.
Counterfactual
What would have happened had we acted differently — e.g. “had we re‑aligned the coupling in March” — computed, not guessed.
Data residency attestation
A signed record that the platform kept all data inside your boundary — produced automatically, and something a security reviewer can ask to see.
Differential privacy
A mathematical guarantee that a shared model update cannot be traced back to any one site’s records.
Digital twin
A virtual copy of a machine that behaves like the real one, used to test ideas safely.
Federated learning
Many sites improve one shared model by exchanging model updates, never raw data — each site’s data stays where it is.
Historian / SCADA / CMMS
The systems a plant already runs — the sensor archive, the control system, and the maintenance work‑order system — which Ryedore reads without changing.
Human-gated
The AI may propose an action, but a person must approve it before anything happens.
Industrial multi-task encoder
Ryedore’s own model: one shared representation learned across industries that serves several tasks (remaining life, anomaly, forecast, classification) — built by us, with no third‑party model inside.
Label flywheel
The loop where each failure your engineers confirm makes the next prediction sharper — the labels are yours and never leave your site.
Neural operator
A model that learns the physics of a class of equipment so it can answer what‑if questions with physics‑level fidelity in milliseconds.
Never-degrade gate
A rule that blocks any new model version that would be worse on any task than the one currently running.
Number status
The label under every figure on this site — signed (re‑runnable proof), attested (measured on the platform), industry figure (a named external source), target (a goal, not a result), illustrative (a worked scenario) or roadmap (not shipped).
On-premises
Runs on your own hardware; nothing leaves your site.
Physics check
A test that the AI’s answer obeys the physical limits of the equipment.
Promotion (of a model)
The moment a new model version is allowed to serve — only after it has passed every quality gate and been logged as admitted.
Promotion attestation
A signed certificate that a new model version passed every gate before it was allowed to serve.
Provenance
Where a result came from — which data, which step, which specialist produced it.
Quick What-if
Ryekronix’s fast setpoint ranking — proposes moves in seconds; it never dispatches them.
Reasoning trace
The step-by-step record of how the AI reached a conclusion — what it saw, what it inferred, and why.
Remaining useful life (RUL)
How long a machine has before it fails, as a prediction with an uncertainty range.
Root cause
The underlying reason a problem happened, as opposed to the alarm that reported it.
Ryekronix
Ryedore’s operations agent: ask in plain English, it plans, investigates, answers with a trace, and asks a human before acting.
Signed benchmark
A published test result on public data, cryptographically signed so anyone can verify it has not been altered — and re-run it.
Specialists Lab
Ryedore’s six-agent experiment team: it designs and runs a controlled trial on a digital twin and reports the statistics. It never touches the plant.
What-if / counterfactual
A question of the form “what happens if we change X?” answered by simulation rather than by trying it on the real machine.

Words are cheap. Check the proof.

Signed benchmarks, a reasoning trace and the gate log are one click away.