The vocabulary of autonomous operations.

Authority ◆

The right to cause an action that affects the process. The scale's central variable: human at Levels 0–3, shared within an envelope at Level 4, systemic under human objectives at Level 5.

Autonomy level ◆

A classification (0–5) of an operation's systems by what they do unprompted and who holds authority over actions affecting the process. → The six levels

Bounded closed loop

Level 4 of the scale: the system writes back to the control system within a defined operating envelope — named conditions, confidence thresholds, hard limits — and degrades to a safe state outside it. Humans set objectives and handle exceptions. → Level 4

Confidence score ◆

A quantified statement of the system's certainty attached to a recommendation, enabling proportionate human review.

Degrade to safe state ◆

The mandatory Level 4 behavior of relinquishing closed-loop action and settling the process to a defined safe condition when the operating envelope is exited.

Domain model ◆ (the Domain)

A single connected representation of an operation's physical and digital elements — assets, hierarchies, sensors, analytical models, computations and documents — and their relationships. The prerequisite of Level 2. → Deep dive: the Domain

Goal-directed operation ◆ (goal-directed plant)

Level 5 operation in which human input consists of business objectives and their governance, and the system plans and acts across all units and conditions to meet them. → Level 5

Instrumented

Level 0 of the scale: the plant is measured but not understood by its systems — data acquisition, visualization and fixed-rule automation exist; all interpretation is human. → Level 0

Integrated assistant

Level 2 of the scale: a unified domain model with orchestration to run analysis across it on demand — plant-wide, cross-domain reasoning in human language. The system remains reactive: humans must know what to ask. → Level 2

Operating envelope ◆

The explicitly declared set of conditions, thresholds and hard limits within which a Level 4 system is permitted to act on the control layer.

Point analytics ◆

An analytical or machine-learning model attached to a single asset or narrow function, with its own data path and asset definition. Characteristic of Level 1. → Level 1

Recommendation ◆

A proposed action emitted by the system, carrying its diagnosis, the alternatives considered and a confidence score, for human approval. The unit of Level 3 operation.

Supervised autonomy

Level 3 of the scale: the plant monitors itself continuously and unprompted — detecting deviations, diagnosing causes, simulating options and emitting recommended actions with confidence scores. Humans approve and execute every action. → Level 3

The Six Levels of Autonomous Plant Operations

The open scale (Arundo 2026) classifying plant operations from Level 0 (instrumented) to Level 5 (goal-directed), anchored in authority and unprompted capability, modeled on the structural precedent of SAE J3016. → The scale · Download the standard

Unprompted operation ◆

System activity — monitoring, diagnosis, recommendation — initiated by the system on its own schedule or triggers, rather than in response to a human query. The Level 2→3 boundary.

Write-back

A system action applied directly to the control layer (e.g., a setpoint change), as distinct from a recommendation to a human. Permitted only at Level 4 and above, and only within an operating envelope.

These definitions may be reproduced unmodified with attribution. Cite as: Arundo Analytics AS (2026), The Six Levels of Autonomous Plant Operations, v1.0, §4 · https://doi.org/10.5281/zenodo.23476035