What is the difference between a DEA score of 1 and a score of 2?

The approval. At score 1, encoded logic makes the decision and a human confirms it before it takes effect — which means the human is still the decision-maker. At score 2, the logic's decisions take effect within designed Intervention Thresholds, and human judgment is reserved for the decision classes the Exception Architecture explicitly assigns to the Steward. The distinction is not about how good the logic is. It is about whether a person stands between the logic and its consequences.

Does a high DEA score mean no human ever makes decisions?

No. It means the decisions humans make are the ones the architecture assigned to them at design time — the Judgment Layer. The autonomous business removes the pretence that every operational determination requires judgment, not judgment itself. The Judgment Layer is not a residual category; it is a deliberately designed domain where genuine human assessment adds value that encoded logic cannot replicate.

Why does the approval step matter if it only takes seconds?

Because encoded preparation is nearly free at the margin and human approval is not. As volume grows, the Escalation Rate — the proportion of decisions requiring human confirmation — becomes the throughput ceiling of the operation. The Automation Paradox compounds it: the faster the prepared decisions arrive, the more relatively expensive the human confirming them becomes. An approval step that takes ten seconds per case becomes the binding constraint at ten thousand cases a day.

How does Decision Execution Autonomy (DEA) relate to Task Execution Autonomy (TEA)?

[Task Execution Autonomy (TEA)](https://arcoventure.studio/lexicon/task-execution-autonomy) scores who moves the work; DEA scores who decides how it moves. A business can score 2 on TEA — agents execute everything — while scoring 0 on DEA if a human approves every routing decision, prioritisation, and escalation. This is the most common autonomy misread: a business that has automated its execution while keeping its decision-making human looks and feels autonomous because the work is no longer done by people. DEA makes the misread visible as a number.

What is the relationship between DEA and the State Machine and Exception Architecture?

The State Machine is the architectural implementation of high DEA: every routing, prioritisation, and escalation decision is encoded as a state transition that the logic resolves without human initiation. Building the State Machine requires the decision inventory — the explicit enumeration of every decision class the operation contains and the conditions that govern each. The Exception Architecture governs the decision classes that fall below the DEA score: it specifies which operational states the encoded logic cannot resolve and must escalate to the Steward. Taken together, the State Machine encodes what DEA scores; the Exception Architecture bounds what DEA cannot yet score.

The Decision Belongs to the Logic

Beneath every task a business executes sits a decision that someone made: that this task runs next, that this exception escalates, that this case closes, that this resource goes here rather than there. The task is visible; the decision is not. Which is why a business can transfer all of its visible work to agents and remain, in every structural sense, a human operation — because the invisible stream of operational decisions still flows through people.

Decision Execution Autonomy (DEA) is the second axis of the Autonomy Spectrum Framework: the proportion of a business's operational decisions made by encoded logic rather than human judgment, scored 0–2 against who decides how work is routed, prioritised, escalated, and resolved. It is the axis where automation and autonomy diverge.

The decision inventory

Operational decisions are not strategy. They are the continuous, high-frequency stream of small determinations that keep an operation moving: which incoming item is handled first, which case matches which workflow, which anomaly is an exception and which is noise, which exception resolves automatically and which escalates, which capacity is allocated to which queue. A mid-sized operation makes thousands of these decisions a day. In a traditional business, every one of them passes through a person — usually a manager, which is most of what management is.

DEA scores the transfer of that stream. The first step in scoring the axis is building the decision inventory: an explicit list of the decision classes the operation contains. The decision inventory is the human-readable form of the State Machine — every routing condition, every escalation trigger, every exception class, made visible as a list before any of them are encoded. Most businesses have never written this list, because when humans make all the decisions, nobody needs to know what the decisions are. The act of inventorying is itself diagnostic: a business that cannot enumerate its operational decisions cannot have encoded them.

The scoring semantics

A score of 0 indicates that people make the operational decisions, with systems informing them — dashboards, recommendations, alerts that a human acts on. A score of 1 indicates that encoded logic decides within bounded domains, but people retain approval authority over decisions the logic has already made. A score of 2 indicates that operational decisions are made by encoded logic within designed Intervention Threshold, with human judgment reserved for the Judgment Layer / Execution Layer — the decision classes the architecture deliberately assigns to the Steward through the Exception Architecture.

The boundary between 1 and 2 is the approval. This is the subtlest and most consequential line in the framework, because the score-1 business looks autonomous from a distance. Its logic routes, classifies, prioritises, and proposes. Its dashboards are full of machine-generated determinations. And then, at the end of the pipeline, a person reviews and confirms — and in that confirmation, the entire transfer is undone. When logic decides and a human confirms, the decision has not been transferred. Only its preparation has. The human is still the decision-maker; the system has merely become a very sophisticated way of drafting their decisions for them.

The approval bottleneck

The approval pattern is not an accident of caution. It is the structural signature of the Automated Business — a company that uses technology to execute existing human workflows more efficiently while keeping human decision-making at the centre of its operations. The original workflow had a manager approving things; the automated workflow has the same manager approving more things, faster, with better-prepared options. The decision architecture is untouched.

The cost of the untouched architecture compounds with scale. Encoded preparation is nearly free at the margin; human approval is not. As volume grows, the Escalation Rate — the proportion of cases requiring human judgment — becomes the throughput ceiling of the entire operation, and the business discovers that it has spent its AI budget making its bottleneck more comfortable. The Automation Paradox lives here: the faster the prepared decisions arrive, the more relatively expensive the human confirming them becomes.

A score of 2 does not mean no human ever decides anything. It means the decisions humans make are the ones the architecture assigned to them — the Judgment Layer, defined at design time, bounded by Intervention Thresholds the logic enforces. The autonomous business does not remove judgment. It removes the pretence that every operational determination requires it.

The Operator's Verdict

Audit one day of your operation and separate what the systems prepared from what the systems decided. Every recommendation a person acted on, every queue a person ordered, every escalation a person classified is a human decision wearing automated clothing. The businesses that clear this axis did not get there by trusting models more. They got there by designing the decision inventory, encoding each class with its threshold, and accepting that an approval step is not a safety measure — it is a declaration about who runs the operation.

Technology changes what is possible. Decision ownership determines what is autonomous.

KEY TAKEAWAY

What is Decision Execution Autonomy (DEA)?

Decision Execution Autonomy (DEA) is the second axis of the Autonomy Spectrum Framework — the proportion of a business's operational decisions made by encoded logic rather than human judgment, scored 0–2 against who decides how work is routed, prioritised, escalated, and resolved. A score of 0 means people decide with systems informing them; a score of 1 means logic decides within bounded domains but humans retain approval authority; a score of 2 means encoded logic decides within designed Intervention Thresholds, with human judgment reserved for the Judgment Layer — the decision classes the Exception Architecture assigns to the Steward. DEA is where automation and autonomy diverge: when logic decides and a human confirms, only the decision's preparation has been transferred, not the decision. The first step in scoring DEA is building the decision inventory — an explicit list of the decision classes the operation contains — which is the human-readable form of the State Machine.