What counts as an intervention when scoring this axis?

An intervention counts when the system could not continue correctly without it — an exception above the Intervention Threshold or a halt the architecture itself triggered. A Steward who voluntarily inspects, tunes, or improves a running system has not intervened in the scored sense. The distinction is between required human input and elected human input. Curiosity is not dependency; it is governance, and governance is what makes the score valid.

Why does Intervention Dependency use both MTTI and the Escalation Rate?

Because they catch different failure shapes. MTTI measures the rhythm of dependency over time — how long the system runs before it surfaces something. The Escalation Rate measures the proportion of work that needs a person, regardless of when. A system with a long average MTTI and a high Escalation Rate has a burst problem: long quiet stretches punctuated by concentrated human load. A system with a short MTTI and a low Escalation Rate has a chronic miscalibration in one task class. A single number makes both look the same; the pair distinguishes them.

How does Intervention Dependency relate to the first three axes?

The four axes score different layers of the same operation and none can be inferred from the others. Task Execution Autonomy (TEA) scores who executes the work. Decision Execution Autonomy (DEA) scores who makes routing and resolution decisions. Process Continuity Score (PCS) scores whether work moves between stages without a human restart. ID scores the operational result of all three: how long the system built from those architecture decisions actually ran before it needed a person. A business can achieve high TEA, DEA, and PCS while scoring 0 on ID if the architecture's exception protocol is miscalibrated and produces constant escalations. The axis is the reality check.

What is Proof of Action and how does it validate the ID score?

Proof of Action is the immutable record of every agentic decision, output, and intervention that the system produces during operation — the structured log the Steward reviews on the Audit Surface. It validates the ID score in two ways: it confirms that interventions counted were genuinely required rather than elected, and it confirms that the Steward was actively reviewing the governance surface during the measurement window rather than simply absent. An ID score without a Proof of Action trail is an assertion. With it, it is an auditable claim — the difference that matters to an operator managing the architecture, an acquirer evaluating it, and a Steward defending it.

The System That Does Not Call

Q: Can a system score 2 just by running quietly for weeks?

No. The score requires evidence that the Steward was actively engaging with the Audit Surface during the measurement window. A long quiet stretch on an unmonitored system is Nominal MTTI — a measurement of absence, not autonomy — and scores zero regardless of its length. The validation pairing — intervention frequency plus confirmed Steward engagement — is what distinguishes Intervention Dependency from a passive count of how often someone was called.

Intervention Dependency (ID) is the fourth axis of the Autonomy Spectrum Framework: the frequency with which a business's core systems require human intervention to continue operating, scored 0–2 and operationally expressed through MTTI and the Escalation Rate — and the only axis that scores the operational result rather than the architectural design. The first three axes — Task Execution Autonomy (TEA), Decision Execution Autonomy (DEA), and Process Continuity Score (PCS) — can all be assessed from design documents and workflow maps, which means all three can be engineered to look right on paper. ID cannot. Either the system ran for three days without needing a person, or it did not, and the logs know which. It is the inverse measure of Architectural Certainty — the state in which a business’s core logic is so robust that it requires no human decision-making for days or weeks at a time.

The two operational expressions

Intervention Dependency is read from two instruments. The first is MTTI (Mean Time to Intervention) — the average time between required human interventions in an agentic system, with Arco's target greater than 72 hours on core Revenue Loops. MTTI measures the rhythm of dependency: how long the system runs before the architecture itself summons a person.

The second is the Escalation Rate — the proportion of agentic task executions that escalate to a human Steward. The two instruments catch different failure shapes. A system can have a long MTTI and a high Escalation Rate concentrated in bursts, or a short MTTI driven by one chronically miscalibrated task class while everything else runs clean. The axis is scored on both, because a single number hides exactly the pathology an acquirer or operator needs to see.

The word required carries weight in both definitions. An intervention counts when the system could not continue correctly without it — an exception above the Intervention Threshold, a halt the architecture itself triggered. A Steward who chooses to inspect, tune, or improve a running system has not intervened in the scored sense. Curiosity is not dependency.

The scoring semantics

A score of 0 indicates continuous steering: the system requires human input multiple times per day, and the people operating it are functionally part of the execution path regardless of how the architecture is described. A score of 1 indicates scheduled or frequent intervention — the system runs unattended for hours but not days, and intervention is routine rather than exceptional. A score of 2 indicates that MTTI on the core Revenue Loop exceeds 72 hours and intervention occurs by exception only, at the conditions the Intervention Threshold was designed to surface.

The 72-hour line is not arbitrary. A system that crosses it has survived the full variety of a multi-day operational cycle — the volume peaks, the malformed inputs, the integration hiccups — without once requiring a human decision to continue. Below that line, a business does not have an autonomous operation with occasional support. It has a human operation with an unusually capable toolset, staffed by people whose job is to be interruptible.

The number that must be watched to be true

Intervention Dependency carries one validation requirement that no other axis does, and it is the reason the axis cannot be gamed by neglect. A long MTTI has two possible causes. The first is Architectural Certainty: the system genuinely did not need anyone. The second is Nominal MTTI — the condition in which measured MTTI is long not because the system achieved Architectural Certainty but because the Steward has stopped engaging with the Audit Surface. Both produce identical numbers and opposite realities.

This is why ID is always scored as a pair: the intervention frequency, and the evidence that the Steward was actively engaging with the Audit Surface throughout the measurement window. A 200-hour MTTI on a monitored system is a strong score. The same number on an unmonitored system is Nominal MTTI — a measurement of absence, not autonomy — and it scores zero. The full anatomy of that failure mode, and the specific risk it creates as the architecture matures, is documented in The Problem That Success Creates.

The Operator's Verdict

Pull ninety days of Proof of Action records and compute two numbers: the average gap between interventions your system actually required, and the proportion of executions that escalated. Then ask the harder question — for every long quiet stretch, can you prove someone was watching? An operation that cannot produce all three answers is not autonomous and not measurable; it is simply undocumented, which is a different problem and a prior one.

Technology changes what is possible. The audit determines what is certain.

KEY TAKEAWAY

What is Intervention Dependency (ID)?

Intervention Dependency (ID) is the fourth axis of the Autonomy Spectrum Framework — the frequency with which a business's core systems require human intervention to continue operating, scored 0–2 and expressed operationally through MTTI (Mean Time to Intervention) and the Escalation Rate. A score of 0 means continuous steering with intervention multiple times per day; a score of 1 means the system runs unattended for hours but not days; a score of 2 means MTTI exceeds 72 hours on core Revenue Loops and intervention occurs by exception only, at the conditions the Intervention Threshold was designed to surface. The axis is valid only when scored against an active Audit Surface, because a long MTTI produced by a Steward who has stopped watching is Nominal MTTI — a measurement of absence, not autonomy — and scores zero regardless of the number.