How can you tell genuine Architectural Certainty from Nominal MTTI?

By the audit record. Genuine Architectural Certainty produces a long MTTI alongside continuous Steward engagement with the Audit Surface — engagement that leaves a Proof of Action record. Nominal MTTI produces the same MTTI number with no engagement record. The framework treats the pair as the measurement: either number alone is uninterpretable. A 300-hour MTTI with a continuous Proof of Action trail is a strong score. The identical number without the trail is a finding — it measures absence.

Doesn't requiring audit evidence defeat the point of autonomy?

No. Autonomy means the system does not require human execution or routine intervention — it does not mean the system runs unobserved. The Steward's audit engagement is governance, not intervention, and it is what makes every other autonomy metric the operation reports interpretable. The distinction matters practically: a Steward reviewing the Audit Surface for ten minutes at a pace designed for operational tempo has not intervened in any task. They have confirmed that the system's silence is earned.

Why publish how the framework can be gamed?

Because a framework that conceals its failure modes asks to be trusted, while one that publishes them asks to be checked. The Autonomy Spectrum will face sceptical scrutiny when applied to real companies, and the methodology survives that scrutiny only by answering the gaming question before it is asked. An autonomy score produced by a framework that could be defeated by neglect would be cheaper than the marketing claim it was supposed to replace.

What happens to a system operating under Nominal MTTI over time?

The accumulation is compound and silent. Knowledge Debt grows: without active Steward review, resolved exceptions are not encoded into the Operational Ledger and the system's context quality stagnates. Execution Divergence patterns accumulate undetected: the leading indicators of context drift that the Audit Surface was designed to surface never reach the Steward. Logic Decay advances unchecked: the Continuous Regression Loop may fire, but without Steward review the Deterministic Failures it produces are unresolved. The Exception Architecture stops improving. When the system does break visibly — and unmonitored systems break loudly — the business must reconstruct weeks of operational history from logs that recorded actions but not the decisions about those actions.

How does the metric pairing principle extend to other autonomy measurements?

Resolution Integrity is paired with the ticket-closure rate, because closure without resolution is the same gaming move — a metric that improves when the system stops caring about output quality. The Escalation Rate is paired with active Audit Surface review, because a low rate in the absence of Steward oversight confirms only that nobody was available to escalate to, not that the system resolved cases correctly. The 80 Percent Threshold is paired with the Proof of Action trail, because cross-departmental handoffs that executed without intervention are only meaningful if the system was in a position to surface exceptions. The pairing principle applies wherever a metric can improve through disengagement rather than through architectural quality.

The Metric That Lies

Here is how to achieve a world-class autonomy score without building anything: deploy an agentic system, stop watching it, and wait. The interventions will fall to zero — not because the system stopped needing them, but because nobody is there to perform them. The MTTI (Mean Time to Intervention) will stretch to weeks. On paper, the operation has achieved what we spend entire architectures pursuing. In reality, it has achieved a measurement of absence. Nominal MTTI is the condition in which measured MTTI is long not because the system achieved genuine Architectural Certainty but because the Steward has stopped engaging with the audit surface — a metric that scores zero precisely because what it measures is not autonomy. The Autonomy Spectrum Framework names this failure mode specifically because it needs to be nameable.

Any framework that scores autonomy must answer this attack before it scores anyone, because the attack is not hypothetical — it is the default trajectory of every unmonitored deployment. This memo is the Autonomy Spectrum Framework answering it.

The two causes of silence

A system that does not call for help is silent for one of two reasons. The first is Architectural Certainty: the logic genuinely handled everything the operational cycle produced, and the silence is the sound of a system that works. The second is neglect: the system encountered conditions that should have triggered escalation, and the escalations landed nowhere, and the failures are compounding quietly in the gap where a Steward used to be.

The two causes produce identical metrics and opposite realities. We named the second one in the Lexicon precisely because it needs to be nameable: Nominal MTTI — the condition in which a system's measured MTTI is long not because it has achieved genuine Architectural Certainty but because the Steward has stopped engaging with the audit surface, producing a metric that appears to confirm autonomous operation while the system is in fact running unmonitored and unrefined.

The condition is dangerous in proportion to how good it looks. A short MTTI announces its own problem; a nominal one announces a triumph. The dashboard turns green at exactly the moment governance ends, and every reporting layer above the dashboard inherits the green. By the time the silence breaks — because Logic Decay compounds in the dark and Execution Divergence patterns accumulate undetected — the business has weeks of unaudited operation to reconstruct, and the Knowledge Debt it accumulated without active Steward encoding is not recoverable from the logs alone.

The validation rule

The framework's defence is structural, not procedural. Intervention Dependency, the axis MTTI feeds, is never scored as a single number. It is scored as a pair: the intervention frequency, and the evidence that the Steward was actively engaging with the Audit Surface throughout the measurement window. The Audit Surface exists to make that engagement observable — it is the governance digest designed for Steward comprehensibility at operational tempo, and engagement with it leaves a Proof of Action record exactly the way the agents' own executions do.

The rule, stated plainly: an autonomy metric is valid only over a window in which someone was demonstrably in a position to intervene. A 300-hour MTTI with continuous audit engagement is among the strongest results an operation can produce. The identical number without audit engagement scores zero — not as a penalty, but as a finding. What was measured in the second case was the Steward's attention, and the measurement is accurate.

The rule generalises beyond MTTI. A ticket-closure rate is valid only alongside Resolution Integrity, because closure without resolution is the same gaming move in a different department — we made that argument in Closed Is Not the Same as Resolved. Every operational metric that improves when humans disengage needs a paired observation that detects the disengagement. The pairing is not bureaucracy. It is what makes the number mean what it claims to mean.

Why we publish the attack

A measurement framework that conceals its failure modes is asking to be trusted. A framework that publishes them is asking to be checked — and only the second kind earns citation from people who do not already agree with it. The Autonomy Spectrum will be applied to real companies, by us and eventually by others, and every application will face the same sceptical question: how do you know the quiet systems were autonomous rather than abandoned? The honest answer is that without audit evidence, we do not — and the methodology says so, scores accordingly, and survives the question.

The deeper point is about what measurement is for. The purpose of scoring autonomy is not to award the claim. It is to make the claim expensive — and a framework that could be defeated by neglect would make the claim cheaper than ever, because neglect is free.

The Operator's Verdict

Look at your own quietest metric — the one that has been green the longest — and ask when a human last verified that the green was earned. If the answer is not on record, the metric is not evidence of anything except how long it has been since someone checked. The systems most worth trusting are the ones whose silence is audited, and the operators most worth trusting are the ones who can prove they were listening to it.

Technology changes what is possible. Verification determines what is honest.

KEY TAKEAWAY

What is Nominal MTTI and why does it matter for measuring autonomy?

Nominal MTTI is the condition in which a system's measured Mean Time to Intervention is long not because it has achieved genuine Architectural Certainty but because the Steward has stopped engaging with the audit surface — producing a metric that appears to confirm autonomous operation while the system is running unmonitored. It matters because any autonomy measurement can be gamed by neglect: interventions fall to zero when nobody is there to perform them. Under Nominal MTTI, Logic Decay compounds in the dark, Execution Divergence patterns accumulate undetected, Knowledge Debt grows without active Steward encoding, and the Exception Architecture stops improving — while every operational metric appears healthy. The Autonomy Spectrum Framework defends against this by scoring Intervention Dependency as a pair: the intervention frequency plus evidence of active Audit Surface engagement during the measurement window, delivered through the Proof of Action record the engagement produces. A long MTTI without audit evidence scores zero because what it measures is absence, not autonomy.