DemoPricing

What is the Autonomy Readiness Index (ARI)?

The Autonomy Readiness Index (ARI) is a score from 0 to 1 for how far a rule set can safely be left to decide without a person. It combines four component scores: Logic Integrity (LIS), Data Reliability (DRS), Governance Maturity (GMS) and Risk Exposure (RES).

How it is calculated

ARI = 0.35 × LIS + 0.30 × DRS + 0.20 × GMS − 0.15 × RES, clamped between 0 and 1. LIS falls with exceptions, conflicts, deep nesting and undocumented rules. DRS reflects missing and inconsistent data and, with decision logs, how often practice departs from the rules. GMS counts owners, an audit trail, escalation paths and versioning. RES rises with financial, irreversibility and regulatory exposure.

What the bands mean

At 0.60 or above a rule set is in the autonomous band; from 0.30 to 0.60, bounded, where a person reviews more cases; below 0.30, human-governed. Because Risk Exposure is subtracted, the highest score in practice is about 0.85.

With and without decision logs

A policy sealed as written (Declared) shows the logic, governance and risk components. Data Reliability and the composite need decision logs, so a Declared artifact does not show a composite score.

Example

Scenario

A lender wants to let a model approve small personal loans without review.

Without governance

The model is 92% accurate on historical data, so it is switched on. Accuracy says nothing about whether the rules around it conflict, whether the input data is reliable, or whether anyone owns the decision.

With NOMOS

The rule set scores 0.41: logic is clean, but data reliability is low and there is no escalation path for edge cases. It runs in the bounded band, with more cases sent to a person, until those are fixed.

Frequently asked questions

Is ARI an accuracy score?+

No. It says nothing about how accurate a model is. It scores the rules, the data they read, who owns them and what is at stake.

Can ARI change after sealing?+

Yes, when new decision logs or outcomes are added and the analysis is run again. The sealed rules themselves do not change; a new version would.

Related concepts

See it in practice

Upload a policy, confirm its rules, seal them: any system can then check an action against them.

Open Studio →