Predictive validity of AgentStripes stripes

Determine whether higher AgentStripes stripe assignments for released AI agents predict improved runtime behaviour, specifically whether agents with more stripes fail less in production.

Background

AgentStripes assigns released AI agents between zero and seven stripes by combining capability ratings, evidence strength, severity-weighted scoring, and basic, safety, and proof blockers. The model is intended to measure engineering maturity rather than task performance or operational reliability directly.

The paper explicitly distinguishes the proposed maturity score from runtime outcomes and states that no evidence yet establishes a relationship between stripe assignments and production failures. Establishing this predictive relationship is therefore a central unresolved empirical problem requiring runtime or production outcome data.

References

Whether this combination predicts anything about runtime behaviour is an empirical question that the specification itself lists as open (``nothing yet shows that agents with more stripes fail less in production'').

— AgentStripes: An Evidence-Graded Maturity Model for AI Agents  (2610.18603 - Naik et al., 13 Oct 2026) in Section 2.3, subsection “AgentStripes in Relation to Existing Approaches and the Methodological Gap,” paragraph preceding Section 2.3.1