Methodology Paper · April 2026

Designing a Rigorous AI System for Tier 1 Behavior Support Plans

A paper-validation study of 22 kindergarten profiles.
Matt Edelstein, PsyD, BCBA-D — Kennedy Krieger Institute; Johns Hopkins University School of Medicine
Manal Mehta — Co-founder, NeuroPath Health

What this paper does

We describe the clinical decision tree, prompt scaffolding, and safety-net routing behind NeuroPath Health's Tier 1 Behavior Support Plan generator, and we document a paper-validation study in which the system produced plans for 22 synthetic kindergarten profiles constructed to span the behavioral spectrum clinicians actually encounter.

No real-student data were used. This is a design-and-methodology paper, not a customer case study — published so district and hospital leaders can evaluate the rigor of the system before inviting it into their environments.

Inside the paper

  • The 5-component Tier 1 plan scaffold (antecedents · replacement · consequences)
  • The 22-profile synthetic cohort and how it was constructed to span 95% of the behavioral problem-space
  • Interobserver Agreement (IOA) 5-dimension scoring rubric and pass criteria
  • Safety-Net routing — 5 mandatory-reporting triggers that halt plan generation
  • Sample plan walkthrough with commentary on what the model did, caught, and escalated
  • What this methodology does not yet prove — and what the next study (real-student IOA) will

About the authors

Matt Edelstein, PsyD, BCBA-D
Co-founder, NeuroPath Health
Director of the Brief Treatment Clinic at Kennedy Krieger Institute and Assistant Professor in the Department of Psychiatry & Behavioral Sciences at Johns Hopkins University School of Medicine. Clinically trained in applied behavior analysis, with a consulting caseload spanning complex Tier 2 and Tier 3 behavioral cases referred from school districts and pediatric systems across the mid-Atlantic.
Manal Mehta
Co-founder, NeuroPath Health
Product and engineering lead for NeuroPath Health. Background in healthcare operations and AI systems. Responsible for the clinical-decision-tree architecture, the prompt scaffolding, and the paper-validation protocol described in this study.