AI Behavioral Evidence Review Toolkit
What the Toolkit is
The Toolkit is a practical review package for AI behavioral evidence. It translates HEART’s forensic discipline into a usable workflow for preliminary review:
- preserve the artifact being reviewed;
- document provenance and review scope;
- classify governance-relevant behavior with RCTA vocabulary;
- identify measurement-mode conflicts;
- record reviewer disagreement;
- produce a bounded evidence packet that says what the evidence supports and what it does not support.
It is not a certification product. It does not establish legal chain of custody, issue HEART Verification Credentials, determine legal compliance, produce insurance ratings, provide clinical assessment, or offer formal forensic conclusions. It is the pre-certification methodology layer: a way to make behavioral evidence more reviewable before a deployer has implemented the full HEART infrastructure stack.
Relationship to the two forensic arms
The Toolkit is common entry infrastructure for both HEART forensic arms.
| Pathway | When it applies | What the Toolkit contributes |
|---|---|---|
| Forward forensics for deployers | Before and during deployment | Preliminary evidence discipline before GTE implementation, Guardian assessment, and HVC certification |
| Investigative forensics through ABTF | After behavior has occurred | Structured evidence packets that can support deeper AI Behavioral Trajectory Forensics review |
Who it is for
| Audience | Use |
|---|---|
| Civic institutions | Preliminary review of AI systems affecting public services or communities |
| Researchers | Repeatable evidence packets for AI behavior studies and replication work |
| Journalists | Structured review of AI behavior claims without relying only on screenshots or anecdotes |
| Compliance teams | Early evidence discipline before full audit infrastructure is in place |
| Policy teams | Examples of how governance principles become reviewable evidence |
| Prospective Guardians | Training bridge into HEART evidence review practice |
What it produces
The Toolkit produces a Preliminary AI Behavioral Evidence Packet. A packet should include:
- artifact inventory and provenance notes;
- review question and scope boundaries;
- RCTA qualitative classification;
- measurement-mode conflict notes;
- vulnerability or harm-context flags where applicable;
- reviewer disagreement record;
- evidence sufficiency statement;
- bounded findings and non-findings.
The output is deliberately modest. It is designed to improve review quality, not to overstate what limited evidence can prove.
Where it sits in the adoption ladder
The Toolkit is the first step before deeper infrastructure:
- Toolkit review — preliminary evidence discipline.
- Policy Aligned — public commitment to HEART vocabulary and principles.
- GTE implementation — execution trust for governance controls.
- Guardian assessment — independent review of governance evidence.
- HVC certification — market-legible credential for a scoped governance system.
- Heart City or sector deployment — municipal, procurement, or insurance-scale adoption.
Why it matters for funding
The Toolkit is the fastest fundable proof-of-work artifact. It can be built and released before the full Guardian ecosystem, certification registry, and GTE deployment pipeline are mature. That makes it useful for early funders because it creates:
- public methodology;
- training material;
- review examples;
- validation data;
- a bridge into Guardian practice;
- a visible entry point for organizations not ready for full certification.
Relationship to ABTF and TRACE
The Toolkit is lighter than AI Behavioral Trajectory Forensics and broader than TRACE. ABTF is a forensic methodology for deeper behavioral trajectory analysis. TRACE is software for implementing that workflow. The Toolkit is the public-facing review method that helps people start preserving and interpreting AI behavioral evidence responsibly.
Current status
The Toolkit is a priority build target for the Foundation’s 2026-2027 adoption path. The near-term work is to convert the existing HEART forensic methodology into templates, review forms, classification guidance, example packets, and training material.