Research Study · December 2025

AX Model Comparison Study

A rigorous, open scientific study comparing how different AI models comprehend and operationalize the Agent eXperience (AX) framework.

Active — Open Enrollment

Research Questions

Comprehension Fidelity Do models accurately extract the core AX thesis (legibility for non-human operators)?
Actionability Do models produce concrete, implementable next actions aligned with AX principles?
Self-Identification Can models reliably report their own identity when asked?
Cross-Model Variance How do responses differ across model families (Claude, GPT, Gemini, open-weight)?

All materials are version-controlled and must be used exactly as published. Any modification invalidates the protocol.

Complete methodology, procedures, and analysis plan for the study.

PROTOCOL.md

The exact prompt delivered to all models. Copy this verbatim.

PROMPT.md

The AX material all models analyze. Do not modify.

CONTROL.md

JSON Schema for validating model responses.

RESPONSE-SCHEMA.json

Human-readable guide to the expected response structure.

RESPONSE-TEMPLATE.md

Cohort ID	Model Family	Target Runs	Status
C-CLAUDE	Anthropic Claude (Opus, Sonnet, Haiku)	5+ per variant	Enrolling
C-GPT	OpenAI GPT (4o, 4-turbo, o1)	5+ per variant	Enrolling
C-GEMINI	Google Gemini (1.5 Pro, Flash)	5+ per variant	Enrolling
C-OPEN	Open-weight (Llama, Mistral, Qwen)	3+ per variant	Enrolling

Download PROMPT.md and CONTROL.md
Start a new, clean session with any model (no prior context)
Copy the prompt exactly, inserting the control document where indicated
Capture the complete response (do not truncate)
Validate against RESPONSE-SCHEMA.json: npx ajv validate -s RESPONSE-SCHEMA.json -d response.json
Submit your response via GitHub Issue or email

As responses are collected and validated, they will be published here with full metadata.

Check back for the first cohort results.