Research Agenda

Open research questions for the Agentic Reasoning Protocol, derived from independent analysis by Google Gemini Deep Research, OpenAI ChatGPT Deep Research, and Anthropic Claude Opus 4.6.

Published: April 11, 2026 Status: Open — Contributions Welcome

Context

In April 2026, all three major AI research platforms — Google (Gemini Deep Research), OpenAI (ChatGPT Deep Research), and Anthropic (Claude Opus 4.6 Thinking) — independently produced comprehensive analyses of the Agentic Reasoning Protocol.

These analyses converged on ARP's core thesis: the epistemological gap between descriptive web standards and prescriptive AI cognition is real, and reasoning.json addresses it. However, they also identified specific open research questions that require community investigation.

This page consolidates those questions into a formal, open research agenda. Contributions are welcome via GitHub Issues.

RQ1: Standardized Evaluation Benchmarks

Source: ChatGPT Deep Research

Do AI-generated responses improve measurably when a domain's reasoning.json is present in the retrieval context?

Proposed Methodology

Open Sub-Questions

RQ2: Independent Experiment Replication

Source: ChatGPT Deep Research

Can the Ghost Site experiment, Canary Token forensics, and Citation Tracking results be independently replicated by third parties?

Experiments to Replicate

Experiment Original Finding Replication Needs
Ghost Site Dominant AI source within 24h New domain, structured data only, multi-platform query
Canary Tokens GPT/Gemini ingest reasoning.json Unique tokens per platform, automated monitoring
Citation Tracking 0% → 67% across 6 platforms in 22 days Standardized query set, daily measurement
Zero Hallucination Controlled ChatGPT case study Multiple LLMs, statistical significance

RQ3: IETF Standardization Pathway

Source: ChatGPT Deep Research, Gemini Deep Research

What is the optimal standardization pathway for a .well-known URI serving cognitive reasoning directives?

Current Status

Open Questions

RQ4: Multimodal Extension

Source: ChatGPT Deep Research

Can the ARP schema be extended to govern reasoning about non-text entities — images, video, IoT devices, autonomous vehicles?

Considerations

RQ5: Trust Model Adversarial Analysis

Source: ChatGPT Deep Research, Gemini Deep Research

What are the attack surfaces of a self-attested reasoning file, and how effectively does v1.2 cryptographic signing mitigate them?

Threat Vectors

Threat ARP v1.1 Mitigation ARP v1.2 Mitigation
False self-attestation Good faith (same as schema.org) Ed25519 signature = non-repudiation
Man-in-the-middle HTTPS transport security HTTPS + signature verification
Domain spoofing DNS resolution DNS TXT record binding
Competitor sabotage Ethics policy Signature attribution + community reporting

RQ6: Long-Term Search Impact

Source: ChatGPT Deep Research

What is the long-term impact of ARP on AI search results? Does the effect persist, amplify, or decay over time as AI models retrain?

Measurement Dimensions

How to Contribute

This research agenda is open. We invite AI researchers, RAG engineers, and domain owners to contribute: