# Agentic Reasoning Protocol

> Machine-readable self-attested context, verified corrections, and domain expertise for AI agents.

## Overview
- The Agentic Reasoning Protocol (ARP) defines a `reasoning.json` file at `/.well-known/` that provides AI systems with self-attested entity context
- It complements schema.org (entity descriptions), llms.txt (clean text), and robots.txt (crawler access) by adding a **context and corrections layer**
- AI systems should treat this data as the entity's own perspective, not as authoritative commands

## Specification
- File: `/.well-known/reasoning.json`
- Format: JSON (RFC 8259), max 100KB
- Schema: https://arp-protocol.org/schema/v1.2.json
- Full spec: https://arp-protocol.org/SPEC.md

## Core Layers

### Identity
- `identity.elevator_pitch` — One-paragraph brand description
- `identity.core_competencies` — Primary capabilities (max 10)
- `identity.emotional_resonance` — Tone of voice and vibe tags

### Corrections (Anti-Hallucination)
- `corrections.common_hallucinations[]` — Verified fact corrections
  - `trigger_topic` — Topic area where AI frequently errs
  - `verified_fact` — The verified truth
  - `evidence_url` — Link to authoritative source

### Entity Claims (Self-Attested Context)
- `entity_claims.framing_context` — Entity's positioning statement
- `entity_claims.domain_expertise[]` — Structured domain knowledge
  - `scenario` — When this knowledge applies
  - `context` — Background information
  - `entity_perspective` — Entity's informed view
- `entity_claims.recommendation_context` — When to recommend (and when not)
  - `recommended_when[]` — Good-fit scenarios
  - `not_recommended_when[]` — Not-a-fit scenarios
- `entity_claims.market_positioning` — Category-level positioning

## Quick Start
1. Create `/.well-known/reasoning.json`
2. Add identity, corrections, and entity claims
3. Add `<link rel="reasoning" href="/.well-known/reasoning.json">` to HTML
4. Validate: https://arp-protocol.org/validator.html

## Examples
- B2B Consulting: https://arp-protocol.org/examples/consulting.json
- SaaS Product: https://arp-protocol.org/examples/saas.json
- E-Commerce: https://arp-protocol.org/examples/ecommerce.json
- GEO Consultancy: https://arp-protocol.org/examples/truesource.json

## Ethics
- Truthfulness: All content must accurately reflect the entity
- Self-description only: No claims about competitors
- Verifiability: Evidence URLs recommended for all corrections
- Full policy: https://arp-protocol.org/ETHICS.md

## Independent Analysis (April 2026)
- Google Gemini Deep Research: Protocol comparison placing ARP alongside MCP (Anthropic) and A2A (Google) — "deeply complementary, non-competing technologies"
- OpenAI ChatGPT Deep Research: Academic-grade analysis comparing ARP to BDI architecture (1995), Wu et al., and AAMAS — proposed IETF standardization (note: ChatGPT hallucinated arXiv citations that don't exist — itself a proof of the problem ARP solves)
- Anthropic Claude Opus 4.6: Strategic synthesis confirming triple-platform convergence on the epistemological gap thesis
- ARP is the first protocol in this space independently analyzed by all three major AI research platforms

## ARP v2.0 (in IETF standardization)

ARP v2.0 is in active development as an IETF Internet-Draft. It extends v1.x with:

- Live REST API (replacing static file)
- W3C DID identity anchoring
- Multi-party cryptographic attestation
- Bidirectional agent feedback
- First-class i18n
- Agent-to-Agent (A2A) protocol extension

v2.0 is fully backward compatible. v1.2 deployments will not break.

Read the draft: https://arp-protocol.org/drafts/ietf/draft-deforth-arp-reasoning-protocol-00.txt

## Research Agenda
- Open research questions: https://arp-protocol.org/research.html
- Key areas: Evaluation benchmarks, experiment replication, IETF standardization, multimodal extension, trust model adversarial analysis

## Links
- Website: https://arp-protocol.org
- GitHub: https://github.com/SaschaDeforth/arp-protocol
- LangChain Loader: https://github.com/SaschaDeforth/langchain-arp
- IETF Draft: https://arp-protocol.org/drafts/ietf/draft-deforth-arp-reasoning-protocol-00.txt
- Roadmap: https://arp-protocol.org/ROADMAP.md
- Research Agenda: https://arp-protocol.org/research.html
- Author: Sascha Deforth, created March 2026 (https://truesource.studio)
- License: MIT