examplesOverview

Examples Overview

This section provides complete, validated ORP documents demonstrating how the protocol works in practice.

What Makes a Good Example?

A strong ORP example document:

  1. Addresses a real policy question — Not a hypothetical exercise
  2. Shows all layers — Demonstrates how L1-L5 work together
  3. Makes substantive claims — Takes positions that can be debated and forked
  4. Documents uncertainties — Honest about data limitations and assumptions
  5. Invites disagreement — Explicitly encourages forking

Available Examples

ORP v0.1: Policy Analysis

Danish Property Tax Reform (ORP-Full)

Danish Property Tax Reform: LVT + Smoothed Profit Tax

A complete ORP-Full proposal replacing Denmark’s current property tax system with a Land Value Tax (LVT) combined with a profit tax on gains at sale. Demonstrates all five layers with real economic modeling, stakeholder analysis, and transparent decision documentation.

Compliance: ORP-Full (all 5 layers)
Status: Open for comment
Author: Asbjørn
Domain: Policy — Tax reform

ORP v0.2: AI Training Datasets (Post-Hoc Reconstruction)

The following examples demonstrate post-hoc reconstruction of major AI training datasets using ORP v0.2. These show how ORP can be applied retrospectively to document datasets that lack adequate transparency documentation.

Key features demonstrated:

  • Post-hoc reconstruction (post_hoc: true flag)
  • Extension system (orp-ai-training-v1, orp-license-v1)
  • Accountability gap documentation
  • Vulnerable population assessment
  • Multi-modal data handling

Comparative Analysis

AI Training Datasets: Comparative Analysis

Comprehensive analysis across all four AI training dataset reconstructions, revealing patterns of constitutive opacity, systemic harm, and accountability gaps across the AI ecosystem. Shows what ORP enables for prospective governance.

Datasets analyzed: ImageNet, LAION-5B, Common Crawl, GitHub Copilot
Total documentation: 8,328 lines of ORP YAML
Status: Post-hoc reconstruction

ImageNet ILSVRC-2012 Training Data

ImageNet Training Data Reconstruction

The most influential training dataset in AI history (1.28M images). Post-hoc reconstruction revealing how constitutive decisions from 2009 (WordNet taxonomy, web scraping) shaped computer vision AI for over a decade.

Compliance: ORP-PostHoc
Key findings: 80%+ North American/European images, 54% person categories later found offensive
Domain: AI Training — Computer Vision

LAION-5B Training Data

LAION-5B Training Data Reconstruction

Web-scale image-text dataset (5.85B pairs) used to train Stable Diffusion. Post-hoc reconstruction revealing how absence of prospective safety filtering led to 3,226 CSAM URLs distributed for 21 months.

Compliance: ORP-PostHoc
Key findings: 3,226 CSAM URLs, medical privacy violations, no hash-matching
Domain: AI Training — Multi-modal (image-text)

Common Crawl Training Data

Common Crawl Training Data Reconstruction

The largest open web crawl dataset (250TB, 250B pages) training GPT-3, GPT-4, LLaMA, Claude, Gemini. Post-hoc reconstruction revealing how constitutive decisions from 2008 resulted in 90%+ English hegemony and linguistic extinction.

Compliance: ORP-PostHoc
Key findings: 90%+ English content, 4-5B people affected by linguistic exclusion
Domain: AI Training — NLP (text)

GitHub Copilot Training Data

GitHub Copilot Training Data Reconstruction

Code training dataset (100M+ repositories) powering AI pair programming tool with $100M+ revenue. Post-hoc reconstruction revealing GPL violations, license attribution stripping, and open source labor exploitation.

Compliance: ORP-PostHoc
Key findings: $1-10B lawsuit pending, 10M+ OSS maintainers receive $0 compensation
Domain: AI Training — Code

How to Use These Examples

As a Learning Tool

  • Start with the overview — Understand what the proposal is trying to achieve
  • Read layer by layer — See how each layer builds on the previous one
  • Study the YAML — See the actual validated ORP document structure

As a Template

  • Fork the document — Download the YAML and modify it for your own policy
  • Adapt the structure — Use the same layer organization for your domain
  • Keep what works — Copy sections that apply to your use case

As a Debate

  • Challenge the assumptions — Fork and change the variables
  • Test alternative scenarios — Modify L2 to explore different outcomes
  • Add stakeholders — Expand L3 to include groups you think were missed
  • Propose alternatives — Fork the entire document with your preferred policy

Validating Examples

All examples in this section are validated ORP documents. You can verify them yourself:

# Download the example
curl -O https://publicreasonproject.org/examples/danish_property_tax_reform.yaml
 
# Validate it
orp validate danish_property_tax_reform.yaml
 
# Check compliance level
orp check danish_property_tax_reform.yaml

Expected output:

Valid ORP document
Compliance Level: ORP-Full
Layers Present: 5/5 (Header + L1-L5)

Contributing Your Own Examples

Have you created an ORP document you’d like to share? We welcome contributions!

Requirements:

  • Must validate as at least ORP-Standard
  • Must address a real policy, research, or organizational question
  • Must be open for forking (preferably with explicit invitation)
  • Should demonstrate thoughtful use of each layer

See our Contributing Guide for how to submit your example.

Example Formats

Examples are provided in multiple formats:

  • This website — Web-friendly walkthrough with explanations
  • YAML — The actual validated ORP document
  • Source repository — Raw file in the GitLab repo

You can use whichever format is most helpful for your needs.

Next: Danish Property Tax Reform →

ORP vs Existing StandardsDanish Property Tax Reform