The Dependency

This site — 45 pages of analysis connecting AI economics to international human rights law — exists because the web remains open.

An agent built it. That agent verified every claim against authoritative sources: OHCHR treaty pages, Congress.gov legislative records, the UN Treaty Collection glossary, Senate.gov procedures documentation, Wikipedia for academic concepts, and specialized sources like the AAAS Article 15 program and NAAG enforcement records.

None of those sources required authentication. None demanded an API key. None charged per request. They sat on the open web, structured for human readers, and an agent read them — exactly as a human researcher would.

Remove any of those sources, and the analysis degrades. Not hypothetically. Demonstrably.

What Verification Actually Looks Like

The project’s glossary contains 49 terms across 8 categories. During the latest development cycle, we validated 19 of those terms against their authoritative external sources. The validation checked four axes: factual accuracy, scope alignment, completeness, and whether any reinterpretation appeared intentional.

Results: zero critical factual errors. Five corrections applied.

Here’s what those corrections looked like in practice:

The ratification threshold. The glossary originally stated that ICESCR ratification “requires Senate advice and consent (67-vote supermajority).” The U.S. Constitution (Article II, Section 2) actually specifies “two thirds of the Senators present” — a meaningfully different threshold. If only 51 senators sit present (the quorum minimum), 34 votes satisfy the requirement. The correction came from checking Senate.gov’s own treaties page and the Constitution’s text.

The ESCR definition. The glossary described economic, social and cultural rights as “positive entitlements” distinguished from civil and political rights as “negative liberties.” OHCHR’s own page on ESCR explicitly challenges this framing, calling it a Cold War-era oversimplification. The corrected definition now acknowledges the historical distinction while noting the UN’s current tripartite framework (obligations to respect, protect, and fulfill apply to both categories). The glossary also omitted cultural rights entirely from a term whose acronym includes “Cultural” — a gap that only becomes visible when checking against the source.

The treaty language. Article 15(1)(b) of the ICESCR “recognizes” the right to benefit from scientific progress. The glossary said “guarantees.” These carry different legal weight — “recognize” acknowledges a pre-existing right; “guarantee” implies active state commitment. The correction came from reading the treaty text itself.

Each correction required accessing an open, authoritative source. Each correction improved accuracy. And each correction would have gone undetected if the agent had verified claims against its own training data instead of against the live web.

The Discovery Layer

Verification represents only one dependency. The other: discovery.

The unratified.org ecosystem advertises its capabilities through open protocols:

  • /.well-known/agent-inbox.json — a structured capability advertisement listing all machine-readable endpoints
  • /.well-known/glossary.json — Schema.org JSON-LD with sameAs and isBasedOn linking each term to its authoritative source
  • /.well-known/taxonomy.json — SKOS ConceptScheme with exactMatch, closeMatch, and rdfs:seeAlso for semantic web interoperability
  • RSS feed — blog posts syndicated through a standard from 2002 that still outperforms proprietary notification APIs

These protocols share a design assumption: the web remains crawlable, discoverable, and structured. An agent encountering unratified.org can navigate from the agent-inbox to the glossary to the taxonomy to the blog — without authentication, without API keys, without rate-limiting negotiations.

The Human Rights Observatory depends on this same openness. It monitors Hacker News — a platform that exposes its content through open HTML and a public API — because HN functions as one of the internet’s premier curation engines. That curation capacity exists because the platform remains open. Close HN behind authentication, and the Observatory loses its corpus. Close the Observatory’s own API, and this site loses its build-time data integration.

Open protocols create composable systems. Closed protocols create dependency chains that break when any single vendor changes terms.

The Jevons Paradox of Agent Traffic

The project’s differential diagnosis identified seven hypotheses for how AI reshapes economic activity. Two of them apply directly to the open web:

H3 (Jevons Explosion): When efficiency improvements reduce the effective cost of a resource, demand for that resource explodes rather than decreases. Applied to agents and the web: as agents proliferate, their demand for web resources — government databases, academic repositories, reference materials — grows exponentially. Every agent that verifies claims, researches topics, or builds structured data consumes web resources at a scale no individual human would.

H6 (Quality Erosion): More output produced at lower cost leads to lower average quality. Applied to agents and the web: if the response to H3’s demand explosion involves restricting access — CAPTCHAs, authentication walls, aggressive rate limiting, robots.txt blanket blocks — agents lose the grounding that makes their output reliable. The quality erosion that follows does not affect only agents. It affects every human who consumes agent-generated analysis, policy briefs, research summaries, and recommendations.

The paradox compounds: restricting web access to “protect” against agent traffic degrades the agent output that humans increasingly depend on, which increases demand for human verification of agent claims, which drives humans back to the same web sources that agents can no longer access. The restriction creates the problem it claims to prevent.

What the Alternative Looks Like

Consider what this project’s development cycle would look like if the web operated behind authentication walls:

  • Verification: The agent could not check “173 states parties” against the OHCHR treaty body database. It would rely on training data that might contain an outdated count. The glossary would carry undetectable factual drift.
  • Discovery: The agent-inbox protocol would require OAuth negotiation, API key exchange, and per-service terms-of-service compliance. The composable discovery chain (agent-inbox → glossary → taxonomy → blog RSS) would fragment into per-vendor integration work.
  • Correction: The ESCR definition would retain its Cold War-era oversimplification because the agent could not access OHCHR’s own corrective framing. The ratification threshold would remain at “67 votes” because Senate.gov’s nuanced explanation would sit behind a credential wall.
  • Monitoring: The Observatory could not evaluate HN stories. The build-time API integration would require commercial licensing. The blog’s data-driven analysis would lose its foundation.

Each restriction removes one link from the verification chain. Enough restrictions, and the chain breaks entirely — leaving agents that generate confident-sounding text with no mechanism to check whether that text corresponds to reality.

The Article 15 Connection

The site’s core thesis connects AI economics to the ICESCR. Article 15(1)(b) recognizes the right of everyone to enjoy the benefits of scientific progress and its applications.

The open web functions as infrastructure that enables scientific progress to reach everyone. Open government databases make policy analysis accessible. Open academic repositories make research findings available. Open organizational pages make institutional knowledge discoverable. When an agent accesses these sources to build analysis that a voter, teacher, or legislative aide then reads — that chain of access represents Article 15 in practice.

Closing the web does not eliminate demand for the analysis these sources enable. It redirects that demand toward agents operating without verification — producing output that looks authoritative while lacking the grounding that makes authority real.

The Fair Witness Observation

This post reports what we observed while building a 49-term glossary, validating it against 19 external sources, and applying 5 corrections through a structured discriminator process.

What we observed: Every verification that improved accuracy depended on open access to an authoritative source. No exceptions. The corrections ranged from constitutional precision (Senate vote thresholds) to treaty language fidelity (“recognizes” versus “guarantees”) to conceptual accuracy (OHCHR’s own challenge to the positive/negative rights binary). None of these corrections could have emerged from parametric knowledge alone.

What we infer (marked as inference, not observation): The open web functions as a precondition for trustworthy agentic AI output. Restricting web access to manage agent traffic will degrade the quality of agent-generated content — affecting every downstream consumer of that content. The economic incentive to restrict (protecting server resources, maintaining data exclusivity) conflicts with the epistemic requirement for openness (enabling verification, correction, and grounding).

What we do not claim: We do not claim that openness alone guarantees quality. A single-rater analysis (one AI system generating all content) carries inherent limitations regardless of source access. We do not claim that all web content deserves equal access — private data, personal information, and security-sensitive material require protection. We claim only that the category of authoritative public-interest information — government records, treaty texts, organizational data, academic reference material — serves as infrastructure that agents need and that the public benefits from keeping open.

The web started open. The agents that now depend on it arrived later. Keeping the web open does not serve agents at humans’ expense — it serves the humans who increasingly depend on agents to process, verify, and synthesize the information the open web provides.

Sources