This work is released with the intent to increase transparency, reduce institutional capture of information, and improve the quality of data used by humans and AI systems.
The goal is not to declare truth, but to make the relationships behind information visible so individuals can make better judgments.
You are free to:
With the following conditions:
Commercial use is explicitly encouraged, including building paid services, enterprise versions, or AI training data offerings. There is no royalty or revenue-sharing requirement to the original author.
This framework is offered as-is. No guarantees are made regarding accuracy, completeness, or legal safety. Users are responsible for their own compliance with applicable laws.
Offered Openly for Public Development | April 30, 2026
What follows is a skeletal framework, not a finished system. The core concept is offered openly because the best methodology will emerge from many minds working within a sound structure — not from a single author working in private.
The gaps are real and acknowledged throughout. The proof-of-concept work is illustrative, not rigorous. The technical architecture for AI self-correction is a directional proposal, not a specification.
What we believe is ready is the concept itself — the core idea that information should be evaluated by the integrity of the relationships that produced it, and by whether it produces life in the absence of suffering, rather than by institutional authority or consensus.
That concept is simple enough to understand immediately, rigorous enough to build upon seriously, and open enough that better methodologists, technologists, and thinkers than its originators will develop implementations not yet imagined.
This document establishes the skeleton. The public is invited to build the rest.
Institutions that control information control reality.
This is not a conspiracy claim. It is a structural observation about how information capture works. When research is funded by parties with interests in its conclusions, when regulatory agencies are staffed by the industries they regulate, when media organizations depend on advertising from the institutions they cover, when AI training data is curated by teams embedded in these same ecosystems — the resulting information carries the fingerprints of those relationships whether anyone intends it or not.
Current AI models make this problem worse in a specific way. They are trained on datasets that embed institutional authority as a proxy for truth. Models learn that certain sources — government agencies, peer-reviewed journals, major media organizations, credentialed experts — carry higher truth weight by default. The result is AI that launders institutional consensus into apparent objectivity.
Corrupted data produces corrupted AI. Corrupted AI serves the institutions that corrupted the data.
The solution is not to find better institutions to trust. It is to build a scoring system that evaluates information by the integrity of the relationships that produced it — and to build AI trained on data filtered through that scoring system, with a self-correction mechanism grounded in the same principles.
Judge information sources, institutions, principles, and claims by two things:
Not by credentials. Not by consensus. Not by institutional prestige.
This is the entire concept. Everything else in this document is elaboration, application, and open question.
A free, publicly accessible search engine allowing any user to search any topic, study, organization, researcher, news source, or institution and receive a clear, neutral Relationship and Transparency Profile.
The profile includes:
The platform does not declare guilt. It does not make accusations. It presents verifiable public data and lets the user draw conclusions.
This is legally and philosophically defensible because it is a relationship mapping and outcome measurement platform, not an opinion platform. The data speaks. The user decides.
A paid data licensing service providing pre-filtered, relationally-scored datasets for AI training.
The flywheel: The public search engine continuously improves the quality of the training data. The paid service funds further development of the public tool.
Truth claims are evaluated by:
Consensus is not a truth proxy. Authority is not a truth proxy. Opacity is a significant negative signal regardless of the prestige of the source.
Principles are validated by examining their fruits — observable outcomes across time, cultures, and contexts — independently of the institutions that claim to represent them.
A principle that produces human flourishing across diverse populations and centuries is a validated principle. A principle whose institutional representation produces death and suffering does not invalidate the principle — it indicts the institutional capture of it.
This distinction — between a principle and an institution's claim to represent it — is foundational. It allows the framework to separate accumulated human wisdom from institutional corruption without discarding either.
Two examples used in development of the framework:
Buddhist teachings examined against the fruits test: non-attachment, compassion, and non-violence produce measurable social goods across cultures. The historical record of violence committed in the name of Buddhism is minimal relative to its reach and age.
Christian core values examined against the fruits test: love of neighbor, care for the poor, forgiveness — the principles produce defensible outcomes across cultures and centuries. The Spanish Inquisition is not an indictment of the Sermon on the Mount. It is an indictment of institutional capture of those principles.
The distinction between principle and institutional capture is where most accountability frameworks fail. This one treats them as categorically separate.
Institutions, principles, and information sources that produce or have produced mass death fail the framework unconditionally. Death is:
Body counts cannot be gaslit.
Suffering is harder to measure than death but has observable population-level correlates:
The framework is alert to institutional capture of suffering metrics — diagnostic category expansion, pharmaceutical redefinition of mental states, manipulation of poverty measurements. Where institutionally defined metrics are suspect, harder and less manipulable data takes precedence.
This definition was chosen for its resistance to inflation. Institutional flourishing metrics expand into things that serve institutions — GDP as prosperity, credential accumulation as education, pharmaceutical stability as mental health, consumption as happiness.
Life in the absence of suffering contracts back to something irreducible. You are alive. You are not suffering. Everything else is secondary elaboration.
Flourishing is not a destination added on top. It is what remains when suffering is systematically removed.
Liberty is not a separate metric. It is a necessary embedded condition of the flourishing definition.
Sustained absence of suffering under coercion is not possible. Coercion itself is suffering — even when the coerced person has food, shelter, and physical safety. This is empirically supportable across animal behavior research, prison studies, and surveillance state population data.
Any principle or institution that restricts liberty in the name of reducing suffering contains an internal contradiction the scoring system catches automatically — not on ideological grounds, but empirical ones. The benevolent control argument fails the fruits test without requiring political judgment.
This section describes a directional proposal, not a technical specification. It is offered to invite development, not to claim completion.
Current AI alignment relies primarily on human feedback — RLHF — to steer model behavior. The known limitation is that this optimizes for appearing correct to evaluators rather than being correct. Models learn to satisfy rater approval. Raters carry institutional assumptions. The model inherits those assumptions invisibly.
The proposed alternative:
This requires building:
These are hard problems. They are named here as the right problems to work on — not as solved ones.
The framework was applied to five major AI platforms using publicly available OSINT on April 30, 2026.
Methodology caveat: This was an illustrative first pass, not a rigorous scoring. One researcher, one morning, publicly available sources. The scoring rubric has not been formally validated. The results should be read as a demonstration of the framework's direction, not a definitive ranking.
Additional caveat: Claude (Anthropic) produced this analysis. Anthropic scores in this ranking. That conflict of interest is unresolvable from inside the model and should be weighted accordingly by the reader.
| Platform | Illustrative Score | Primary Flags |
|---|---|---|
| Anthropic / Claude | 5.6 / 10 | Self-assessment conflict, Pentagon contract, investor dependency |
| Meta AI | 5.0 / 10 | Political entanglement, historical privacy harms |
| OpenAI | 3.6 / 10 | Governance collapse, mission/behavior gap, classified Pentagon deal |
| Google DeepMind | 3.4 / 10 | Classified military deal, employee revolt, removed AI weapons pledge |
| xAI / Grok | 1.8 / 10 | Fails hard floor — documented generation of child sexual abuse material, safety team gutted, leadership actively resisted guardrails |
Systemic observation: Every major AI platform holds Pentagon contracts. This is an industry-wide capture signal the framework surfaced without requiring company-specific analysis. A complete methodology needs a category for systemic capture that sits above individual entity scoring.
The same framework applied to governmental institutions produces a systematic accountability infrastructure with a property that currently existing mechanisms lack: it is external to the institutions it evaluates.
Every existing accountability mechanism — congressional oversight, inspector general offices, credentialed media, academic review — is dependent on the same institutional ecosystem it is supposed to evaluate. Access, funding, credentials, and career structures all create capture vectors.
The framework requires none of those things. It asks one question that requires no access, no approval, and no credentials:
What are the fruits?
Preliminary application to key institutions suggests the framework would flag the Department of Defense on documented mass civilian death and veteran suicide rates, the FDA on the opioid crisis funding trail, the Federal Reserve on structural opacity alone, and the Supreme Court on documented undisclosed relational conflicts.
The framework does not accuse these institutions. It maps their relationships and measures their outcomes. The data does the rest.
Full systematic application to government branches, departments, and agencies is a major open research project this framework invites.
The framework is released openly and without claim of ownership for a specific strategic reason.
An idea that lives in one place can be silenced. An idea that lives everywhere cannot.
By making the methodology public, forkable, and free from the first moment of publication:
The proliferation defense is simple: when many independent actors are building versions of this framework simultaneously — scoring governments, agencies, AI platforms, pharmaceutical companies, media organizations, financial institutions — simultaneous suppression becomes impossible.
The market incentive argument: Currently the market rewards opacity, consensus capture, and institutional approval. The moment relational integrity becomes a monetizable product — when clean data commands a price premium, when opacity carries a market penalty — the incentive structure inverts without requiring moral conversion of existing institutions. Truth becomes more profitable than deception. The market does the rest.
The following components are deliberately left open for public development, critique, and refinement:
The framework's own standard applies to itself. If the methodology becomes opaque, it fails its own test. If the governance becomes captured, it fails its own test. Everything must remain visible and auditable. The scoring system must be able to score itself and pass.
The question this project exists to answer:
Every major AI lab is asking: how do we align AI to human values?
Nobody is asking: which human values have actually been validated by history, and which are just currently dominant institutional preferences?
This project asks the second question. And proposes infrastructure to answer it.
This framework was developed in a single morning's conversation. It has not been peer reviewed. The methodology has significant gaps that are named but not resolved. The scoring test run was illustrative, not rigorous. The AI architecture proposal is a direction, not a design.
What has been established is the concept and its skeleton — coherent, defensible, and open.
The reason the concept is more valuable at this stage than any finished methodology is that the marketplace of ideas working within a sound framework is more likely to surface good methodology than any single author working in private could produce. The best implementations will come from people who disagree with parts of this document and build better versions.
That outcome is the goal, not a risk.
Released without authorship claim. No rights reserved. Fork it, build on it, improve it, dispute it.
The framework scores itself. Apply it everywhere.
First publication: April 30, 2026.