Epistemology

The Foundation's premise

Every measuring instrument has a bounded cone of illumination. The discipline of asking what an instrument can and cannot adjudicate is the foundation of honest measurement. The Open Honest Foundation operates on this discipline as a methodological commitment. The Foundation's three open standards, the Honest Framework, the Slop Audit, and MÉTRON, are three instruments operating on this discipline applied to different domains: the structural correctness of code, the quality and compliance posture of enterprise software, and the linguistic foundations of artificial-intelligence systems. The premise is older than any of the standards. The standards are the operational expression of the premise.

The cone of light

In operational terms, an instrument reveals the structure it was designed to reveal and is silent on structure outside its cone. A microscope cannot adjudicate questions about galaxies, and a telescope cannot adjudicate questions about cell membranes. The same is true of less obvious instruments: a test suite reveals the cases its author thought to write and is silent on the cases its author did not. A compliance audit reveals what its dimensions were designed to expose and is silent on properties outside those dimensions. A language-model benchmark reveals capability differences along the axes it measures and is silent on capability differences along axes it does not.

The consequence matters. Defects, bugs, mistranslations, miscalibrations, and misunderstandings are often instrument-cone artifacts, not problem-domain necessities. A problem that resists more effort within the original instrument can sometimes dissolve when a different instrument is brought to bear, or when the limits of the current instrument are explicitly named. Refusing to make this distinction is what produces the recurring pattern in which more measurement, more testing, and more compute fail to close gaps that an instrument-shift would close trivially. The Foundation's standards are designed against this pattern.

The unnamed instrument

The highest-stakes instruments in the world today are frontier AI systems, and they are the clearest case of a cone left unnamed. The phrase needs unpacking, because none of its three parts is obvious on its face.

Confounded. A frontier model is trained by changing many things at once, the data, the scale, the architecture, the tokenizer, the training procedure, and tuning the whole to maximize benchmark scores. When everything varies together and the goal is to win the benchmark rather than to isolate a cause, no capability can be attributed to any one factor. This is the opposite of a controlled experiment, where one variable is changed and the rest held fixed; it is optimization, not explanation, and optimization cannot tell you why.

Private. The training data, the exact procedure, and much of the architecture are undisclosed, for competitive reasons. What cannot be seen cannot be checked or reproduced, and a result that no outside party can reproduce is not yet knowledge.

No falsifier. Nothing about the system was committed in advance as a prediction that reality could have refuted. The model is built to work and then described after the fact. A claim that could not have failed any test is not a scientific claim about why the system behaves as it does.

A further problem reaches the measuring stick itself. Even the benchmark the optimization chases may not measure what it is taken to measure. The Foundation's own cross-linguistic work found that standard child-scale benchmark scores responded principally to properties of the tokenizer and the answer template rather than to the task competence they are assumed to track, and reported several of its own apparent results as artifacts on those grounds. A system tuned to raise such a score may be raising something other than the capability the score is taken to name. The cone-of-light discipline applies to the benchmark as much as to the model.

Put together, the frontier model is an instrument that will not name its own cone. Two consequences follow. As epistemology, it cannot ground a claim about why it behaves as it does, because the process that produced it was neither controlled nor disclosed. As safety, the controlled, public, reproducible knowledge needed to understand, predict, or trust such a system is exactly the knowledge its construction does not produce, so independent verification has no object to work on.

This gap is not an oversight, and naming why it persists is part of the discipline. The institutions with the means to close it are structurally directed away from it: their value networks reward capability at scale, and controlled, small-scale, public, falsifiable measurement neither advances that nor can be done without exposing proprietary method. The Foundation reads this not as a market opening but as a public-interest gap: knowledge the well-resourced are disincentivized to produce has to be produced by those whose purpose is the knowledge itself. MÉTRON is built for exactly that, a controlled cross-linguistic experiment a researcher outside any lab can run and independently verify, because a result no one outside the lab can reproduce is not yet knowledge.

Three instruments, one discipline

Honest Framework

Cone: the structural correctness of code at the type and composition level. What it makes visible: which categories of bug are eliminated by construction rather than caught by testing. The framework is an instrument for reading code in a way that reveals its invariants. Where conventional engineering practice waits for defects to surface and then writes tests to catch them, the Honest Framework treats the structural shape of code as the thing under inspection: a function whose type signature precludes a class of failure cannot exhibit that failure regardless of test coverage. Errors are designed out at the structural level rather than caught after the fact. The framework's cone is narrow and deliberate, and the structure inside it is exactly what other instruments are silent on.

Slop Audit

Cone: enterprise software quality across eighteen named dimensions, mapped to compliance frameworks including SOC 2 Trust Services Criteria, NIST SP 800-53, OWASP ASVS, and ISO/IEC 25010. What it makes visible: structural properties that conventional metrics cannot illuminate, such as whether an authentication system fails closed under load, whether an entitlement layer is uniform across endpoints, and whether secret-handling is consistent across services. The audit is an instrument designed to catch what other instruments miss by construction, by inspecting compliance-mapped artifacts and qualitative markers that aggregate metrics flatten away. Its cone is the structure that lives between the line of code and the line of policy.

MÉTRON Framework

Cone: the structural properties of natural language as deposited in pre-training data. What it makes visible: that capabilities attributed to neural-network scale are observable through controlled cross-linguistic ablation as properties of training-language morphology. The platform is an instrument for calibrating capability claims that compute-and-architecture explanations cannot adjudicate. Where conventional language-model evaluation treats the model as a unit and asks how big or how trained, MÉTRON treats the training-language as the unit and asks what structural features of that language permit which capabilities. Its cone is the training language, not the system.

The convergence-of-signals method

The Foundation formalizes a methodological commitment that follows directly from the cone-of-light premise: when multiple independent instruments (empirical experiments, philosophical inquiries, theological traditions, scholarly disciplines with different observational commitments) meet at the same boundary, the convergence licenses an inference that no single instrument alone could license. This is the convergence-of-signals method. The Foundation publishes a peer-reviewable specification of the method with worked examples drawn from its cross-linguistic empirical work and from cross-traditional scholarly convenings, released as a standalone open-access deposit under permissive license with a persistent digital identifier.

The method is the methodological foundation that gives the Foundation's empirical research program its interpretive coherence. It is also the protocol the Foundation uses to scope cross-traditional convenings and to assess the validity of inferences drawn from heterogeneous data sources.

A worked example

The discipline shown here is the Foundation's own premise, and researching the methodology of empirical inquiry itself is one of its stated purposes, alongside the measurement and production of verifiable software. The worked example runs that method on the cross-linguistic and cognitive line of inquiry behind MÉTRON: a claim about the nature of intelligence is proposed at the ontological tier, tested at the empirical tier, and governed throughout by the epistemological tier, the cone-of-light discipline articulated in The Robot in the Dark (cited below). The method is the Foundation's; the deepest of these claims draw on the founder's scholarly work, which the Foundation cites rather than authors.

How the three tiers build an evolving discovery

A claim is proposed, tested under the rules, and what the evidence supports becomes the ground for the next, deeper claim.

Rows are the three tiers: ontological (proposes truth claims), empirical (tests them), and epistemological (the governing rules). Columns are stages of one discovery, from supported through being established, proposed, and the open question.
Supported Being established you are here Proposed The open question
Ontological proposes truth claims Intelligence is not tied to one physical medium. Intelligence, cognition and consciousness are three different things. The same intelligence shows up across many independent measures. Such intelligence can act on its own, as an agent.
Empirical tests them Train identical models on different languages; compare what they learn. Separate the three; show consciousness narrates but does not do the thinking. Many unrelated experiments; see whether they point the same way. Define agency so it can fail the test; look for it across substrates.
Epistemological The rules: what would count as support, and where each instrument's reach ends. It governs every stage, and grows as the experiments map its edges.
Solid arrows, mutual support: a claim is sent down to be tested, and the evidence travels back up to support it or send it back for revision.
Dotted arrows, governance: the foundation governs each test, and every closed door (a failed probe, a named limit) maps a new edge back into the foundation.

Why this stance matters

The discipline matters operationally for three reasons. First, it is the through-line that makes the three Foundation-governed standards cohere as a portfolio rather than a pile: they are not three unrelated tools but three instruments built on the same epistemological commitment. Second, it differentiates the Foundation's instruments from the larger field's tendency to make existing instruments brighter rather than to ask what those instruments cannot illuminate in principle. More tests do not reveal what tests cannot see. More benchmarks do not reveal what benchmarks cannot measure. The Foundation's standards are designed to extend the inventory of cones, not to amplify any single one. Third, the discipline is what lets the Foundation's standards survive paradigm shifts in the fields they measure: an instrument whose cone is honestly named is portable across changes in the surrounding theory.

The discipline turns inward

An instrument-aware stance is worthless if it exempts its own instruments. The Foundation applies to itself the same discipline it applies to the systems it measures. It names the cone of each of its own standards rather than claiming a view from nowhere. It pre-registers its research so that a failed prediction counts against it. It reports its own false positives as readily as its findings: MÉTRON treats a result that proves to be an artifact of its own pipeline as an artifact, not a finding. And it stakes its central claims on public refutation rather than private assurance, from the Slop Audit's standing challenge to break its auto-generated test suite to the open invitation to refute the Honest Framework's correctness claim. An instrument that audits others while exempting itself is the failure this Foundation exists to correct.

Cross-traditional scholarly convenings

The Foundation organizes and hosts cross-traditional scholarly convenings that bring together scholars from philosophy, theology, computational linguistics, cognitive science, and adjacent humanities and social-science fields around the convergence-of-signals method. The convenings produce peer-reviewable proceedings and method specifications, contributing to the scholarly literature on multi-traditional methodology and on the relationship between empirical and contemplative inquiry. The Linguistic Telescope, a Foundation publication authored by the founder as part of his charitable scholarly service, provides the focus and locus around which these convenings are structured.

Sources

The Foundation cites these works as the lineage of its epistemological stance. The works themselves remain the property of their authors and publishers.