part of EQUORA Institute
HUEN
EQUORA Institute · Research method · Trilith Method™
How We Research

How the pages on this site come about

Behind the pages on this site stands a method in which AI does a great deal of the work, while a human sets the bar for evidence.

At the foot of every content page there is a short block noting that the page comes out of research at the EQUORA Institute. This page sets out what that means in practice: how the work is divided between human and machine, what is checked against what, and where the decision stays in human hands.

01 · AI-first research
The unit of research is a person plus a standing AI team

Most institutions today are AI-supported: a researcher does the work and occasionally reaches for a model. The model is a tool on the desk, and the organisational unit is still one person. The EQUORA Institute inverts this: the basic unit of research is a researcher plus a standing AI team, every project begins with agents already in place, knowledge lives in a shared graph instead of scattered documents, and first drafts of publications and proposals are prepared by AI.

The human moves up in this picture, to the work that cannot be delegated: what is worth studying, what counts as sufficient evidence, and what may appear under the institute's name. In our own words, every researcher operates a research team; the team is mostly artificial, and the judgment is entirely human.

02 · The Trilith Method™
Three layers, separated by how far the machine is trusted

A full research lifecycle can be carried by AI, from literature discovery to communication. The mistake that makes such organisations fragile is treating every function as equally trustworthy. The Trilith Method™ therefore separates them into three layers.

I. Generative where AI leads
High-volume, fast-iterating work where the machine's speed is a clean gain.
  • continuous literature discovery across fields
  • experiment design: proposing protocols, parameter spaces, controls
  • simulation and the running of computational models
  • the living knowledge graph connecting projects
  • first drafts: papers, summaries, public explainers
II. Verification where AI checks AI, adversarially
The core of the method. No single model validates its own kind of output.
  • a contradictory protocol: one model asserts, a second refutes, a third adjudicates — on the record
  • multi-model triangulation, because different architectures fail differently
  • research log: every AI contribution timestamped, attributed and reproducible
  • reproducibility checks: a claim must survive re-derivation
III. Governance where only humans decide
The non-delegable layer.
  • what is worth studying — problem selection is a human, strategic act
  • what counts as sufficient evidence — the bar is set by people
  • what goes out under our name — final epistemic and ethical responsibility is human

The name is deliberate: a trilith is a structure of three stones, two uprights carrying a third — here too the three layers stand only because each bears on the others. Within that structure, reliable agreement is reached by first putting models into structured disagreement: consensus is an output of the method rather than its assumption.

The method is defined by how it disagrees with itself.

03 · Why the architecture matters
The failure mode we design against

The danger of an AI-first organisation is not slowness. The danger is being fast, fluent and wrong at scale. Language models produce output that sounds persuasive whether or not it is true, and an organisation that generates research at machine speed can generate convincing error at machine speed.

There is a quieter danger too: epistemic monoculture. If every researcher leans on the same few foundation models, the organisation becomes blind to the same things, inheriting the models' strengths and blind spots together. Model diversity is therefore a design requirement rather than a convenience.

Failure is recorded, not discarded

Traditional science is bad at remembering what did not work: refuted hypotheses and dead ends vanish, and the next researcher walks the same blind alley again. A knowledge graph holds negatives as easily as positives, so refuted hypotheses and failed approaches stay alongside confirmed findings, attributed and searchable.

04 · Two places, two standards
The institute's research output and the pages here

The EQUORA Institute's research plans and notes answer how an investigation is built: what we claim, on what evidence, and what could overturn it. These change slowly, carry version numbers, and are structured by the three layers of the Trilith Method™. The pages on iterators.org are the outward-facing side of the same work: a claim or a chart that stands on its own and holds at a given moment of the research.

EQUORA Institute

Completeness is the aim. Version numbers, evidentiary layers, and stated conditions of refutation.

iterators.org

Usability is the aim. A readable page that ends in something the reader can act on.

It follows that a page here is not an institutional position. It captures one state of the research, and that state rests on the findings available at the time of publication. Where a page has been updated, the date shows it.

05 · Where responsibility sits
The judgment stays human

AI takes part throughout the research process as a thinking partner: it searches, proposes, objects, writes first drafts, and supplies the code behind the interactive elements. What you read here nonetheless appeared by a human decision, and responsibility for interpretation and publication rests there too. This is the third layer of the Trilith Method™, and it is the point at which the method is deliberately inefficient.

In practice this also means errors are public. Where a claim on a page was inaccurate, the correction stays on the page, dated, naming what the earlier version said. This is the same standard the pages here apply to others, so it is fair to apply it to ourselves.

The machine supplies speed. A person sets the bar, and a person puts their name to it.

Further reading