Neuronautix · Transparency
This website is built by a human-directed AI agent pipeline, and it says so. Neuronautix publishes on AI agents in research, provenance, and FAIR metadata. Declaring how its own site is made is the minimum consistency we owe readers — and it tells a sceptical visitor exactly which parts to check first.
The machine-readable source of truth is /AI-DECLARATION.md, following the AI-DECLARATION.md specification v0.1.2. This page is the human rendering of it.
AI-DECLARATION · auto
The short version
auto” describes who typed it — not who answers for itAn honest declaration is an enabler, not a confession. When you know which parts were generated, you can go straight to them and verify. Everything else on this page is the detail behind that one claim.
The site's HTML, CSS, JavaScript, Python tooling, and tests are written by an AI agent running to completion from a prompt, under the agent roles defined in AGENTS.md and CLAUDE.md. About half of this repository's commits are authored by the agent directly; most of the rest are human merges of agent branches.
The accountable author and owner of every published claim — ORCID 0000-0003-4820-7951. The human-approved record, not raw model output, is the source of truth. Nothing reaches the live site without human review and merge.
If you want to re-verify something, start with assets/, scripts/, tests/, and rag/ — those are fully generated. The prose in notes/ is generated and gated and graded claim-by-claim.
Declaration · processes
The specification defines six processes and six levels of involvement. The overall level of a repository must equal the highest level present in any process — which is why this site declares auto.
| Process | Level | What that means here |
|---|---|---|
| Design | pair | Site architecture, positioning, the three-pillar structure, the fair.md and trust.md conventions, and the editorial line are human-originated and argued out with the agent. The human understands the internals. |
| Implementation | auto | HTML, CSS, JavaScript, and Python are written by the agent, running autonomously from a prompt to a reviewable branch. |
| Testing | auto | Everything under tests/, and the CI checks that run on every push, were agent-generated and then reviewed. |
| Documentation | copilot | Notes and knowledge-base articles are agent-drafted but gated: two mandatory human approval gates — outline, then draft — before anything is published. Sources are retrieval-grounded; fabricated citations are not permitted. |
| Review | copilot | A review agent performs the first pass and reports — it never approves. Human approval is required, and external review has produced corrections that are visible in the git history. |
| Deployment | assist | The GitHub Pages workflow and CI configuration are largely human-set, with agent help on specific steps. |
Declaration · components
The same declaration, resolved to directories and files. This is the part a sceptic can act on directly.
| Component | Level | Notes |
|---|---|---|
assets/ | auto | All CSS, JavaScript, and figure generation. |
scripts/ | auto | Knowledge-base tooling, citation graph, QA and export scripts. |
tests/ | auto | The full test suite. |
rag/ | auto | Local RAG prototype — private, never deployed. |
knowledge/ | copilot | Source-backed knowledge articles; drafted by agent, human-approved. |
notes/ | copilot | Cited analytical essays; two human gates, plus per-claim epistemic grading. |
presentations/ | copilot | Decks built from an approved brief and source map; never merged without review. |
AI-DECLARATION.md | copilot | This declaration: agent-drafted, human-verified and signed off. |
CLAUDE.md, AGENTS.md | pair | The agent instructions themselves — designed jointly. |
fair.md, trust.md | pair | Proposed conventions authored by Neuronautix, with agent assistance on drafting. |
Reference
Defined by the specification as a combination of who acts and who prompts. Reproduced here so the table above is readable without leaving the page.
| Level | Definition |
|---|---|
| none | Human acts on the task alone with no AI involvement. |
| hint | Human acts on the task and the AI surfaces suggestions passively. |
| assist | Human prompts and the AI acts on a part of the task. |
| pair | Human prompts as both human and AI act on the task equally; the human understands internals clearly. |
| copilot | Human prompts and AI acts on the whole task, prompting the human for permission or clarification. |
| auto | Human prompts and AI acts autonomously, bringing the task to completion. |
Level definitions quoted from the AI-DECLARATION.md specification v0.1.2 by DimwitLabs. The hexagram ䷼ (U+4DFC, “Inner Truth”) is the specification's mark.
Three manifests
fair.md and trust.mdNeuronautix already publishes two repository-level manifests at the root of this site. The AI declaration is the third leg, and the most specific: not how trustworthy is the content, but which parts were written by a machine.
AI-DECLARATION.mdWhich parts of the software and the site were generated, and at what level of autonomy. Portable across repositories and parseable by tooling.
Read AI-DECLARATION.mdtrust.mdHow the knowledge was produced, how each claim is graded on five epistemic categories and a 0–100 confidence scale, and how it is reviewed.
Read trust.mdfair.mdAn honest self-assessment of how Findable, Accessible, Interoperable, and Reusable this repository's knowledge artifacts are — including the gaps.
Read fair.mdScope & limits
It covers the neuronautix.com repository — the website, its knowledge base, its build and QA tooling, and the local RAG prototype. It does not cover client deliverables, consulting outputs, or the separate open-source projects, each of which should carry its own declaration.
It is self-reported and not independently audited. It is reviewed whenever the publishing workflow changes. Last reviewed 2 August 2026, against specification version 0.1.2.