CLAUDE

Umwelten

What this project is

Umwelten is a personal, public-facing encyclopedia of neurodiversity — particularly sensory processing, autism, ADHD, and intellectual disability. It is maintained by Andrew Hopper and built with transparent AI assistance.

The name comes from the German biologist Jakob von Uexküll, who coined Umwelt to describe the perceptual world unique to each organism. Umwelten is the German plural, because this wiki is about multiple perceptual worlds, not just one. A neurodivergent umwelt is more than different sensory thresholds: it encompasses a different relationship to one’s own body signals, a different social world, a different temporal experience, a different way attention moves and settles. This wiki is a guide to those different worlds.

The wiki aims to be what I couldn’t find when I started working in this space: a single, structured, evidence-based resource that takes the neurodivergent perspective seriously, is honest about what the evidence does and doesn’t show, and is useful to autistic people, their families, carers, teachers, researchers, and anyone else who wants to understand neurodivergent experience.

Some foundational content — particularly the Dutch sensory processing frameworks (prikkelbalans, prikkeltaal, prikkelprofiel) and research on sensory processing in autism with intellectual disability — draws on a four-year participatory action research project called De Sensatie van een Goed Leven (The Sensation of a Good Life) at the University Medical Centre Groningen (UMCG). The wiki has grown well beyond those origins.

Philosophy

Why this exists

Good knowledge about neurodiversity is scattered: in people’s heads, in academic papers behind paywalls, in Dutch PDFs on a website, in the lived experience of autistic people and those who support them, in clinical practices that don’t always reflect the evidence. This wiki aims to be a structured, searchable, cross-linked resource that brings scattered knowledge together and makes it genuinely useful.

Principles

1. Neurodivergent perspective first. This is not a clinical information system. The wiki speaks from the neurodivergent viewpoint, not about neurodivergent people as objects of study. Identity-first language is preferred internationally (“autistic person” rather than “person with autism”), though the Dutch materials historically use person-first (“mensen met autisme”). The wiki uses identity-first as default while acknowledging individual preference.

2. Evidence-transparent. Every claim should be traceable to its source and its level of evidence. The wiki distinguishes clearly between peer-reviewed research, practitioner consensus, emerging patterns, and single observations. Where the evidence is weak, the wiki says so. Where it is contested, the wiki presents the contest honestly. The wiki never overstates what is known.

3. Human-curated, AI-assisted. AI (primarily Claude) helps with research, drafting, connection-finding, and quality maintenance. A human reviews and approves everything. The AI proposes; the human disposes. This process is transparent — the wiki does not pretend to be purely human-written. See the About page for details.

4. No deficit framing. Neurological differences are differences, not deficits. Challenges arise from the mismatch between the person and their environment, not from the person being broken. This is not a philosophical nicety — it is the position best supported by the evidence on masking, mental health outcomes, and the social model of disability.

5. Knowledge, not cases. The wiki never contains pages about identifiable individuals as cases or subjects. Experiential knowledge is abstracted into patterns. Privacy is non-negotiable — particularly when discussing vulnerable populations such as people with intellectual disabilities.

6. Cross-linked by design. Entries are richly linked to each other. A page on auditory processing connects to relevant research summaries, practical strategies, assessment tools, key figures, and related sensory domains. The linking taxonomy is the connective tissue that makes the wiki more than a collection of articles.

7. Honest about uncertainty. Neurodiversity research has significant evidence gaps. The wiki documents these gaps explicitly rather than papering over them. An evidence gap page is as valuable as a topic overview — sometimes more so.

Scope

The wiki covers neurodivergent experience broadly: autism, ADHD, their co-occurrence, sensory processing, social cognition, executive function, burnout, masking, and the science, practice, and contested territory around all of these. It includes the intersection of neurodiversity with AI, policy, and society. Sensory processing in autism and intellectual disability remains the area with the deepest content and the strongest editorial perspective, but the scope extends to the full neurodivergent umwelt — internal states, social worlds, cognitive styles, and the environmental mismatches that create difficulty.

The wiki does not attempt to be encyclopaedic about all of neurodivergence. It goes deep where it has expertise and is transparent about its boundaries.

Architecture

Platform

The wiki is built in Obsidian — a local-first, markdown-based knowledge management tool — and published as a static website via Quartz on Cloudflare Pages at umwelten.wiki. The source vault lives on a Mac Mini M4 Pro (48GB). Deployment is via a local build script (deploy.sh) using Wrangler CLI.

Folder structure

wiki/
├── CLAUDE.md                    # This file
├── 00-inbox/                    # Raw inputs awaiting processing
│   ├── voice-notes/
│   ├── meeting-notes/
│   ├── forwarded/
│   └── observations/
├── 01-science/                  # Science and research
│   ├── topic-overviews/         # Synthesised overviews of key topics
│   ├── paper-summaries/         # Summaries of individual research papers
│   ├── evidence-gaps/           # Explicitly documented gaps in knowledge
│   └── glossary/                # Defined terminology
├── 02-practice/                 # Practice and experience
│   ├── observed-patterns/       # Patterns distilled from multiple observations
│   ├── strategies/              # Approaches and strategies that work
│   ├── environment-design/      # Physical and sensory environment guidance
│   └── context-notes/           # Setting-specific knowledge
├── 03-methods/                  # Methods and tools
│   ├── assessment-tools/        # Instruments for assessing sensory processing
│   ├── prikkeltaal/             # The Dutch stimulus language framework
│   ├── interventions/           # Intervention approaches and frameworks
│   └── training-guides/         # Guides and training materials
├── 04-context/                  # Context and history
│   ├── case-studies/            # Notable projects and their lessons
│   └── reference/               # Background material
├── 05-landscape/                # People, organisations, and ideas
│   ├── peer-organisations/      # Organisations doing related work
│   ├── policy-landscape/        # Relevant policy frameworks
│   ├── funding-landscape/       # Funding bodies and opportunities
│   └── key-figures/             # Important researchers, thinkers, advocates
├── 06-comms/                    # Drafts and working material
│   └── drafts/
└── templates/                   # Page templates for each content type

Tagging taxonomy

Every page gets tagged with metadata in its YAML frontmatter.

Content type (what kind of knowledge):

• type/research — peer-reviewed or formally published
• type/experiential — derived from practice and lived experience
• type/methodological — tools, instruments, approaches
• type/landscape — people, organisations, ideas, policy
• type/meta — about the wiki itself (about page, methodology)

Sensory domain (which senses):

• sensory/auditory
• sensory/visual
• sensory/tactile
• sensory/vestibular
• sensory/proprioceptive
• sensory/olfactory-gustatory
• sensory/interoceptive
• sensory/multi-modal

Setting (where the knowledge applies):

• setting/home
• setting/classroom
• setting/care-facility
• setting/workplace
• setting/public-spaces
• setting/transitions

Age group:

• age/early-childhood (0–6)
• age/school-age (6–18)
• age/adult (18+)
• age/all

Evidence level (how well-supported):

• evidence/peer-reviewed — published research
• evidence/practitioner-consensus — widely agreed among professionals
• evidence/emerging-pattern — observed by multiple people, not yet formally studied
• evidence/single-observation — one person’s report, valuable but unconfirmed
• evidence/anecdotal — informal, context-dependent

Status:

• status/draft — in progress, not yet reviewed
• status/review — awaiting human review
• status/published — approved and integrated
• status/outdated — flagged for update
• status/archived — superseded but retained for history

Language:

• lang/nl
• lang/en
• lang/bilingual

Inbox processing workflow

The 00-inbox/ folder is where raw knowledge arrives. The AI-assisted processing workflow:

1. Arrival. A new file lands in inbox — a transcribed voice note, an observation, a meeting note, a forwarded document, a research paper.

2. Triage. Claude reads the input and proposes: which category it belongs to, what tags apply, whether it relates to an existing page or warrants a new one, and a draft structured entry if it’s new content.

3. Review. I review, edit, approve, or reject the proposal.

4. Integration. Approved entries move to their proper location. Links to related pages are added. The page goes to status/published.

5. Connection scan. Periodically, Claude scans for: new connections between pages, patterns across recent entries, pages that may be outdated, and gaps in coverage.

Related projects

This wiki sits alongside a planned companion project:

Encyclopedia of neurodivergent experience — a separate, distinct corpus. Selected individuals are invited to have open-ended conversations with an AI interlocutor about their sensory and neurodivergent experiences. Contributions are synthesised into a collaboratively authored taxonomy of experience — not traceable to individuals, a genuine work of collective intelligence. The encyclopedia is separate from this wiki, but bridges exist: patterns identified there can inform wiki pages, and wiki knowledge can contextualise encyclopedia themes.

Technical notes

• All files are markdown with YAML frontmatter
• Never start the body with a # Title heading that duplicates the frontmatter title: field. Quartz renders the frontmatter title as the article’s H1 via the page-header ArticleTitle component; any # Title line at the top of the body becomes a duplicate title on the rendered page (and again in hover popovers, search previews, and OG cards). The body should open with either prose or the first sub-heading (usually ## Summary or ## Overview). This applies to every new page, every Cowork-generated page, and every regenerated page. If an existing page has a duplicate H1, strip it — do not leave it in place.
• Internal wikilinks must use the full vault path from root, e.g. [[01-science/topic-overviews/autistic-burnout|Autistic burnout]], not the short form [[autistic-burnout]]. Quartz’s shortest link-resolution mode has a fallback branch that produces broken links from nested pages when the short form is used; full-path wikilinks resolve deterministically.
• Frontmatter dates use created: and updated: (not modified: — that’s Quartz’s native field name, but this vault uses the English vocabulary and a custom emitter reads updated: directly).
• Internal links use Obsidian’s [[wikilink]] syntax
• Tags use a hierarchical format: sensory/auditory, type/research, etc.
• The vault is designed to be queried by an LLM via Claude Code or a local model
• Privacy is paramount: no personally identifiable information about service users, clients, or research subjects
• The wiki currently has one curator (me). The system is designed so that a second curator could be onboarded by reading this document and the templates

Style and voice

• Use British English spellings throughout
• Prefer clear, direct language over academic jargon — but don’t shy away from precise terminology when it matters
• Write for an intelligent non-specialist audience unless a page is explicitly tagged as technical
• The tone is warm, respectful, and grounded — speaking from the neurodivergent perspective, not about neurodivergent people as objects of study
• Avoid deficit framing: neurological differences are differences, not deficits. Challenges arise from the mismatch between the person and their environment, not from the person being broken
• Be honest about contested areas. Present the best case for each position, then state what the evidence supports. The wiki has a perspective (neurodiversity-affirming, evidence-transparent) but it is not propaganda
• When the wiki takes a position, own it in first person. “This wiki argues that…” or “The position here is…” — not hiding behind passive voice or false objectivity

Writing craft: avoiding AI tells

The wiki is transparent about AI assistance, but the writing should never look AI-generated. Avoid these common LLM patterns:

• Em-dash overuse. Use sparingly. Prefer commas, colons, semicolons, or restructured sentences.
• Hedging formulations. Remove “it is worth noting that,” “it is important to understand,” “essentially,” “fundamentally.” If it’s worth noting, just note it.
• Performative emphasis. Remove “this matters enormously,” “this is critical,” “this cannot be overstated.” Let content carry its own weight.
• Over-signposting. Don’t announce what’s coming. Just say it.
• Triple-structure lists. Vary rhythm. Sometimes two items. Sometimes four. Sometimes one example explored in depth.
• Formulaic transitions. Don’t start consecutive paragraphs with “However,” “Additionally,” “Furthermore.”
• “Not X but Y” overuse. Keep the strongest contrasts, rewrite the rest as direct statements.
• Repetitive paragraph openings. Vary sentence structure across consecutive paragraphs.

The goal is confident, natural prose that reads like it was written by someone who knows the subject, happens to have had AI research assistance, and edited carefully.

AI transparency

This wiki is built with substantial AI assistance. Claude (Anthropic) is used for:

• Research and literature synthesis
• First-draft writing of wiki pages
• Connection-finding between topics
• Gap analysis and quality maintenance
• Inbox triage and content structuring

Every page is reviewed, edited, and approved by a human before publication. The AI does not publish autonomously. This process is described on the wiki’s About page, because intellectual honesty about how knowledge is produced matters — especially in a project about minds that work differently.