# Welcome Visitors This site is a reflection. It argues one thing: **an AI agent working against a persistent, governed wiki is the natural shape of a self-improving research loop** - and that the protocols I have been building, the Hitchhiker rewilding, and Andrej Karpathy's LLM-Wiki idea are all the *same logic in different skins.*
*For David:* this is a draft I am leaving on the homepage so you can absorb the understanding and reflection, then **fork it into the right subdomain.** Read the four pages below in order; the whole argument is self-contained and interlinked, ready to lift across.
## The one-line claim RAG re-derives knowledge on every query and throws the answer away. An LLM-Wiki *compiles* knowledge once, keeps it current, and **files good answers back as new sources** - so the work compounds instead of resetting. That single move - query output becoming wiki input - is the self-improving loop. We did not theorise it; we ran it on two real corpora.
## Read in order 1. The LLM-Wiki and the Self-Improving Research Loop - the pattern, the three layers, and why agent + wiki self-improves for free. 2. Governing an LLM-Wiki: the Eight Gates - the governance the federation adds (G0-G7) so the loop is trustworthy, not just maintained. 3. The Hitchhiker Expression of the Same Protocol - the same loop wearing the Mouse House / Don't Panic / 42 story. 4. One Logic, Many Skins - the correspondence table: engineering, formal, and mythic registers side by side, plus where City Keys and the value model sharpen it.
## Hitchhiker Briefs Standalone case studies that branch from the loop above - each on its own path, read on its own. The same structure, applied to live events in the wild. - Hitchhiker Brief 001: When Story Became Data - a protection regime rebranded as an engine: Story Protocol becomes DATA Foundation, and what that teaches about IP, provenance, and where value actually accrues.
## For the Vision Fish A single shelf of creative documents to share with the people we invite to pledge time, judgement and a name. The whole two-year story told six ways - briefing, strategy, narrative, saga, the Guide's own future entry, and proverbs - with the timeline figure and every document downloadable. - The Hitchhiker Vision Fish Docs - read whichever voice fits you, then take the files.
## The trust layer The loop above compounds *knowledge*. This thread compounds *trust* - the bilateral agreements a first person proposes (IEEE 7012 / MyTerms), and the agents, devices, and trust communities that carry and witness them. - myTerms: The Agreement Layer - one agreement in three registers (human / legal / machine-readable); why Godel makes the *perfect* agreement impossible, so trust needs a witness from outside; and the consent-first LAN ceremony - around a shared model that names itself - that earns one. - Bonjour by Hand, and the Trust Layer Above - the LAN reachability barrier (a `.local` name that dies on a new router), David's host-side fixes, and why trust is the storey above mere reachability.
## The evidence trail This is filed as a working example of the loop it describes. Building a wiki skill from Karpathy's essay *produced* the eight-gates synthesis - a page that existed in no source, discovered by querying the pattern and filed back. The same loop, run on the workshop transcripts, produced the whole Hitchhiker matrix. Two corpora, one process, three voices. Source idea: Andrej Karpathy, [LLM Wiki](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f). Reflection by Mitch, 2026-06-28.