The domain wiki

The Knowledge a Market Runs On

The engine is generic; the knowledge is not. What a market knows is kept as a wiki that its sponsor writes and keeps. It is easy to start, it is never finished, and everything built afterwards rests on it.

More than one way through

Most tasks a participant must get through have more than one way through them, and the people a thin market fails are often the ones who know only the obvious way. A worker with years in a trade and no certificate is told a certificate is required, when the law may let an employer hire on demonstrated skill.

A requirement written as fixed refuses those people silently. Nobody complains, and nothing records who was turned away. One sourced page in the wiki turns the refusal into a route.

WITHOUT THE PAGE WITH ONE SOURCED PAGE 55 workers skilled, no certificate Certificate required Employers 55 refused, silently 55 workers skilled, no certificate Certificate required Employers 45 still refused One wiki page, with its source a lawful route: hire on demonstrated skill 10 reached The requirement is written as fixed, so nobody sees who it turned away.
A trade where the law lets an employer hire on demonstrated skill. The counts come from a twin whose participants are synthetic.

So the wiki records, for each task, the ways through it: who grants each, what it confers, and what it still requires. Two rules bind every entry. The wiki says that a path exists, never that a person qualifies for it. And no path is recorded whose granting authority cannot be named.

What a domain wiki is

Plain-text pages, each tied to its sources, drafted with an AI model and ruled on by a person. It holds what only people in the market know: what a certificate means in one trade, which requirement has a lawful alternative, how the two sides describe the same thing in different words.

It is easy to start. An LLM does the drafting, an open wiki engine supplies the tools, and DeeperPoint’s guides serve as the starter kit. It is never finished. It keeps improving as the market puts questions to it.

How it is built and kept

  1. Gather. Research is compiled into notes, each fact tied to the source it came from.
  2. Ingest. An AI model drafts typed pages from the notes and links them to one another. A person rules on what stands.
  3. Check. A validator runs on every change, and a review pass looks for contradictions, duplicates and pages nothing links to.
  4. Gate. Every page is marked as shareable or not. The export stops if a shared page would carry something that must not leave.
  5. Publish. The shareable pages go out as one fixed version that people, and a twin, can read.

Then the cycle repeats, because the market changes and so do the questions.

What it feeds

A twin is the marketplace engine configured for one market and filled with synthetic, marked participants. The wiki supplies four things to it.

THE WIKI Sourced pages checked, and gated for what may leave The sponsor keeps it a fixed, recorded version THE TWIN Reference library answers cite the wiki page Schema values each tagged by where it came from Alternative routes ways around a hard requirement Synthetic population generated from the wiki’s figures THE MARKETPLACE Real participants The wiki stays, and is kept current A sponsor runs it synthetic profiles never cross
The knowledge carries from the twin into the marketplace. The synthetic participants do not.
  • The reference library. CommonContext splits the shared pages, embeds them and prepares them for the engine to load, so an answer can cite the page it came from.
  • The backing for schema values. Each value in the market’s schema is tagged by where it came from: stated in a wiki page, inferred from one by a written rule, or invented for the demonstration. In the twins built so far the schema was written by hand, and the wiki backs and checks it.
  • Alternative routes. A hard requirement must name its approved alternatives, and those come from sourced pages.
  • The facts behind the synthetic population. The participants are invented; the figures they are drawn from are the wiki’s.

Why a twin reads a fixed version

A wiki changes whenever its keeper learns something. A twin is meant to read one named version of it, record which, and move to a newer one deliberately. Then an answer can always be traced to the page as it stood, and a change in the wiki cannot quietly change what the market does.

Asking the wiki

A question page can be put over the wiki: answers made only from its shared pages, with citations, and a plain statement where the wiki does not hold the answer. It is an early check on whether the wiki holds anything of value, before any twin exists, and what it cannot answer becomes the next research. See the whole path →

Who keeps it

The sponsor decides who maintains the wiki. DeeperPoint helps start it; keeping it current is the sponsor’s standing job, and blocked matches and unanswered questions show where a page is missing. The engines are shared openly. A market’s wiki is not: it is the sponsor’s own.

Where to start

The sponsor’s guide covers what to gather, how to judge it and how to know when you are ready. The wiki-to-twin recipe is the build procedure, stage by stage. The wiki engine is open source under MIT. What a build costs gives the hours and the cash that a wiki and a twin take.

MarketForge: the map and the method → What we’re building →