Let us demonstrate what QuKaiZen can do.
Four instruments, one corpus. Each panel below routes over the taxonomy artifact or calls the public API this site serves, weighs the responses in your browser, and runs the same gate law the build runs — 1,176 sourced terms across 15 Worlds today, and the counter climbs.
Everything routes. Watch it decide.
If you're sending data, it has to land in a World — a capture with nowhere to land is just a chat log. This panel fetches the engine's own taxonomy artifact (every parent–child edge cites a classification authority) and routes a named subject in this tab: the same deterministic, single-world routing the artifact's own status line declares. Name a subject; watch the walk.
What this panel deliberately does not do: recognize a raw capture — a photo, a voice note — as a subject. That recognition, the multi-world fan-out, and the live capture stream arriving here are the ahead half of the router.
→ we aim to· the llm-router + the wired capture streamAsk the dictionary. Watch the wire.
Pick a World and ask for a term. You get both answers: the card a person reads and the JSON a machine gets. Then press “explain another way” — the AI Dictionary's own button — and a small model on the project's own box rephrases the entry, streamed as it writes. Every request line is copyable; the same calls run from any terminal.
Fetch the fact. Skip the haystack.
An agent that mounts a World looks facts up instead of carrying the whole corpus in its context window. Below: the full Ecology corpus and a single-term slice, both fetched and weighed in this tab just now. The ratio comes from those two measurements.
the whole corpus
fetching…
one term's slice
— not yet weighed
GET /what?term=trophic-cascade&world=ecology · the exact response body
the ratio
Weigh a slice and the ratio renders from the two live figures.
the docent reads the slice
A small model on the project's own box answers from the entry it just looked up — the slugs it grounded on ride along in a response header.
Break the corpus. The gate reads it back.
This page imports the same assertClosedSourcedGraph the build runs and points it at your in-tab copy of the Ecology corpus. Erase a citation, invent a category, remove a term that others cite — the verdict re-renders on every edit. Reset restores the corpus exactly as fetched.
your in-tab copy · waiting for the corpus
waiting for the corpus…
The function above is vendored from the DDaC engine with a provenance stamp in its header; npm run check re-hashes the file body against that stamp on every build. The same law gates every World before it ships.
the shipped bytes, verified
Your edits above lived in this tab's copy alone. Below: SHA-256 over the exact bytes this tab fetched, beside the hash committed with the corpus and the hash in the build receipt.
That's serve, the middle of the loop: sourced terms fetched live over an open API, weighed as lookups instead of context, and gated by a law you just ran yourself. Part 2 continues from here — the same World, baked into a model and run without a full GPU.