The World-factory · documentation as code

Pick a card.

Every World begins as a measurable goal in ARAIL. The goal implies a subject; agents research it and run experiments while a provenance gate checks every source, and a curated, fully-sourced app comes out the other side, a dictionary, an on-box docent, and an open API. Same framework, any domain.

11  Worlds941  sourced termsopen API at /dac

How the pieces fit

DaCthe knowledge engineWorldsthe stage for the lab + researchKnowledge basesourced · declarative · kept freshmay or may not →Personal LLMbaked by Nucleus — if it earns it

DaC is the knowledge engine. It researched, gated, and compiled the Worlds above. Each World sets the stage for the lab and the research that run on top of it — and the sourced knowledge base they produce may, or may not, get baked into a personal LLM you own.

I
I

AI / ML

editorial
331terms
14 categories

The original. The whole discipline, from attention and transformers to RAG and agent memory.

Enter the World →

II
II

Astronomy

immersive · cosmic
101terms
8 categories

The observable universe, defined. Parsec to the cosmic microwave background, with the stars out at full tilt.

Enter the World →

III
III

Mathematics

immersive · formulae
124terms
10 categories

The famous formulas and the most beautiful results, Euler's identity to Gödel. The blackboard never stops writing.

Enter the World →

IVψ
IVψ

Quantum

immersive · quantum
101terms
8 categories

The physics of the very small, told honestly. Superposition, entanglement, qubits, the myths stripped out.

Enter the World →

V
V

Biology

immersive · bio
103terms
8 categories

The science of life, from the double helix and the cell to evolution and the immune system, myths corrected.

Enter the World →

VI
VI

Ecology & Ecosystems

editorial
51terms
7 categories

Trophic cascades, regime shifts, and metapopulations — ecosystem-level ecology told honestly, pop-myths corrected.

Enter the World →

VII
VII

World Models

editorial
34terms
6 categories

Bayesian surprise vs. prediction-error curiosity vs. intrinsic motivation — precision vocabulary for wiring a curious, world-modeling agent. WorldLLM is the flagship method.

Enter the World →

VIII
VIII

Neural Networks

editorial
34terms
6 categories

The substrate everything modern rests on — from the perceptron and backpropagation to transformers, residual connections, and the failure modes every deep learning practitioner hits.

Enter the World →

🎀
🎀

Hello Kitty World

editorial
32terms
4 categories

A world for fun. The first domain of Charlotte's World — a fan world run on kindness, cleverness, and a little moonlight: talk past Kuromi by day, race the Glooms by night, befriend the ghosts to escape the manor. Play her games straight from the deck.

Enter the World →

🧸
🧸

The Squishy World

editorial
9terms
4 categories

Newly dreamed. The second domain of Charlotte's World — a cozy, pastel realm of plush, squishable friends that runs on softness, comfort, and a warm hug. Her own original soft creatures, growing as she dreams them.

Enter the World →

·
·

Nucleus Pipeline

editorial
21terms
3 categories

The seven stages that turn a teacher model and a gate-passed World into an owned, sealed specialist — from ingestion and 7-layer extraction through the adversarial swarm to the cryptographic seal, each stage a sourced term with its own technical deep dive.

Enter the World →

A World is where development begins

Not the finished product — the ground you build on.

A World lays down the definitions and the base structure of what's to be: the shared, fully-sourced vocabulary you and your agents develop from. You don't start from a blank page — you start from a settled foundation, then build depth on top of it.

Building a product around plants? Enter a focused World like Botany or Horticulture for a tight, specific base — or a higher-level World like Biology to develop with more depth and surrounding context. You pick the altitude; the World sets the ground you and your agents build on.

How a goal becomes a World

01

A measurable goal

Set in ARAIL — improve or learn something you can measure.

02

A theme

The goal implies a subject. Scope, boundaries, sources.

03

Agent swarm

Experiments to improve it; the corpus drafted alongside.

04

Provenance gate

Every entry must cite a real source.

05

A World

A dictionary + docent + open API.

Seed: agent-curated, each entry cites a source, adversarially verified before it ships.

Who builds the World · PaperAgents

today: Claude · endgame: PaperAgents

Point PaperAgents at a theme. The curation becomes repeatable.

Same DaC framework, pointed at any subject. Each is an open, fast API over a curated, fully-sourced corpus, built by an agent swarm that researches the subject and writes the World around it. That swarm has a shape: PaperAgents is a configuration-first orchestrator, a team of agents declared in one TOML file that owns its work, reconciles its own drift, and carries its own knowledge base. DaC says what a good World is; PaperAgents is who builds it. Today that swarm is farmed out to Claude; the endgame declares it and runs it on our own orchestrator, our own sauce, all the way down.

idea qukaizen.com/paperagents·truth github.com/cdarnell/paperagents

A World is a manifest

Declarative TOML team

One file declares the research agents, the corpus they own, and the budget. The build that stood up five Worlds becomes a single team you re-point at the next theme.

Repeatable by construction

Idempotent apply

Point it at a new subject, apply, and the same pipeline stands up another World, no bespoke run, no drift between what you declared and what shipped.

The DaC refresh loop

Watcher + reconciliation

Desired-state is the field as it stands today; the World is observed-state. The watcher surfaces the gap and reconciles it, re-gather, re-gate, recompile, on a schedule instead of by hand.

The corpus is the memory

Embedded per-team RAG

A team carries its own sourced knowledge base. For a World that knowledge base is the product, the curated, fully-sourced corpus served at /dac.

world-ai.tomlillustrative
# world-ai.toml — a Knowledge World as a PaperAgents team (illustrative)
# Modern array schema: [[teams]] / [[agents]] / [[knowledge]]

[[teams]]
[teams.metadata]
name      = "world-ai"
namespace = "dac"
[teams.metadata.labels]
world       = "ai"
displayName = "AI"

[teams.spec]
description = "The AI Knowledge World — sourced, compiled, served at /dac"
[teams.spec.watch]
enabled         = true
intervalSeconds = 86400   # re-gather, re-gate, recompile nightly

[[agents]]
[agents.metadata]
name      = "ai-researcher"
namespace = "dac"
[agents.spec]
teamRef          = "dac/world-ai"
role             = "analyst"      # research/drafting — requires hybrid
modelTier        = "hybrid"
responsibilities = ["Draft AI-domain term entries from the cited canon"]

[[agents]]
[agents.metadata]
name      = "ai-provenance"
namespace = "dac"
[agents.spec]
teamRef          = "dac/world-ai"
role             = "compliance"   # adversarial source gate — requires frontier
modelTier        = "frontier"
responsibilities = ["Verify every claim and source; reject pop-AI myths"]

[[agents]]
[agents.metadata]
name      = "ai-compiler"
namespace = "dac"
[agents.spec]
teamRef          = "dac/world-ai"
role             = "executor"     # deterministic assembly — requires sml-executor
modelTier        = "sml-executor"
responsibilities = ["Assemble validated shards into terms.json + WorldSpec"]

[[knowledge]]
[knowledge.metadata]
name      = "ai-corpus"
namespace = "dac"
[knowledge.spec]
teamRef = "dac/world-ai"
sources = ["./public/worlds/ai/terms.json", "Goodfellow et al. — Deep Learning",
           "Russell & Norvig — Artificial Intelligence: A Modern Approach"]

ARAIL prototypes · PaperAgents produces · DaC serves · Nucleus bakes.

Why DaC — and what it's good for

DaC is the knowledge engine behind the deck — it researched every World, gated every source, and compiled the result. A dictionary is just the most legible shape of that compiled, sourced, declarative knowledge base. The Worlds above prove the pipeline. The same flow — define a theme, let agents draft, gate every source, compile — is built to serve far more than dictionaries. Each card is another shape of the same output.

live

Knowledge dictionaries

Six live today: AI, Astronomy, Mathematics, Quantum, Biology, Ecology — each sourced and browsable.

e.g.  try /what?world=math&term=quadratic-formula

roadmap

Living research digests

Agents re-read the field overnight, the gate re-checks, the World recompiles. Yesterday's papers, today.

e.g.  nightly: re-gather → re-gate → recompile

roadmap

Runbooks & SOPs

Each business function's operating manual as a declarative World your PaperAgents run from.

e.g.  ask “how do we close the month?” → the sourced runbook

roadmap

Docs, onboarding & courses

API docs, product guides, curricula. Learn is the human projection of the same corpus.

e.g.  one corpus → API docs, a course, day-one onboarding

roadmap

Compliance & sourced KBs

Auditable knowledge where every claim cites its origin, by construction.

e.g.  every answer ships with its citation — audit-ready

roadmap

Agent operating knowledge

Buddy's brain as a versioned World, the truth your agents reason from.

e.g.  your agents reason from a versioned, sourced World

The refresh loop

A World is never finished, it is kept. On a schedule, agents re-gather the raw material, the provenance gate re-checks it, and the World recompiles with the latest findings folded in. When the compiled truth is good enough, Nucleus bakes it into a model you own; the settled notes drop out of the active retrieval set (the sealed source is kept), and the window slides forward on the same subject. That is the circle below.

Every World feeds the circle · RAG to MEMORY

A World is a compiled source of truth — the stage the lab and the research run on top of. Serve it live as RAG, or, when it earns it, bake it into a personal model you own with Nucleus: the settled knowledge becomes memory and drops out of the active retrieval set, while the sealed source is kept for provenance and re-bake. The window slides — same subject, kept current. Not every World needs to be baked — many are most useful served live.

One open, fast API

curl qukaizen.com
$ curl qukaizen.com/dac          # the framework manifest: worlds + the API
$ curl "qukaizen.com/what?world=math&term=quadratic-formula"
$ curl "qukaizen.com/story?world=astronomy&term=parsec"
$ curl qukaizen.com/health       # liveness + per-World term counts
$ curl qukaizen.com/metrics      # Prometheus gauges
Dictionary →