Skip to content
Cognitive
Orchestration Spec of record ↗

Domain · Build your own

Derive CO for your world.

Codegen, research, education, governance — the named domains prove one thing: the methodology travels. But the list is a template, not a limit. If AI does real work in your field, you can derive a Cognitive Orchestration application for it. Here's how — and how to make it part of the family.

§ · First, the honest test

Does CO even fit your domain?

CO earns its place when AI fails your work in three specific ways — amnesia, convention drift, and safety blindness. So the first step is a test, not a build: try to name all three in your domain.

If you can, CO fits, and the rest of this page is your blueprint. If you can't find all three, CO may not be the right tool — and that's a real answer worth having before you invest. We'd rather you know now.

§ · Step 1 — Analyze your domain

Five questions that surface the five layers.

  1. 01

    What do your experts know that novices don't?

    The institutional knowledge — the conventions, judgment, and hard-won lessons that separate a practitioner from a beginner. This is what Layer 2 will hold.

  2. 02

    What goes wrong when AI lacks that context?

    Your domain's specific failure modes. If you can name all three (below), CO fits. This shapes your guardrails.

  3. 03

    What rules are non-negotiable?

    The things that must never break — the compliance step, the safety check, the boundary. These become deterministic guardrails.

  4. 04

    What is the standard workflow?

    The phases work moves through, and where a human must sign off. This becomes Layer 4.

  5. 05

    What patterns emerge from repeated work?

    What recurs often enough to capture and reuse. This is what Layer 5 learns and compounds.

§ · Step 2 — Map the three failure modes

Write yours down.

  1. F1

    Amnesia

    How does your domain's context get lost as a session grows? Which instructions quietly revert to generic defaults?

  2. F2

    Convention drift

    Where do your practices diverge from the textbook? What does an AI do “by the book” that your organization does differently?

  3. F3

    Safety blindness

    What compliance, safety, or regulatory step does the AI skip because it isn't the most direct path to a result?

§ · Step 3 — Design the five layers

Build one layer at a time.

  1. 1 · Intent Name your specialists

    Define at least two domain-specialized configurations and the rules that route work to them. Who should hold which task?

  2. 2 · Context Write down what your org knows

    A master directive loaded every session, plus a knowledge hierarchy — quick indexes, topic files, deep reference pulled on demand.

  3. 3 · Guardrails Make the non-negotiables enforceable

    Soft rules the AI interprets, and hard checks that run outside the model and block a violation before it lands.

  4. 4 · Instructions Define the phases and the gates

    A structured workflow the process can't skip, with human authority at the points that matter, and evidence required to call something done.

  5. 5 · Learning Close the loop

    Observe what works, capture the patterns, and promote the best into durable artifacts — on human approval, so knowledge compounds.

§ · Then — contribute it back

New domains are born in the open.

A domain you derive doesn't stay yours alone. Propose it in the open — a public sketch, reviewed by the community — and a strong one joins the family as an official “CO for X.” This is how the methodology grows: not decreed from the center, but extended by the people who practice it. Name it (CO for [Domain], short-name CO[X]), say who it's for, and share it.

Read the reference Join the movement

Derivation methodology per the CO Domain Application Template (Terrene Foundation, CC BY 4.0). Propose new domains via the Terrene Foundation; specification of record at terrene.foundation.