Most educational programs live in a facilitator's head or a PDF no one reads. When they scale — to new sites, new staff, new AI tools — they degrade.

OLM is a structural layer that makes programs portable, inspectable, and reproducible — without replacing the curriculum.

v1.0 — May 2026 Open framework Built by Meta Humans

What OLM is

OLM is not a curriculum. It is a governed vocabulary and structural layer that any program can be expressed in — making the structure of learning inspectable alongside the content, without changing the content.

Every program expressed in OLM consists of the same kinds of building blocks: routines (what learners do), artifacts (what they produce), evidence (what is observed), constraints (what limits the work), and patterns (named clusters of blocks that recur). A program is a specific, versioned composition of these blocks.

The layer model

An OLM program packet has six layers. Each layer serves a distinct audience and is generated from the layers above it.

Core Mapping
The learning architecture — programs, patterns, routines, artifacts, evidence, constraints. Delivery-agnostic. This is what OLM governs.
Playbook
How the program is typically delivered — pedagogy, group configuration, educator stance, adaptation rules.
Runbook
The executable session flow — timed steps, facilitator actions, participant actions. Internal pipeline stage.
Educator Brief
What educators see — preparation checklist, materials, session flow, key prompts, pitfalls. No OLM IDs.
Parent Brief
What families see — accessible description, what kids will do, why it's valuable.

A hub buyer can inspect the Core Mapping. An educator reads the Educator Brief. A parent reads the Parent Brief. All three documents derive from the same structural source.

The canonical elements

OLM governs a registry of canonical elements — the only valid building blocks for any program mapping. No generated packet may reference an ID not in the registry. This is what makes OLM machine-readable and validator-enforced.

12Patterns
21Routines
19Artifacts
5Evidence types
9Constraints
8HDDs
Routines Artifacts Evidence Constraints Patterns HDDs
noticing_wondering hands_on_activity reflect peer_feedback evidence_synthesis
reflection_log kanban_board scientific_model cad_part_file
question_board prototype_cycle consensus_model capstone_pitch
artifact_presence numeric_match time_limit design_specification curiosity persistence

Browse the full canonical registry →

Reference programs

Four programs mapped end-to-end — selected to show that OLM handles real variety across domain, age range, format, and authorship type.

Venture & Making Authored

Popcorn Factory

Ages 7–17 · Week-long camp

Teams design, build, and pitch a popcorn product venture. The primary reference implementation — shows OLM from the ground up across a full week-long program.

pattern.project_planning pattern.prototype_cycle pattern.capstone_pitch
View mapping →
Culinary Authored

Homemade Pizza

All ages · 60-minute workshop

Learners explore the history of pizza, then make one. Demonstrates OLM at its simplest — a single-session program with one perishable artifact and a clean inquiry-to-application arc.

routine.new_material routine.hands_on_activity artifact.physical_product
View mapping →
Technical / CAD Integrated

CAD Missions

Ages 7+ · Self-paced series

Mission-based 3D modeling using Onshape. Demonstrates OLM on an externally-authored program: one Core Mapping covers the full catalog. Surface-area verification produces the strongest evidence signal in the library.

routine.apply_feature_with_parameters routine.verify_with_measurement evidence.numeric_match
View mapping →
Science Inquiry Integrated — OpenSciEd

OpenSciEd Unit 6.1 — Light & Matter

Ages 11–12 · 6-week unit

OLM as a translation layer over a widely-used NSF-funded curriculum. OpenSciEd's five "routines" become OLM patterns — naming the difference is what makes the translation possible.

pattern.phenomenon_anchor pattern.investigation_cycle pattern.consensus_model
View mapping →

What a translation looks like

OpenSciEd uses "routine" to mean "named instructional move." OLM uses "pattern" for the same idea. Naming the difference is what makes the translation possible.

OpenSciEd constructOLM translation
Anchoring Phenomenon Routinepattern.phenomenon_anchor
Navigation RoutineRitual — Storyline Navigation (no evidence produced)
Investigation Routinepattern.investigation_cycle
Putting the Pieces Together Routinepattern.consensus_model
Driving Question Boardartifact.question_set (existing canonical ID)

OLM and AI systems

OLM is designed to be legible to both humans and AI systems. Most educational frameworks are too narrative or loosely defined to constrain an LLM. OLM's governed vocabulary and explicit layer rules eliminate the most common failure modes in AI-generated educational content: activities masquerading as programs, skills asserted without derivation, evidence claims without artifacts.

When a model operates within OLM, it cannot hallucinate structural relationships. The constraints become guardrails. This makes OLM a shared semantic layer between humans and AI tools working on the same content.

System prompt
Drop into Claude, ChatGPT, or any capable LLM to get OLM-aligned generation.
Download olm_system_prompt_latest.md
Context bundle
Constitution, canonical registry, generation rules, and worked examples — everything an LLM needs in one fetch.
Download olm_context_bundle_latest.md
Canonical registry
Machine-readable YAML. The authoritative source for ID validation — for humans, AI agents, and automated scripts.
Download canonical_registry_latest.yaml

All AI resources, usage guide, and changelog →

Contribute

OLM grows through practitioners mapping the programs they actually run. The most valuable contribution is a Core Mapping — a structural description of a real program expressed in canonical OLM vocabulary.

Apply OLM to your program

Paste your brief below to build a complete prompt — the canonical registry, layer rules, and your brief — ready to paste into Claude, ChatGPT, or any capable LLM. Generate a Core Mapping for the learning architecture, or a Playbook for the delivery model. Your AI tool, your output, framework-compliant.

Propose a canonical element

If your program uses a routine, artifact, or evidence type with no existing canonical equivalent, propose it. Requires framework fluency and a justification for reusability.

Flag an error

Open a GitHub issue. Lowest barrier, still valuable. Structural feedback helps the registry stay accurate.

Build a prompt from your brief

Paste a program brief and get a complete prompt ready to drop into Claude, ChatGPT, Gemini, or any capable LLM. Choose OLM Mapping for a YAML Core Mapping (the learning architecture), or OLM Playbook for a YAML Playbook (the delivery model). Both are valid contributions to the OLM library. Nothing leaves your browser.

Community

OLM educators and hub operators discuss programs and framework questions in the Meta Humans community forum under the OLM category. Technical contributors work directly through the GitHub repository.