Education has countless routes—grades, courses, textbooks, tests, and apps—but no widely shared street map of the knowledge and capabilities beneath them.
The project takes the form of an OpenStreetMap-like base layer for education. Stable records distinguish concepts, knowledge, representations, practices, observable capabilities, and composites; typed relationships carry direction, rationale, confidence, provenance, lifecycle, and review state. Curricula, projects, assessments, learner records, mastery algorithms, and apps remain downstream consumers rather than being baked into the terrain. Its purpose is to give people, institutions, software, and AI agents a common map of what can be learned and how the pieces relate.
The question behind Open Learning Atlas
Open Learning Atlas begins with the terrain. It treats a concept, a representation, an observable capability, and a prerequisite claim as different kinds of public records with stable identities, provenance, confidence, lifecycle, and review state. A curriculum can choose a route through that terrain without becoming the terrain itself. Learners may benefit indirectly through better routes and diagnostics; educators, researchers, curriculum designers, institutions, software teams, and AI agents can use the map directly. A weak or falsely authoritative map could misdirect learning and assessment, so uncertainty and accountable review are part of the infrastructure.
How Open Learning Atlas took shape
The first public domain is mathematics. The alpha grew through a constitution, typed schema, source records, deterministic compilers and validators, versioned release packages, agent query fixtures, a human review ledger, and a mobile-first atlas. AI agents materially accelerated the build, while every generated mathematical record remained explicitly candidate pending accountable review. Josiah originated the open educational-map direction, the OpenStreetMap analogy, the separation between canonical terrain and downstream curricula or apps, and the requirement that people and AI agents share the same inspectable architecture. He directed an AI-assisted build of the ontology, governance, tooling, releases, and public interface and required candidate status and explicit review limits instead of presenting generated breadth as consensus.
Architecture diagram · open educational terrain
A shared map separates mathematical terrain from the routes built across it.
Curricula, projects, assessments, learner state, mastery models, and apps remain downstream overlays.
What the evidence supports
The repository, release packages, validators, conformance fixtures, review queue, and live atlas are public. All 87 territories pass structural floors, but a release-pinned adversarial audit found four P0 decomposition problems and 28 P1 specialist-review priorities; all 87 human territory reviews remain pending. The architecture is now real enough to expose its own weakness. Structural checks pass, but the map is not yet a trustworthy prerequisite graph: an adversarial audit found concrete omissions, broad bundled territories, thin specialist evidence, and mostly unreviewed dependencies. Publishing those problems is part of building open infrastructure rather than a reason to hide the alpha.
This is a candidate alpha, not a validated standard or a learner-ready curriculum. Only four active relationships currently assert prerequisite necessity, 511 of 519 assessable records have unreviewed dependencies, and no territory has accountable human approval. The map's breadth, granularity, evidence criteria, and dependency corridors require specialist review. The next stage is accountable specialist validation: resolve the four P0 findings, review the 28 P1 territories, complete all eight review dimensions for every territory, and build scoped dependency corridors before Mathematics v1.0.