The Federated Knowledge Graph

Dependencies, not sequences

A knowledge graph is a structured map of concepts and their relationships — what depends on what, what enables what. It is the system's understanding of the structure of knowledge itself.

The crucial distinction: L5 models dependencies, not sequences. "Fractions require basic arithmetic" is a dependency — it belongs in L5. "First fractions, then decimals" is a didactic decision — it belongs in L3. If the graph prescribed sequences, it would become the curriculum. The architecture refuses that role for L5 deliberately.

Two governance layers

The graph has a two-layer architecture. The first layer is the universal STEM core — globally shared, versioned, maintained by independent rotating expert panels. Every change is recorded in an open ledger.

The second layer is the sovereignly curated cultural layer — history, society, values, ethics — maintained by legitimated curatorial authorities. A national education ministry opens the governance cockpit, sees the universal graph, marks where its narrative diverges, and sees the consequences. The system stores both perspectives as attributes, not errors.

When the layers disagree — for instance, when a cultural authority contests a dependency the STEM panel considers universal — the resolution principle is explicit: Layer 1 governs empirically falsifiable claims (the boiling point of water, the prerequisites of calculus); Layer 2 governs interpretation, narrative and values. Boundary disputes are escalated to a standing boundary committee with rotating membership drawn from both panels, whose rulings are published with reasoning and are revisable.

What lives inside the graph

Within these governance layers, the graph operates with richer internal structure. A concept layer maps abstract knowledge and dependencies. A manifestation layer maps the same concept across domains — "ratios" as fractions in mathematics, rhythm intervals in music, recipe scaling in cooking — enabling the Mentor's polymorphic delivery. A competency layer connects to L6: what one must be able to do. A cross-cutting competency layer maps skills that grow through everything — creativity, critical thinking, frustration tolerance — observed bidirectionally across learning situations.

These internal layers are part of the original architectural contribution. They do not exist in any current system.

The graph is alive

Initial construction is AI-assisted: automated methods extract prerequisite relationships with usable precision (F1 scores of 70–90% depending on domain), providing a starting point for human validation. Continuous refinement happens through expert review, implicit validation through millions of learning trajectories, and ingestion of new findings.

The system discovers relationship types never pre-modelled. After millions of trajectories, it might find that mastering concept X systematically blocks concept Y — not a "requires" relationship but an "interferes with" edge that no one anticipated.

Adoption, not regulation.

The cold start

Adoption follows a specific dynamic: the graph becomes irreversible when it is referenced — by Mentors generating experiences, by evidence anchored to its nodes, by universities recognising its structure, by employers using its competency definitions. De-facto standardisation through utility, not regulation.

But this dynamic has a cold-start problem: the graph needs users to become valuable, and users need the graph to be valuable. The natural first vertical is mathematics from primary school through university-level calculus — a domain with clear dependencies, minimal cultural contestation and existing prerequisite research to validate against. The cold start is the single hardest practical challenge in the entire architecture.

Reference

Architecture paper, Section 5, L5. DOI: 10.5281/zenodo.18759134. CC BY 4.0.