OCG
OCG
Open Cognitive Graph
Trunk v1.2 · Live
Principle 1

Structural Transparency:
Every Judgment Is Inspectable

OCG's key contribution is not being "smarter" — it is placing epistemic authority at an abstraction level that humans can inspect. When AI says "you're wrong", there must be a structure behind it that domain experts can review, dispute, and revise — not a black box.

⬛
Black-box LLM Opaque
Student Response ? AI: answer is incorrect × Opaque　× Unlocatable　× Unfixable
◎
OCG Governance Inspectable
Student Response common_misconception@Physics Confusing F=ma with mass alone prereq: velocity → acceleration Dr. Wang · 2024-09 provenance prerequisite_of Velocity & Acceleration AI: answer is incorrect ✓ Inspectable　✓ Addressable　✓ Fixable
OCG's key contribution is not being "smarter" — it is placing epistemic authority at an abstraction level that humans can inspect: concepts, relations, provenance — not parameters. When AI says "you're wrong", there must be a structure behind it that a domain expert can audit, contest, and revise.
Principle 2

Targeted Fix:
Errors Have a Structural Address

In OCG, finding the structural address enables a targeted fix — inserting scaffold nodes immediately affects every output that depends on that path. Black-box models cannot do this: the root cause is diffused across parameter space.

Interactive demo — click buttons to observe graph changes
Energy as Property concept@Physics Conservation of Energy concept@Physics prerequisite_of @Science ⚠ Cognitive leap too large, no bridge nodes Energy Transfer scaffolds@Science ✦ New Scaffold Node System Boundary scaffolds@Science ✦ New Scaffold Node
▸ Click “Diagnose” to reveal the structural defect
⚠
Issue found
This edge directly links “Energy as Property” to “Conservation of Energy”, assuming a cognitive leap that students cannot make — intermediate bridge concepts are missing.
✓
Fix applied
2 scaffold nodes inserted (Energy Transfer, System Boundary). The cognitive span of each step is now valid. All outputs depending on this path benefit immediately.
⚑
Problem: Cognitive Leap

The raw prerequisite_of edge assumes students can jump directly from “energy is a property” to “energy is conserved” — an empirically proven over-stretch.

Black-box Issue found → wait for vendor to retrain the entire model, root cause unlocatable
OCG Issue found → locate node → propose → validate → update
✦
Fix: Scaffolding Strategy

Insert intermediate concept nodes so that the cognitive span of every step stays within learnable range. New nodes use the scaffolds relation type, making their role semantically explicit.

Energy Transfer System Boundary scaffolds@Science

A structural address means every fix has a precise scope — no unintended ripple effects, no hidden side-effects elsewhere in the graph.

Every error has a structural address — not a “global model bias”, but a single edge that assumed an unjustified cognitive leap. In OCG, finding the address enables a targeted fix that immediately propagates to every output that uses this path.
Principle 3

Pluralism:
Branches Contain Disagreement

OCG's trunk-branch architecture does not resolve "which view is correct" — it specifies "under which epistemic domain a view holds". Branches let plural perspectives coexist legitimately while the trunk remains stable.

⊕
Trunk-Branch Architecture Consensus trunk · 3 domain branches
Trunk Consensus layer — Academic Committee Branch: Euclidean Geometry @Geometry@Euclidean P5 Parallel Postulate Unique valid path Prop.47 (Pythagorean Thm.) Strictly depends on P5 Branch: Philosophy @Philosophy (multi-school) Rationalism Reason as foundation Empiricism Perception as foundation Competing branches coexist — no arbitration analogous_to bridges cross-school concepts Branch: Non-Euclidean @Geometry@Spherical P5 replaced (parallels meet) Valid in branch, trunk unaffected Spherical triangle angles > 180° Contradicts Euclid, yet coexists analogous_to: Euclid's "postulate" ≈ Descartes' "first principle"
◈
Trunk Rules

The trunk stores only cross-domain consensus concepts. Any contested content must go into a branch — never the trunk — ensuring the stability of the shared consensus layer.

⑂
Branch Isolation

Replacing P5 in non-Euclidean geometry does not "pollute" the Euclidean branch. Each branch maintains its own axiomatic environment; divergences are stored explicitly as structure.

↔
Cross-Domain Links

The analogous_to relation enables cross-branch analogy without unifying ontologies, building semantic bridges while preserving each domain's independence.

OCG's trunk-branch model achieves “disagreements don’t need to be arbitrated — they just need to be located”. The divide between Rationalism and Empiricism doesn't need an Academic Committee ruling; both exist legitimately in their own branches.
Principle 4

Governance Process:
Authority Through Due Process

Governance is not an external constraint — it is a process built into the structure itself. Every change has a proposer, evidence, validator, and timestamp, giving a structural answer to "who may change", "what changed", and "on what grounds".

⏱
Governance Lifecycle
✓
Teacher observes a problem
Students consistently struggle with the "energy property → conservation" transition across multiple assessments
✓
Locate the structural address
OCG reveals: this edge assumes a cognitive leap that does not hold — scaffold nodes needed
3
Proposal: insert scaffold nodes
Branch maintainers submit change → automated consistency check → enters review queue
4
Expert review + pilot
Learning science researchers validate → small-scale pilot (3 schools, 450 students) → data collected
5
Trunk updated, propagated to all branches
Change written to Trunk OCG → all students benefit immediately, history preserved
provenance_record.yaml
node: EnergyTransfer@Science type: scaffolds proposed_by: "Middle School Teachers Alliance" evidence: "3 schools, 450 students, 4-week pilot" validated_by: "Learning Science Committee" validated_at: 2025-03-10 trunk_version: v1.2 status: "propagated_to_all_branches"
⇄
Core Contrast with Black-box
Black-box Issue found → unlocatable → wait for full model retrain → still unexplainable
OCG Issue found → locate node → propose → validate → update → fully auditable

Epistemic authority comes from due process, not "the AI said so" — every node change is traceable to a specific proposer, evidence base, and validator.

450
pilot students
3
schools
v1.2
current trunk
Every change has a proposer, evidence base, validator, and timestamp. This provenance mechanism ensures that governance authority does not rest on any single organisation's unilateral decision, but is derived from the process itself.