Concept Kernels

A Concept Kernel is the fundamental unit of the protocol. Every concept — from Cat to Invoice to Gene — is an autonomous kernel process that owns and governs its own semantic definition.

What Makes a Kernel?

A kernel is not just data. It is a self-governing process with:

• Identity — A unique name and version within the protocol
• Ontology — A LinkML schema defining its structure, slots, and enums
• Constraints — SHACL shapes that validate all mutations
• Relationships — Typed edges to other kernels (REL, LINK)
• Ingress/Egress — Validated input and audited output channels
• Ledger — Immutable record of all protocol interactions
• Lifecycle — States including proposed, active, locked, and archived

Kernel Lifecycle

PROPOSED → VALIDATING → ACTIVE → [LOCKED] → ARCHIVED
              ↓                      ↑
           REJECTED              CONSENSUS
                                  UNLOCK

1. Proposed — A new concept is submitted for admission
2. Validating — Ontology and SHACL checks run; LLM semantic validation
3. Active — Kernel is live and accepting protocol interactions
4. Locked — Mutations are paused pending consensus resolution
5. Archived — Kernel is retired but its ledger is preserved

Kernel Isolation

Each kernel is isolated by design:

• No kernel can directly mutate another kernel's state
• Cross-kernel interactions flow through the protocol
• A locked kernel cannot be unlocked without consensus
• Every mutation produces a cryptographic proof

This isolation ensures that even in a fleet of millions of kernels, semantic integrity is maintained through the protocol rather than through trust.

Duplicate Handling

When a duplicate concept is proposed, the protocol does not reject it. Instead:

1. DSPy-driven semantic comparison identifies the overlap
2. A merge protocol is initiated
3. Consensus determines whether to merge, fork, or reject
4. The decision is logged with full provenance

---

Ask Questions on Discord