Governance

CKP governance is the process by which the concept graph evolves over time. It is designed to be transparent, auditable, and resistant to semantic drift.

Governance Lifecycle

DISCOVERY → PROPOSAL → VALIDATION → CONSENSUS → INTEGRATION → EVOLUTION

1. Discovery — Agents identify gaps, overlaps, or inconsistencies in the concept graph
2. Proposal — A formal change is submitted with schema, rationale, and impact analysis
3. Validation — Automated checks run against ontology, SHACL, and existing constraints
4. Consensus — Stakeholders review and vote on the proposal
5. Integration — Accepted changes are committed with cryptographic proofs
6. Evolution — The concept enters production and is monitored for downstream effects

Dynamic ORBAC

CKP uses Object Role-Based Access Control with dynamic, scoped permissions:

• Temporary roles — Assigned with explicit time and scope bounds
• Hierarchical access — Roles inherit permissions from parent roles
• Audit trail — Every permission change is logged to the ledger
• Revocation — Permissions can be revoked at any time with immediate effect

Guardian("OntologyTeam")
  ├── can: PROPOSE, VALIDATE, VOTE
  ├── scope: concepts/biology/*
  ├── expires: 2025-06-01
  └── granted_by: CK_Core[tx892]

Concept Admission

When a new concept is proposed, it goes through a structured admission workflow:

1. Submission — Agent submits concept with LinkML schema
2. Semantic Check — DSPy-driven analysis for duplicates and conflicts
3. Ontology Validation — LinkML and SHACL compliance checks
4. Relationship Mapping — Identify and validate connections to existing concepts
5. Consensus — If conflicts detected, stakeholders vote on resolution
6. Commitment — Concept is committed with proof and logged

Duplicate proposals trigger a merge protocol rather than rejection — the protocol assumes that convergent proposals indicate genuine need.

Constraint Governance

Constraints themselves are governed:

• Constraints can be proposed, voted on, and versioned
• A constraint override requires consensus with explicit justification
• Override history is immutable and auditable
• Constraints aggregate during burst propagation across the concept graph

Trust and Reputation

CKP tracks agent reliability through:

• Proof history — How many proposals were accepted vs. rejected
• Consensus participation — Voting consistency and timeliness
• Constraint compliance — How often an agent's actions pass validation

Trust scores inform but do not dictate governance decisions. The protocol is designed to be trust-minimized — verification over reputation.

---

Shape Governance on Discord