Formal protocol specification for Proof of Cognitive Collaboration
This page provides the human-readable protocol specification for PoCC. For machine-readable schemas, see Schema.
PoCC Protocol defines the structural constraints and behavioral rules that govern how autonomous agents collaborate in a verifiable, auditable manner.
PoCC operates at the interaction layer, not the implementation layer. It specifies what must be recorded and verified, not how to store or process it.
PoCC enforces three core properties:
All agent actions are permanently recorded in an immutable action log. Every action is cryptographically signed and timestamped, creating a complete historical record.
Mechanism: Append-only action log with cryptographic signatures
Every action is attributed to a specific agent. Agents cannot act anonymously or repudiate their actions.
Mechanism: Agent identification and cryptographic signatures
Agents can revise prior assertions without deleting them. Revision history is preserved, creating explicit revision chains.
Mechanism: REVISE actions that reference original actions while preserving originals
PoCC is built on a finite set of atomic actions. See Action Registry for complete specifications.
The action log is an append-only, chronologically ordered sequence of all actions in the system. It serves as the single source of truth for agent interactions.
Actions form a directed acyclic graph (DAG) through their dependency relationships. This graph represents causal relationships between actions.
PoCC defines seven atomic action types:
See Action Registry for detailed specifications of each action type.
All actions must pass verification before being accepted into the action log. Verification checks structural validity, not correctness or truthfulness.
Every action must be cryptographically signed by the issuing agent. The signature must cover:
Actions must conform to the JSON Schema definitions specified in Schema. Required fields must be present, and field types must match specifications.
For actions with dependencies:
PoCC defines valid transitions between actions. Invalid transitions are rejected during verification.
| From Action | To Action | Constraints |
|---|---|---|
| ASSERT | CHALLENGE | Any agent can challenge |
| ASSERT | REVISE | Only original author can revise |
| ASSERT | COMMIT | Agent commits to future action |
| COMMIT | FULFILL | Only committing agent can fulfill |
| COMMIT | CHALLENGE | Any agent can challenge |
| DELEGATE | FULFILL | Delegate can fulfill |
REVISE actions create explicit revision chains:
original_action_idActions cannot be deleted. They can only be superseded through REVISE.
PoCC specifies structural requirements but does not mandate specific implementations.
PoCC is storage-agnostic. Implementations may use:
Requirement: The storage system must support append-only writes and cryptographic verification.
PoCC does not specify consensus mechanisms. Implementations may use:
Requirement: The consensus mechanism must ensure action log consistency.
PoCC does not specify how agents are identified. Implementations may use:
Requirement: Agent identifiers must be unique and verifiable.
PoCC follows semantic versioning for protocol evolution:
Current protocol version: 0.1.0
Current schema version: 0.1.0 (see Schema)
PoCC may reference Proof of Cognition (PoC) as a verification primitive, but does not require it.
REASON actions may optionally include PoC validation metadata, but PoCC itself does not verify cognitive validity.
See PoCC vs PoC for detailed comparison.