Decision Lineage Protocol
Open governance infrastructure. 19 primitives. 10 invariants. The grammar of organizational decision-making.
What is DLP
The Decision Lineage Protocol is the structural layer that turns a generic prediction engine into a governed decision system.
An LLM is a next-token prediction engine. Raw prediction is statistical pattern matching. It has no physics, no dynamics, no epistemics. It’ll predict whatever is coherent, not necessarily what is true. For consequential decision-making, that’s unacceptable.
DLP provides the missing structural layer:
19 Primitives
The irreducible vocabulary of governance. Every primitive maps to a distinct semantic function. Fewer than 19 is insufficient; more than 19 is redundant.
10 Behavioral Invariants
The physics. What’s structurally possible and impossible. No state machine transition can violate B1-B10.
5-Tier Architecture
Canonical primitives → facet expressions → archetype specialization → user customization → instance state. Each layer compiles to the one below it.
Truth Type System
Four epistemic types: what you know (Authoritative), what you declared (Declared), what you derived (Derived), and what you can’t inspect (Opaque). Every fact carries its epistemic status.
The protocol is runtime-agnostic. JEPA fits. LLMs fit. Future models fit. The invariants and truth types constrain the runtime but don’t care which runtime executes them. This is the socket. Your choice of runtime is the plug.
Specification
The full formal specification of the Decision Lineage Protocol.
- Formal semantics of the 19 irreducible primitives
- Ten behavioral invariants (B1–B10) with SHACL enforcement
- Four truth types and epistemic promotion paths
- State machine definitions (SM-1 through SM-10)
- Governance activation pipeline and policy projection surfaces
- Human capture mechanisms and decision surface design
- Substrate profiles (EAS, BAS, PAS) and actor taxonomy
- Dual-language query architecture (Cypher + SQL)