Governed infrastructure for organizational decision-making.
GrytLabs is a research institute building the governed infrastructure that organizations need to make decisions accountable — at human speed and at machine speed.
The research began with a single question: why does the context behind organizational decisions disappear the moment those decisions are made? Twenty-two research sprints across fifteen academic traditions produced a clear finding — six independent fields had each hit the same structural wall. This is the reconstruction problem. The Decision Lineage Protocol is the formal answer: nineteen primitives, ten behavioral invariants, a truth type system that distinguishes what is known from what is declared from what is derived. The protocol is open, published, and available for anyone to build on.
22 research sprints...
six traditions...
convergence proof...
reconstruction problem...
Protocol: DLP v2.0.0
Primitives: 19
Invariants: B1-B10
MODEL
22 research sprints...
six traditions...
convergence proof...
reconstruction problem...
Protocol: DLP v2.0.0
Primitives: 19
Invariants: B1-B10
The protocol is the grammar. The world model is what you build with it.
What is a world model?
The AI tools most people use today are prediction engines. They generate whatever is statistically coherent — not necessarily what is true, what is allowed, or what your organization actually committed to last quarter. They have no memory that persists between sessions, no understanding of your constraints, no sense of who has authority to decide what. They predict. They do not govern.
A world model is the missing layer. It is a structured representation of how an organization actually works — not a dashboard, not a simulation, not a chatbot’s summary. Formally, it is the governed composition of three models: a data model (what information exists and how it relates), a computational model (how that information changes under decisions and events), and a conceptual model (why the structure matters and what it means).
These compose into ten governed records — purpose, entities, distinctions, relations, time, transition logic, constraints, observation, uncertainty, and memory. When an organization has a world model, every decision carries the full context of every decision that came before it. The AI doesn’t guess what your business does. It knows — because the structure was given, not learned.
That is what makes governance possible at machine speed.
The field uses “model” to mean at least six distinct things — world model, dynamics model, neural network model, mental model, reward model, causal model. A seventh — the governance model — is missing from the discourse entirely. These decompose into three categories:
Without the governance layer, these are three disconnected model types. With it, they compose into a world model.
The two boundaries are nearly aligned because governance is not a separate component — it is the binding force that gives the composition its identity.
The Lineage
The Decision Lineage Protocol did not begin as a protocol. It began as a question — and the path from question to formal grammar followed five stages.
Theory
The reconstruction problem. Twenty-two research sprints across fifteen academic traditions. Six-way convergence. The finding that governance-grade world models cannot be learned from data — they must be architecturally given.
Architecture
The Decision Lineage Protocol. Nineteen primitives, ten behavioral invariants, a truth type system. The formal grammar of organizational governance. Published, open, available for anyone to build on.
Experiment
Three implementations, each proving the protocol works in a different domain. These are research experiments with product surfaces — tangible evidence, not the point.
Implement
Dogfooding only. GrytLabs uses its own protocol to govern itself. The world model index, the governed records, the research agenda — all built on DLP.
Contribute
Publication, open-source release, foundation governance. The protocol is open. The research atlas is public. The specification is available for anyone to build on.
The Equity Gap
The equity gap in organizational infrastructure is structural. The tools, frameworks, and institutional knowledge that large organizations take for granted — governance structures, decision accountability, operational coherence — are inaccessible to the founders, small businesses, and independent operators who need them most. This is not a market failure. It is an infrastructure problem.
84%
of small businesses cannot keep pace with the technology budgets of larger organizations. 42% lack access to resources and expertise to deploy AI.
Sources: Startups Magazine; Goldman Sachs / Old National Bank
GrytLabs was founded to close that gap through research. The investigation spanned twenty-two research sprints across fifteen academic traditions and produced one clear finding: six independent fields had all converged on the same structural wall. The context that makes governance possible — the decision history, the reasoning, the constraints, the commitments — dissipates the moment a decision is made. This is the reconstruction problem. At human speed, it is recoverable through retrospective audit. At machine speed — AI agents making consequential decisions autonomously — it is existential.
The Decision Lineage Protocol (DLP) emerged from that six-way convergence. It is the formal grammar of organizational governance: nineteen primitives, ten behavioral invariants, a truth type system that distinguishes what is known from what is declared from what is derived. The protocol is open, published, and available for anyone to build on. A utility patent (pending) protects the novel architectural claims while ensuring the protocol remains open infrastructure.
Read the research at research.grytlabs.ai. Read the specification at decisionlineageprotocol.io.