State
What an intelligence, system or environment knows, observes, believes and can validly act upon at a point in time.
Trajectory
intelligence.
The science of how intelligences move through time, state, alternatives, interaction and consequence.
Trajectory Intelligence represents, compares, simulates, evaluates and transforms the evolving paths of models, agents, machines and environments. It extends beyond context, memory, retrieval, event logs and generic continual learning by preserving the conditions under which a claim, decision, action or outcome is valid.
Core primitives
The field begins with typed relationships that survive abstraction, handoff and time.
Branch
Realized and unrealized paths, retained without collapsing alternatives, uncertainty or counterfactual structure.
Evidence
The origin, independence, authority, scope and verification state behind every consequential claim.
Consequence
The independently observed change in the world—not merely an intention, tool response or declared completion.
Research layers
Trajectory
A structured path through changing state, retaining decisions, evidence, alternatives, interactions and consequences.
Graph
A typed relational representation connecting trajectories across entities, timescales, environments and abstraction levels.
Field
The dynamics that emerge when trajectories interact, constrain, amplify, redirect or transform one another.
Runtime
Infrastructure that captures, reconciles, branches, compares, retrieves, simulates and governs trajectories during operation.
Evaluation
Measurement of continuity, coverage, attribution, branch quality, consequence, coordination, recovery and transfer.
Open questions
The work begins where the existing stack stops.
- 01How should an autonomous system preserve identity and state across model changes?
- 02When does retrieved context fail to represent the current world?
- 03How can realized and unrealized branches coexist without collapsing uncertainty?
- 04What establishes that a trajectory caused—not merely preceded—an outcome?
- 05How should agents act when their information states and authorities are asymmetric?
- 06When should an experience update a memory, skill, model or world state?
Method
Define. Falsify. Formalize. Build. Measure.
Every proposed primitive is tested against adjacent formalisms and prior art, then operationalized through experiments, benchmarks and infrastructure. A category earns its existence through explanatory and operational power.
Enter the research frontier