Mind · Strategy
MentalSim
Predictive modeling of system dynamics
Definition
Mechanism
System 2 Simulation. Agent constructs a causal graph of the target system. It executes a 'Dry Run' of the proposed plan within that graph, inside an isolated AgentSandbox, to predict State(t+1) — toggling between fast HeuristicSnap and rigorous Deep simulation modes.
Preconditions
- Causal model of target system available
- Initial state known
Postconditions
- Probabilistic outcome vector generated
Invariants
- Builds a causal graph of the target System before running anything.
- The Simulation produces no side effects in the production context.
- Costs at most `cost_fraction` of executing the plan for real.
Risks
Failure modes
Map-Territory Error
Simulation diverges from reality (Hallucinated Physics).
Overconfidence
Agent trusts the simulation execution log more than real-world feedback.
Relationships
Builds on
References
Autonomous Goal-Directed Entity
Society
Isolate agent execution with resource limits and egress controls
Vertical search dimension
Rapid, low-cost decision making via pattern recognition
A virtualized execution
Infrastructure
System condition at time T
Infrastructure
Complex whole of interacting parts
Used by
References
Pre-execution mental model
Reference
Parameters
cost_fractionRatio, (0.0, 0.5]- Share of real execution cost the simulation may consume before simulating is not worth it.
fidelityEnum, {Low, High}- Whether the causal graph models coarse dynamics or detailed ones.
stepsInteger, [1, 10]- How far ahead the dry run looks.