Mind · Strategy
CapacityPressure
Forcing abstraction via resource starvation
Definition
Mechanism
A regularization pattern that forces abstraction by artificially constraining resources (bandwidth, memory, parameter count, or time). By creating a bottleneck where Capacity < Information, the agent is compelled to compress the signal, discarding noise and memorized details in favor of high-level concepts and generalizations. It artificially tightens the Budget, forcing the agent to employ Generalize, ConceptBlend, and ContextCompress to fit the signal within the bottleneck.
Preconditions
- A mechanism to enforce hard limits (e.g., context window, dimensions, token count)
- A stream of information or a learning task
Postconditions
- A highly compressed, abstract representation of the input
- Loss of fine-grained detail/noise
Invariants
- Bottleneck Existence: Available Capacity must be strictly less than Input Information Content
- Lossy Requirement: Output must be a simplified representation, effectively barring exact replication/memorization
- Utility Preservation: The compressed form must retain predictive power for the target task
Risks
Failure modes
Collapse
Constraint is too tight; signal is lost entirely (underfitting).
Adversarial Encoding
Agent finds a way to 'zip' noise rather than abstracting (violating the spirit of the constraint).
False Abstraction
Agent hallucinates simple rules that don't actually exist to satisfy the budget.
Relationships
Builds on
References
Autonomous Goal-Directed Entity
Infrastructure
Resource allocation limit
Atomic fusion of two unrelated concepts into a novel third
Infrastructure
Non-negotiable boundary condition
Semantic compression for long-running contexts
Inductive pattern extraction
Nothing in this library builds on this pattern yet.
Reference
Parameters
compression_ratioRatio, [0.0, 1.0]- Target resource utilization ratio that triggers forced abstraction
resource_typeEnum, {Compute, Memory, Attention, Budget}- Type of resource being constrained (compute, memory, bandwidth)