/ 
Sim-AgentSoar-0.06
/ Sim::AgentSoar::Engine

NAME

Sim::AgentSoar::Engine - Deterministic calibration environment for heuristic AgentSoar

DESCRIPTION

Sim::AgentSoar::Engine defines a minimal deterministic environment used to calibrate and evaluate the Sim::AgentSoar architecture.

The domain consists of integer reachability under a fixed set of operators. Although simple, it is intentionally chosen to expose core AgentSoar dynamics:

  • Local vs. global improvement tension

  • Reversible and irreversible operations

  • Potential regression

  • Cycles and repetition

  • Heuristic bias evaluation

This module contains no probabilistic logic and no LLM interaction. All state transitions and evaluations are deterministic.

Domain Definition

State:

value => integer

Operators:

add_1
sub_1
add_3
sub_3
mul_2
div_2_if_even

Evaluation metric:

abs(value - target)

Domain bounds:

-100 <= value <= 100

Purpose

The integer domain serves as:

  • A calibration environment for testing heuristic quality

  • A controlled benchmark for branching and correction logic

  • A deterministic substrate for measuring LLM efficiency

  • A regression testing environment for architectural changes

Because the environment is fully observable and inexpensive to evaluate, it allows clear separation between structural AgentSoar performance and heuristic proposal behavior.

ReAgentSoar Role

In the broader reAgentSoar program, this module acts as:

  • A wind-tunnel model for cognitive architecture experiments

  • A baseline for comparing deterministic and LLM-guided AgentSoar

  • A controlled testbed for future memory-layer extensions

The simplicity of the domain is intentional: it isolates architectural effects from domain complexity.

METHODS

allowed_operators

Returns the list of valid operators.

apply_operator

Applies an operator to a state value.

valid_value

Ensures state remains within domain bounds.

metric

Computes distance to target.

AUTHOR

Gian Luca Brunetti (2026), gianluca.brunetti@gmail.com