Society · Protocols
EvaluatorOptimizer
Generate-evaluate-refine loop
Definition
Mechanism
A two-role Loop to Optimize a solution where a Generator produces output and an Evaluator critiques it, providing structured feedback. The feedback is fed back to the Generator for refinement. Loop continues until the Evaluator approves or max iterations reached. The Evaluator can be the same model with a different Context, or a separate model. It invokes a MetaCheck to ensure the scoring Criteria remain consistent across optimization cycles.
Preconditions
- Initial solution
- Modification operator
- Scoring function
Postconditions
- High-quality solution found
Invariants
- Feedback loop converges
- Optimizer improves score on Evaluator metric
Risks
Failure modes
Evaluation Drift
Evaluator criteria shift across iterations.
Cosmetic Changes
Generator makes superficial edits that don't address feedback.
Evaluator Capture
Generator learns to game the evaluator's criteria.
Relationships
Builds on
Nothing in this library builds on this pattern yet.
Reference
Signature
Optimize(Loop)