Module bench 

Model evaluation and benchmarking framework (spec §7.10) Model Evaluation and Benchmarking Framework (aprender::bench)

Provides multi-model comparison for evaluating .apr models on custom tasks. Unlike QA (single-model validation), this module compares multiple models to find the smallest model that meets a performance threshold.

Toyota Way Alignment

• Pull Systems (P3): Pareto frontier pulls smallest viable model
• Muda Elimination: Avoid overprovisioning with right-sized models

References

• Deb et al. (2002) “NSGA-II” for Pareto optimization

Example

use aprender::bench::{EvalResult, ModelComparison};

let comparison = ModelComparison::new("python-to-rust");
assert!(comparison.results.is_empty());

Modules

pareto

Pareto Frontier Computation

py2rs

Python to Rust Single-Shot Compile Benchmark (10 Levels)

Structs

EvalResult

Result of evaluating a single model on a single task

EvalSuiteConfig

Evaluation suite configuration

Example

Example input for evaluation

ExampleResult

Result for a single example

ModelComparison

Compare multiple models on the same task

ParetoPoint

Point on the Pareto frontier

Recommendation

Recommendation for a specific scenario

Enums

Difficulty

Difficulty tier for stratified analysis

ExampleStatus

Status of an example evaluation

Traits

EvalTask

Custom evaluation task trait