README.md

Example: Code Optimization

Goal: optimize sort.py so sorting 1,000,000 integers runs below 0.5 seconds while preserving correctness.

Field	Value
Metric	Median runtime in seconds (median_time_s, minimize)
Evaluator	python benchmark.py
Baseline	2.117478s
Best result	0.150459s
Iterations logged	8
Visual result	results.png

Expected Output

This example demonstrates the standard autoresearch artifact bundle:

• research.md: goal, constraints, search space, and iteration history
• research_log.md: detailed hypothesis/evaluation notes
• autoresearch-results.tsv: machine-readable metric trace
• results.png: visual summary of runtime improvement and best-so-far curve
• final_report.md: summary of the best implementation and failed alternatives

Reproduce the Evaluator

From this directory:

python benchmark.py

The evaluator emits the mechanical contract used by the skill:

{"pass": true, "score": -0.150459}

For minimize metrics, score is the negated metric value so higher scores are always better.