whetstone_DSL/datasets/project_benchmarks/README.md

Common Projects Benchmark (100)

This dataset is a practical benchmark list of common production project types. Each line in common_projects_100.jsonl is a single benchmark seed with:

• id: stable benchmark id
• project: short project name
• category: workload family
• language: target implementation language
• spec: concise implementation prompt seed

Use this list to generate taskitems, run A/B execution, and collect training data from gate outputs, remediation traces, token accounting, and final outcomes.