CRAVE: 1,200 Balanced Code Review Verdicts from 600 Pull Requests

CRAVE provides 1,200 high-quality samples for code review classification, released by TuringEnterprises in late 2025. The data covers 600 pull requests across 123 distinct repositories, focusing on binary classification of code changes.

Use Cases

• Training models to predict APPROVE or REQUEST_CHANGES labels for code changes
• Benchmarking code review agents using the 600 pull request samples
• Analyzing characteristics of code changes that trigger a REQUEST_CHANGES verdict

Strengths

• 1,200 balanced samples
• Data sourced from 123 unique repositories
• Filtered for high quality
• MIT licensed

Limitations

• Small sample size of 1,200 records
• Limited to binary classification labels (APPROVE vs REQUEST_CHANGES)

Provenance

Source

TuringEnterprises

Collection Method

Filtered and selected from 600 pull requests

Freshness

Last updated December 2025.

Geography

United States