dataset/README.md

# Buggy Training Dataset

This dataset is organized as:

- `problem_x/buggy.py`: intentionally buggy implementation
- `problem_x/test.py`: correctness tests that should fail before fixes
- optional extra modules (`helpers.py`, `cache.py`, etc.) to support multi-file bug fixing

Current problems: `problem_1` to `problem_18`.

Bug patterns included:
- off-by-one errors
- boundary condition mistakes
- incorrect sorting direction
- exception handling mistakes
- state/recency bugs in cache logic
- recursive base-case bugs
- parsing and whitespace normalization issues
- order-preservation regressions
- matrix transformation direction errors