---
license: cc-by-4.0
language:
- en
tags:
- combinatorial-reasoning
- constraint-satisfaction
- benchmark
- llm-evaluation
size_categories:
- n<1K
configs:
- config_name: default
  data_files:
  - split: test
    path: problems.jsonl
---

# MathConstraint-Easy

A 266-instance benchmark of natural-language combinatorial constraint-satisfaction problems with solver-verified ground truth. 25 problem types drawn from constraint programming (PyCSP3) and SAT-modulo-symmetries on graphs (pysms). 206 SAT / 60 UNSAT; SAT instances split 103 / 103 into hinted and unhinted. A harder companion split is at [MathConstraint](https://huggingface.co/datasets/MathConstraint/MathConstraint).

## Files

- `problems.jsonl` — canonical data file; one JSON record per line
- `instances/<name>.json` — same data, one file per instance
- `manifest.json` — schema, type / SAT / wall-time stats, generation provenance
- `_raw/<name>.{xml,cnf}` — solver model file (XCSP3 for pycsp problems, CNF DIMACS for pysms graph problems)
- `_raw/<name>.out.json` — solver stdout / stderr / command for the reference verdict

The `_raw/` files let any external SAT/CSP solver independently verify the labels.

## License

CC-BY-4.0

## Citation

Anonymous. Citation provided in camera-ready.