This dataset holds the test cases of VeriContest, a benchmark of 1,006 competitive-programming problems from LeetCode and Codeforces for verifiable code generation in Rust with Verus. VeriContest accompanies the paper "VeriContest: A Competitive-Programming Benchmark for Verifiable Code Generation" (arXiv:2605.08553), accepted at NeurIPS 2026 in the Evaluations & Datasets Track.

The problems themselves (descriptions, specifications, code, proofs and output checkers) are in VeriContest and VeriContest Lite. The test cases are kept here, in one place for both, so that the problem datasets stay small and nobody has to download the tests twice. Download them only when you run tests: to score code, or to check specifications against positive and negative cases.

2,012 files, about 49 GiB: two files for each of the 1,006 problems.
tests/testcases.jsonl: positive test cases. Each pairs a valid input with the output of the verified reference program.
tests/mutated_testcases.jsonl: negative test cases. Each pairs a valid input with an incorrect output that a complete postcondition must reject.
lite_problem_ids.txt: the 300 problems of VeriContest Lite, one per line.

codeforces/
  cf1006C/
    tests/
      testcases.jsonl
      mutated_testcases.jsonl
leetcode/
  lc1/
    tests/
      testcases.jsonl
      mutated_testcases.jsonl
lite_problem_ids.txt

Each line is a JSON object with input and output. For LeetCode problems, input is an object keyed by the function's parameter names and output is the return value. For Codeforces problems, both are the complete stdin and stdout strings.

{"input": "1\n4\n4 7 3 5\n", "output": "-3 2 -4 -2\n"}

Some problems accept several correct outputs. For those, the output in testcases.jsonl is one valid answer, and the problem's tests/checker.py in VeriContest decides whether another output is also correct.

Downloading

All test cases:

from huggingface_hub import snapshot_download

snapshot_download("Gax-c/VeriContest_Tests", repo_type="dataset",
                  local_dir="vericontest-tests")

Only the 300 VeriContest Lite problems (about 17 GiB):

from huggingface_hub import hf_hub_download, snapshot_download

ids_file = hf_hub_download("Gax-c/VeriContest_Tests", "lite_problem_ids.txt",
                           repo_type="dataset")
lite = open(ids_file).read().split()
snapshot_download("Gax-c/VeriContest_Tests", repo_type="dataset",
                  local_dir="vericontest-tests",
                  allow_patterns=[f"{p}/tests/*" for p in lite])

One problem:

import json
from huggingface_hub import hf_hub_download

path = hf_hub_download("Gax-c/VeriContest_Tests",
                       "leetcode/lc1/tests/testcases.jsonl", repo_type="dataset")
with open(path) as f:
    testcases = [json.loads(line) for line in f]

The testcases_path and mutated_testcases_path columns of the VeriContest and VeriContest Lite tables are paths in this dataset.

To score programs with the VeriContest evaluator, pass the download folder as --testcases:

python3 benchmarks/verus/evaluation/score.py --split lite --testcases vericontest-tests ...

Versions

The test cases are those of release v2026.10.1, the release the leaderboard results were evaluated on. The tag v2026.10.1 marks them; to pin them, download with revision="v2026.10.1".

License

The test cases are released under CC BY 4.0. Problem statements derive from LeetCode and Codeforces and remain subject to their source terms, as detailed in DATA_LICENSE.md.

Citation

@article{xie2026vericontest,
  title={{VeriContest}: A Competitive-Programming Benchmark for Verifiable Code Generation},
  author={Xie, Zichen and Pawagi, Mrigank and Liu, Yuxin and Rai, Aaditi and Shao, Lize and Berberian, Jr., John and Che, Sicong and Wang, Wenxi},
  journal={arXiv preprint arXiv:2605.08553},
  year={2026}
}