RAS 2026 PSC Release v2.1 — full benchmark package
Public archival release of the INFORMS Railway Applications Section (RAS) 2026 Problem Solving Competition benchmark (Railroad Blocking Problem for North American Class I Railroads), mirrored by the ASU Trans+AI Lab. Official competition pages:
- Kaggle: https://www.kaggle.com/competitions/informs-ras-2026-problem-solving-competition
- INFORMS RAS: https://connect.informs.org/railway-applications/new-item3/problem-solving-competition682
What is in RAS2026-PSC-v2.1-full.zip
Everything in the repository at tag v2.1, with the compressed inputs already expanded, plus the generated sample submission file:
| Path | Contents |
|---|---|
datasets/l1/, l2/, l3/ |
node.csv, link.csv (raw, ~73 MB each), demand.csv, setting.csv, yard_car_demand_summary.csv, layer README |
datasets/schemas/ |
JSON Schemas (Draft 2020-12) and validate_csvs.py |
datasets/DATASET_README.md |
Field-level data reference |
scoring/ |
fast_validator_v2_0.py, metric_fast_v2_0.ipynb, json2csv.py, submission_check.py, solution.csv, od_distance_matrix.csv (raw, 2,000,906 rows), sample_solutions/ (9 JSON files), submission.csv (9-case sample built with json2csv.py), SCORE_README.md |
docs/ |
problem_statement.pdf, GitHub Pages sources |
| root | README.md, DATA_PROVENANCE_AND_USE.md, NOTICE.md, LICENSE-CODE, CITATION.cff, requirements.txt, unpack.py |
The per-layer link.csv.zip, yard_to_yard_min_distance.csv.zip, and sample_solutions.zip files that live in Git are omitted from the full package because their expanded contents are included.
Scoring and validator version
Scoring package v2.0 (June 22, 2026): yard-track limit counts only manifest/coal/grain blocks; 1b_no_subtour and 9_direct_block_rule checks; 9b_demand_volume_consistency retained from v1.1; per-block two-endpoint Class-I-only interchange cost with CSXT → CSX normalization. See scoring/SCORE_README.md.
Data provenance notice
RAS2026-PSC is a research-derived and benchmark-calibrated competition dataset, not an operational or proprietary railroad dataset. Demand records were developed using transformed Freight Analysis Framework (FAF)-based freight-flow references (obtain original FAF data from BTS: https://www.bts.gov/faf/faf5). Physical network connectivity is research-derived in part from OpenStreetMap (© OpenStreetMap contributors, ODbL 1.0: https://www.openstreetmap.org/copyright) and competition-specific research curation. Benchmark yard definitions, capacities, costs, and other parameters are research abstractions for algorithmic evaluation. Users take full responsibility and liability for their use of the materials. Full notice: DATA_PROVENANCE_AND_USE.md. No blanket data license is asserted; project-authored code is MIT (LICENSE-CODE).
Citation
Peiheng Li and Xuesong (Simon) Zhou. INFORMS RAS 2026 Problem Solving Competition. Kaggle, 2026.
https://www.kaggle.com/competitions/informs-ras-2026-problem-solving-competition
Users relying on upstream public source datasets should additionally cite FAF and OpenStreetMap as appropriate.
Checksums
SHA-256 hashes for the release asset and the in-repository archives are in the attached SHA256SUMS.txt (also committed at the repository root). Verify with:
sha256sum -c SHA256SUMS.txt