LeoJiangOR/FrontierOR-Audited-92

FrontierOR Audited 92

This is a derived, evaluator-oriented release of 92 cases from SmartOR/FrontierOR, pinned to upstream dataset commit 37ccd8b6dca3bf7f4e0c58941a6ed156832a6d9e.

The original 180-case release was audited for coherence between the problem description/formulation, reference Gurobi implementation, solution schema, bundled solutions, and feasibility checker. This release contains the 92 cases that were not assigned BLOCK_RELEASE; 91 checkers were hardened and the one previously clean case was brought under the same explicit release contract.

What changed

• The 92 feasibility_check.py files now enforce stable solution fields, finite objectives, non-erased witnesses, derived-field consistency where applicable, and structured fail-closed behavior.
• adulyasak2015 validates the submitted first-stage decision by solving its fixed continuous recourse problem, matching the trusted evaluator behavior.
• A hash-bound release manifest and complete audit evidence are included.

No problem description, mathematical formulation, Gurobi model, instance, bundled solution, or bundled feasibility result was changed relative to the pinned upstream commit. Files with upstream backup markers such as .pre_* and .stale_*, together with six unbound zero-byte solver-log placeholders, are not part of this clean release view. One zero-byte historical log is retained because its hash is explicitly bound by the release manifest; it is not represented as an optimality certificate.

Verification status

Gate	Result
Per-case release status	92/92 EVAL_READY
Checker mutation audit	92/92 clear
Evaluator objective preflight	92/92 passed
Current tiny Gurobi models produced solutions	92/92
Current tiny outputs accepted	92/92
Bundled instance/solution replay	552/552 accepted
Manifest hashes	PASS

The 552 bundled solutions are finite, checker-replayed feasible evaluation baselines. They must not all be described as zero-gap global optima:

• 228 solver-optimal;
• 41 solver-optimal within solver tolerance;
• 199 explicit incumbents;
• 84 objectives without an optimality certificate.

Therefore, 283 references across 66 cases do not carry a global-optimality certificate. This is retained as metadata rather than silently upgraded.

Repository layout

Each case directory contains its description, mathematical formulation, schemas, reference Gurobi code, repaired feasibility checker, instances, bundled solutions, available solver logs, and bundled checker results.

• release_subset_92.json: selected cases, checker contracts, reference status, and SHA-256 bindings for the release-critical artifacts.
• audit/REPAIR_92/: second-stage machine-readable verification logs.
• audit/FrontierOR_Audit_Bundle_20260829.zip: complete phase-1 and phase-2 audit evidence, including 84 confirmed-issue reviews for the original release.
• PUBLISH_FILE_MANIFEST.json and SHA256SUMS: complete published-file inventory.

The repository uses a file-tree layout because every OR task is a self-contained reproducible package. Metadata can be loaded with:

from datasets import load_dataset

meta = load_dataset("LeoJiangOR/FrontierOR-Audited-92", "meta", split="train")
print(meta[0])

For the complete task files:

hf download LeoJiangOR/FrontierOR-Audited-92 --repo-type dataset --local-dir frontieror-audited-92

Provenance and license

• Upstream dataset: https://huggingface.co/datasets/SmartOR/FrontierOR
• Upstream dataset commit: 37ccd8b6dca3bf7f4e0c58941a6ed156832a6d9e
• Evaluator commit used by the audit: 93fb106d98c9d21729aa846f29822e12ad22db90
• Audit date: 2026-08-29 (America/New_York)

The upstream dataset metadata declares CC-BY-4.0; this derived release retains that license metadata and attribution. See LICENSE.md and UPSTREAM_README.md.