When Pooled Flexible Job Shop Scheduling Benchmarks Change the Recommendation: Verification and Reproducibility-Support Package

Description

This dataset accompanies the manuscript “When Pooled Flexible Job Shop Scheduling Benchmarks Change the Recommendation: Claim-Bounded Interpretation under Uncertainty.” It provides a manuscript-level verification and reproducibility-support package for a study of recommendation reliability in uncertainty-aware Flexible Job Shop Scheduling (FJSP) benchmarks. The package contains frozen derived analysis summaries, machine-readable audit records, selected provenance scripts, final manuscript reference tables and figures, SHA-256 checksums, and a package-local verifier. The claim-bearing evidence covers 18 family–subset–preset groups from the Brandimarte, DPdata, and Hurink benchmark families; nominal, light processing-time-noise, and medium processing-time-noise settings; and five transparent dispatching rules: ECT, FIFO/ECT, SPT, LPT, and MWKR. Sixteen groups use validated replication identifiers 0–9, while two Hurink edata groups use targeted 0–19 refinement. The package supports independent verification of the manuscript’s central reported results, including the pooled MWKR recommendation, the six group-level recommendation changes to ECT, exact unrounded flattened-choice penalties, materiality classification, the 6/18 materially worse pooled-choice proportion, bounded worst-tail summaries, replication-basis provenance, and valid-metric-count and comparator checks. The package is intended to verify the manuscript-level evidence summaries and final exhibits. It is not a full end-to-end rerun of the original instance-level experiments and does not include the complete raw experimental execution environment. The scope, file roles, provenance, exclusions, and verification procedure are documented in README.md, MANIFEST.md, and PROVENANCE.md.

Steps to reproduce

1. Download and extract submission_reproducibility_package_v1.0.zip while preserving the internal directory structure. 2. Open a terminal in the extracted submission_reproducibility_package directory. 3. Ensure that Python 3 is installed. The package-local verifier uses only the Python standard library and does not require additional packages. 4. Run: python scripts/verify_core_claims.py 5. A successful verification ends with: SUMMARY: all core manuscript checks passed The verifier checks the central manuscript-facing claims and audit records, including the number of represented groups, the pooled and group-bounded recommendations, the six changed groups, exact unrounded penalties, materiality classification, targeted-refinement provenance, valid-metric-count results, and the five-rule portfolio-relative comparator boundary. 6. The saved expected output is available in verification_report.txt. 7. File integrity can be checked against checksums.sha256. The checksum file contains SHA-256 hashes for every package file except the checksum file itself. 8. Consult README.md for package scope and usage, MANIFEST.md for the role and classification of each file, and PROVENANCE.md for the relationship between the frozen analysis summaries, targeted refinement, manuscript-facing mixed-basis results, and final reference exhibits. The package supports independent verification of frozen manuscript-level summaries and exhibits. It does not reproduce the complete original instance-level experimental campaign from raw inputs.

Institutions

• University of Pannonia

  Veszprém, Veszprém

Categories

Licence