Reproducibility archive for “Identifiability and Exact Recovery of Finite Programs from Compositional Final Labels”

Description

Reproducibility materials accompanying the manuscript “Identifiability and Exact Recovery of Finite Programs from Compositional Final Labels” by Xingjian Zhang, submitted to Information Sciences. The archive contains the implementations of finite-table executors and comparator models; frozen experiment configurations and random seeds; deterministic benchmark generators and included data arrays; trained checkpoints and predictions; per-run metrics and complete run ledgers; SMT identifiability certificates and direct-solving records; tests; and scripts, numerical sources, and LaTeX sources used to produce the manuscript and supplementary document. The benchmarks study modular arithmetic and sampled finite algebras, including shortcut-free compositional supervision, zero-direct-label recovery, solver certificates, and length generalization. Benchmark labels are computed by deterministic program semantics. Some large data arrays are regenerated from the included generators, configurations, and seeds rather than stored in the archive. File: Reproducibility_archive.zip. Extract this ZIP to obtain the archive/ directory. Its README.md describes the software environment and reproduction commands. ARTIFACT_MANIFEST.sha256 records checksums for the archived research files. Directory contents: src/ implements models, benchmark generation, solvers, and verification; configs/ contains experiment settings; data/ contains benchmark artifacts and manifests; results/ contains run records, checkpoints, predictions, and summaries; scripts/ contains integrity checks and analysis entry points; paper/ contains manuscript sources, generated tables, figures, and figure-generation scripts; tests/ contains automated tests. The supplementary PDF is submitted separately with the manuscript.

Steps to reproduce

1. Extract Reproducibility_archive.zip and open the resulting archive/ directory. 2. Read README.md for environment information and experiment-specific entry points. 3. Check file integrity with: python scripts/verify_artifact.py 4. Install the package and test dependencies with: python -m pip install -e ".[test]" 5. Run the tests with: python -m pytest -q 6. Audit stored results with: python paper/audit_repository_results.py 7. Consult README.md for the appropriate benchmark-generation, training, checkpoint-replay, solver, ledger-building, and figure-generation commands. Preserve the included configurations and random seeds when regenerating omitted arrays or repeating experiments. GPU training and solver runs may require substantial time; full retraining is separate from archive-integrity checking.

Institutions

• Carnegie Mellon University

  Pennsylvania, Pittsburgh

Categories

Licence