Data and Code for “Evidence-Governed Admissible State Search for Post-Quantum Cryptography Migration Planning in Networked Banking Systems”
Description
This dataset provides the reproducibility package for the manuscript “Evidence-Governed Admissible State Search for Post-Quantum Cryptography Migration Planning in Networked Banking Systems”. The package contains PQC-MigBench, a controlled publication-safe engineering-decision benchmark for post-quantum cryptography migration planning, together with migration-object records, abstract evidence records, rule descriptions, reference annotations, experimental scripts, baseline configurations, generated outputs, result summaries, LaTeX table files, and documentation. The benchmark contains 360 task instances expanded from 48 seed migration scenarios covering migration-priority assessment, TLS and API gateway upgrade planning, authentication and signature-chain migration, certificate-chain transition, HSM/KMS readiness evaluation, and stage-wise rollout planning. The evidence pool contains abstract evidence records representing standards, vendor statements, protocol descriptions, asset records, scan summaries, configuration records, performance-test records, and review notes. The files are sanitized and publication-safe; they do not contain production network topology, customer information, internal banking asset inventories, proprietary vendor documents, production rollback records, confidential audit logs, or undisclosed institutional rule bases. The package supports reproduction of the main benchmark comparison, stronger engineering baselines, deployability-gating evaluation, search diagnostics, budget and frontier-width sensitivity, no-search pipeline ablation, branch-sensitive diagnostics, hard ablation, stress tests, gate-causal field decoupling, expert-validation preparation, failure diagnostics, and representative LLM baseline protocol. The README file includes a reproduction map linking each manuscript table to the corresponding script, raw output file, CSV summary, and LaTeX table file. The dataset is intended for reproducible evaluation of evidence governance, deployability-gate control, unsafe-escalation prevention, audit-trace preservation, and stage-wise planning under controlled benchmark conditions. It should be used as a reproducibility artifact for the associated manuscript rather than as production banking deployment data.
Files
Steps to reproduce
1. Download and extract the reproducibility package. 2. Enter the package root: ```bash cd egass_final_repair_package/rebuilt ``` 3. Install the required Python dependencies: ```bash pip install -r requirements.txt ``` 4. Reproduce the main benchmark comparison and stronger engineering baselines: ```bash bash RUN_STRONGER_BASELINE_EXPERIMENTS.sh ``` 5. Reproduce search diagnostics and budget/frontier-width sensitivity: ```bash bash RUN_SEARCH_SENSITIVITY_EXPERIMENT.sh ``` 6. Reproduce the no-search pipeline ablation and branch-sensitive diagnostics: ```bash python src/run_no_search_ablation.py --data data/raw --output outputs/no_search_pipeline_ablation_outputs_branchdiag.jsonl --out results/no_search_pipeline_ablation_branchdiag ``` 7. Reproduce the gate-causal field decoupling test: ```bash python src/run_gate_causal_decoupling.py --mode both --data data/raw --output outputs/gate_causal_decoupling_outputs.jsonl --out results/gate_causal_decoupling ``` 8. Reproduce the hard-ablation study: ```bash python src/run_hard_ablation.py --output outputs/submission_hard_ablation_outputs.jsonl python src/run_evaluation.py --pred outputs/submission_hard_ablation_outputs.jsonl --out results/submission_hard_ablation python src/make_tables_and_figures.py --summary results/submission_hard_ablation/summary_metrics.csv --out results/submission_hard_ablation --fig figures/submission_hard_ablation ``` 9. Reproduce the stress-test results: ```bash python -u src/run_stress_test.py --out results/submission_stress --fig figures/submission_stress --max-tasks 360 ``` 10. Prepare and analyze the expert-validation casebook: ```bash python src/prepare_expert_validation.py --pred outputs/submission_method_outputs.jsonl --out results/expert_validation --cases-per-type 3 python src/analyze_expert_validation.py --ratings results/expert_validation/filled_reviewer_forms.csv --out results/expert_validation ``` 11. For the representative LLM baseline, use: ```bash bash RUN_REPRESENTATIVE_LLM_BASELINE.sh ``` This workflow prepares prompts and parses saved responses. Paper-facing LLM-baseline reproduction requires fixed model settings or saved responses as described in the README and protocol documentation. 12. See the README reproduction map for the exact relationship between each manuscript table, execution script, raw output file, CSV summary, and LaTeX table file.