Stage IV Audit Reproducibility Dataset (HVT-A): Deterministic Hash-Based Verification and Human Verification Time Outputs
Description
This dataset contains the complete, analysis-ready outputs used to generate all reproducible statistical results reported in the Stage IV blinded audit validation study. It includes hash-only verification summaries, human verification time (HVT-A) artefacts, extracted comparison tables, and verification certificates required to independently reproduce the manuscript's tables and figures. No raw evidence files are included. All data were generated deterministically from pre-anchored audit executions and are provided to enable independent statistical recomputation and verification of reported results.
Files
Steps to reproduce
## Execution Overview This repository supports deterministic reproduction of all reported Stage IV (HVT-A) analyses using released hash-only verification artefacts and disclosed analysis scripts. Complete instructions are contained in REPRODUCIBILITY.md. Two reproduction modes are provided. --- ## Mode A: Analysis from Released Tables (Recommended) This is the fastest and recommended reproduction pathway. Use this mode to regenerate reported tables and figures presented in the manuscript, without re-running Python extractors. Steps: 1. Download the repository release. 2. Confirm that the following directories are present: • __EXTRACT_OUT__/ — analysis-ready CSV tables • __VERIFY_READY__/ — verification and consistency certificates (JSON/NDJSON) 3. Run the R entrypoint script: R script RUN_ALL_HVT_A_FINAL.R ### Deterministic Extraction scripts (Python) https://github.com/auditlogai/evidence-verification/tree/main/src/python ### Statistical Analyses scripts (R) https://github.com/auditlogai/evidence-verification/tree/main/src/R --- ## Mode B: Full Regeneration from Hash-Only Artefacts This mode performs a full, end-to-end regeneration of analysis-ready tables from released QMS hash-only summary outputs. Use this mode if you want to independently verify: • extractor correctness, • verification logic, and • analysis-readiness checks prior to statistical modeling. Steps: 1. Run the Python extractors to regenerate __EXTRACT_OUT__/ from hash-only artefacts. 2. Run the Python verification scripts to populate __VERIFY_READY__/. 3. Run the R entrypoint script as in Mode A. --- ## Notes on Reproducibility Scope • All analyses reported in the main manuscript are reproducible using Mode A or Mode B, except Manuscript Table 4, which requires direct access to execution environments (Nodes 01–03) and is explicitly documented as non-reproducible from disclosed hash-only exports. • Detailed, step-by-step instructions (directory layout, command sequences, and environment requirements) are provided in REPRODUCIBILITY.md. ---