Stage V Pre-Registration and Stage IV Multi-Site Replication: Hash-Only Verification Packets for International Blinded Challenge Trial of Post-Export Clinical Trial Evidence Verification (WP73B Amendment)
Description
This repository contains two linked datasets supporting the Stage V international multi-site replication study (OSF WP73B Amendment). DATASET 1 – Stage V Pre-Registration: FICTI-PDA Challenge Corpus Hash-Only Packets Blinded hash-only verification packets for the Stage V FICTI-PDA challenge corpus: a synthetic, audit-ready Phase II clinical trial export bundle (FICTIROVAN vs placebo; n=70 simulated preterm neonates; no real patients). Packets are distributed for hash-only mathematical audit of the Stage V international blinded challenge trial. One labelled reference Tk (reference state) and five blinded corpora are provided (labelled QMS_01 through QMS_05), corresponding to: two negative controls (NC01, NC02), and three graded tamper tiers (T1 unintentional error, T2 mild dishonesty and selective omission, T3 overt fraud). Blinding maps (Stage_V_INDEX_BLIND_REVEAL_KEYS___20260606T231407Z.pdf) are pre-committed with hash proofs disclosed; maps are released post-reveal for the interpretive analyses. DATASET 2 – Stage IV Replication: QMS summary hash-packets (12 Arms, re-blinded) QMS summary hash-packets were extracted for the complete Stage IV study from source the original Nodes 01, 02, and 03, and re-blinded under the original 12 audit-arms structure. This dataset complements the prior Stage IV deposit (Telles, Fernando (2026), “Stage IV Audit Reproducibility Dataset (HVT-A): Deterministic Hash-Based Verification and Human Verification Time Outputs”, Mendeley Data, V1, doi: 10.17632/fzw4pzkd83.1), by enabling blinded, international, multi-site replication of detection performance, without disclosing any source data or identifiers. QMSAuditor v5 Professional software was used to enable hash summary extraction to further reduce the data size and volume (source data: 221,708.09 MB), which is sufficient for independent reproducibility using Mode A2 of the software. A new blinding map was applied in accordance to BLINDING_MAP_STAGEV_12ARM_PREHVT.md. Blinding maps are sealed pending blind reveal; SHA-256 of sealed map is disclosed. For of the 12-Arms, the reference corpus (A) is labelled; candidates are presented as QMSv5_01 and QMSv5_02. 01 – IV Baseline Reproducibility (Windows vs Windows): 2 arms 02 – IV Tamper Detection (Windows vs Windows): 2 arms 03 – IIIB Parity (IIIB macOS vs IV Windows): 4 arms 04 – IIIB Positive Controls (IIIB macOS vs IV Windows): 2 arms 05 – IIIA Baseline Reproducibility (IIIA macOS vs II macOS): 1 arm 06 – IIIA Tamper Detection (IIIA macOS vs IIIA macOS): 1 arm VERIFICATION SOFTWARE: DATASET 1 – QMSAuditor v5 Lite (open access, offline, no data collection): https://auditlog.ai/qms-auditor-lite DATASET 2 – QMSAuditor v5 Professional (Mode A2, required for Stage IV replication sub-study): contact corresponding author For instructions refer to OSF pre-registration WP73B. This dataset is released concurrently with amendment lodgement.
Files
Steps to reproduce
This dataset supports the Stage V international blinded challenge trial (OSF pre-registration WP73B Amendment). Verification is performed using QMSAuditor v5 software (AuditLog.AI) operating in offline, zero-custody mode. No source data or identifiers are transmitted at any stage. DATASET 1 – Stage V FICTI-PDA: 1. Download QMSAuditor v5 Lite (free, offline; Windows/macOS): https://auditlog.ai/qms-auditor-lite 2. Download the Tk reference packet and all five QMS_01–QMS_05 blinded packets from this repository. 3. Run five paired comparisons in QMSAuditor v5 Lite (Tk vs each blinded packet). The software generates a machine-readable HVT-A report per comparison. 4. Upload all five HVT-A reports to the designated public results repository using your pseudonymous site identifier as the folder name. 5. Blinding maps and tamper keys are released post blind reveal (after all sites have submitted). Full protocol: OSF pre-registration WP73B Amendment (released concurrently with this deposit). DATASET 2 – Stage IV Replication: 1. Download QMSAuditor v5 Pro (Mode A2). Contact: fernando.telles@auditlog.ai for access 2. Download the 12-arm hash packets from the Stage_IV_BLIND folder in this repository. 3. Refer to Dataset 2 for Stage V blinded classification arms and expected outcomes.md (included in this repository) for the reference/candidate structure per arm. 4. Run paired comparisons labelled reference A vs. blinded QMSv5_01 and QMSv5_02 for each of the 12 audit arms and upload resulting HVT-A reports. Participation enquiries and site registration: fernando.telles@auditlog.ai