Sentinel QMS v4 · Ordinal 14 · Reproducibility

Cryptographic Reproducibility Across a Blinded, Multi-Entity Corpus

Ordinal 14 validates AuditLog.AI and Sentinel QMS v4 under a double-blind, HMAC-randomized challenge involving 7,239 unique files and 21,717 evidence-file instances. Reproducibility is no longer an opinion — it is a hash-level, Bitcoin-anchored property.

21,717Evidence instances100%Dual-hash parity40 / 40HMAC deletions detected0False positives0.24 s / fileHVT

View Ordinal 14 PDF View DOI / ResearchGate

Ordinal 14 is fully zero-custody: only cryptographic digests and Bitcoin payloads are published. No evidence, PHI, or trial data ever leaves the local node.

Stage IIIB double-blind HMAC challenge overview

Stage IIIB — Dual-Arm, Double-Blind HMAC Challenge

Stage IIIB combines a dual-arm design (Multi-Entity + AuditLog.AI) with a dual-operator, double-blind HMAC-randomized deletion challenge.

Arm 1 — Multi-Entity: 6,152 files across SENTINFRA, METAVAL, and corporate/trust entities.
Arm 2 — AuditLog.AI: 1,087 files across 16 historical regulatory runs (REGULATORY-RUN001–016).
Baselines (RUN010, RUN011) are blinded with randomized Bates IDs and cloned to USS positive controls (RUN014, RUN015).
A secret HMAC SALT randomly deletes 20 Bates-labelled files per arm into _HMAC_DELETED/ working copies (RUN012, RUN013).
All six runs are re-audited under a single deterministic compare session (RUN016_AMPLIFY_QMS_AUDIT).

Stage IIIB detection pattern

Baselines + USS controls: 14,478 files · 100% dual-hash parity.
HMAC tamper runs: 7,239 files · 40/40 deletions detected as digest mismatches.
Untouched files: 21,677 files · 0 false-positive mismatches.
All checks performed on blinded Bates IDs — no access to underlying content or trial systems.

For a deeper dive into the pipeline (dual-hash manifests, HMAC engine, QMS comparators, OP_RETURN), see the Technology page, and the Anchors page for OP_RETURN payloads and Ordinal lineage.

Human Verification Time (HVT) — Stage IIIB

Human verification was performed independently by two operators (Dr. Fernando Telles and Dr. Sam Francis). HVT is expressed as seconds per evidence file (Δ ÷ N).

Baseline + USS controls

781 s across 14,478 files → 0.05 s / file

HMAC tamper runs

4,477 s across 7,239 files → 0.62 s / file

Combined total

5,258 s across 21,717 files → 0.24 s / file

HVT quantifies the human work required to confirm deterministic QMS mismatches. In Stage IIIB, verification effort remains low even under blinded, randomized challenge conditions — and is itself cryptographically bound via timestamped HV START/END metadata.