Validation Evidence

All claims made by VLTHRLAB, with evidence, verification status, and sources. This page consolidates validation data from our research programs.

VERIFIED
AUC up to 1.000 on 8 public datasets
Experimental result on 8 public benchmark datasets for industrial anomaly detection
Source: vlthrlab.app/research/mad-system.html | Date: 2026
VERIFIED
5-24 days early detection before failure
Detection window measured across 8 datasets
Source: vlthrlab.app/research/mad-system.html | Date: 2026
VERIFIED
Zero false alarms on cross-dataset transfer
Cross-dataset transfer test: trained on one equipment type, tested on another. AUC 0.9995 with zero false positives.
Source: vlthrlab.app/research/mad-system.html | Date: 2026
VERIFIED
72% downtime reduction in digital twin simulation
Digital twin simulation comparing predictive vs run-to-failure maintenance strategies
Source: vlthrlab.app/research/mad-system.html | Date: 2026
VERIFIED
6x lower maintenance costs in simulation
Cost comparison in digital twin simulation
Source: vlthrlab.app/research/mad-system.html | Date: 2026
VERIFIED
Break-even at 5% detection rate
Economic analysis: only need to prevent 1 in 20 failures to recover implementation cost
Source: vlthrlab.app/research/mad-system.html | Date: 2026
VERIFIED
$1M emergency event reduced to $144K scheduled intervention (86% cost reduction)
Cost model comparing emergency repair vs scheduled intervention
Source: vlthrlab.app/research/mad-system.html | Date: 2026
VERIFIED
40 hours downtime saved per event
Downtime comparison between emergency and scheduled maintenance
Source: vlthrlab.app/research/mad-system.html | Date: 2026
SUPPORTED
$5,000-$25,000/hour deferred production cost on offshore platforms
Industry data on offshore platform production costs
Source: Industry data | Date: 2026
SUPPORTED
222 grid collapses since 2010 in Nigeria
Nigerian power sector reports and public grid incident data
Source: Nigerian power sector reports | Date: 2026
SUPPORTED
110ms retrieval in presentation assistant
System benchmark for dashboard query retrieval
Source: vlthrlab.app/research/presentation-assistant.html | Date: 2026
SUPPORTED
4s explanation in presentation assistant
System benchmark for AI-generated explanation
Source: vlthrlab.app/research/presentation-assistant.html | Date: 2026

Source Hierarchy

  1. Level 1 — First-party primary data (experimental results)
  2. Level 2 — Official documentation and records
  3. Level 3 — Independent authoritative sources
  4. Level 4 — Verified customer evidence
  5. Level 5 — Reputable publications
  6. Level 6 — Community evidence
  7. Level 7 — Unverified claims