AI Validation
Effective: February 7, 2026 | Last updated: 3/3/2026
Table of Contents
1. Validation Methodology
- Internal testing against expert-reviewed image datasets.
- Continuous model evaluation pipeline.
- Comparison against assessments by qualified building professionals.
- Regular re-validation whenever models are updated.
2. Performance by Category
| Defect Category | Detection Performance | Confidence Level |
|---|---|---|
| Structural Cracks | High | High |
| Moisture/Damp | High | Medium-High |
| Mold | Medium-High | Medium |
| Surface Damage | High | High |
| Thermal Issues | Medium | Medium |
Detailed quantitative metrics are available under NDA for enterprise clients.
3. Known Limitations
- Low-light or heavily shadowed images reduce accuracy.
- Concealed defects (behind walls, under flooring) cannot be detected.
- Unusual building materials may not match training data.
- Single-photo analysis is less reliable than multi-angle coverage.
- AI does not assess structural load capacity.
4. Continuous Improvement
- Model updates are tracked via a version manifest.
- Prompt A/B testing framework is active.
- User feedback is integrated into the training pipeline.
- Regular benchmarking against new OeNORM editions.
5. Request Full Report
- Enterprise clients can request the full validation study.
- Contact: info@faultrix.com
- Available under NDA.