External photo datasets for stress-testing the classification engine. These photos are unlabeled — Claude Vision describes each photo, then the rules engine classifies based on those descriptions. No human labels needed.
| Source | Type | Photos | Description | Access |
|---|---|---|---|---|
| Roboflow Home Inspection | Inspection | 471 | Water heater, electrical, roof, plumbing, cracks — real inspection photos with component annotations | Roboflow SDK |
| Pexels API | Stock | 500 - 2,000+ | Search by component: "electrical panel", "water heater", "roof shingles", etc. High quality stock photos | Free API key |
| GitHub Houses Dataset | Rooms | 2,140 | Bedroom, bathroom, kitchen, exterior views — good for room-type classification testing | git clone |
| GitHub BD3 Defects | Defects | 3,965 | Wall cracks, spalling, peeling, stains — building deficiency detection dataset | git clone |
| Kaggle House Rooms | Rooms | 25,000+ | Room types + exterior views — large-scale dataset for classification benchmarking | Kaggle CLI |