How BAO Grades Cards: Evidence, Confidence and Model Updates
From your camera to an evidence report
You submit original front and back photographs through your free account. BAO checks the images and evaluates the visible condition evidence. The report separates centering, corners, edges and surface, and provides confidence and capture guidance. Reports and original grading photographs are stored privately in your account so that you can revisit the evidence across devices.
Geometry before grading
Camera angle can distort printed borders. BAO attempts to detect the physical card plane and correct perspective before measuring centering. Incorrect corner detection or an uncertain printed boundary can still limit the measurement. Cropped edges, blurred detail and glare may require another photograph. The system cannot infer a reliable physical measurement from evidence the camera did not capture.
Visible wear and surface limitations
Corners and edges are evaluated from available photographic evidence. Surface assessment is especially sensitive to lighting, sleeves, slabs and foil reflections. Directional photographs can provide additional evidence when alignment and lighting checks pass. BAO should explain a limitation or request a recapture when the image cannot support a useful assessment. A provisional grade is not an authenticity certificate.
What the number means
The BAO scale runs from 1 to 10 and is intended for photographic pre-grading. Read its subgrades, confidence and evidence alongside the overall estimate. The deployed pipeline combines image analysis with versioned scoring rules; a professional-grade-equivalent model has not been established. Reference-comparison and experimental model channels only affect released grading behavior after their validation and promotion requirements are met.
Every new scan supports reviewed development
Free access supports data collection for model improvement. After acceptance of the required terms, future successful grading scans enter a private review queue with photographic evidence and model-version information. This is a photo-use license, not a transfer of ownership of your card or artwork. Earlier scans are not retroactively enrolled under new terms. Account export and deletion controls remain available.
Validation before model releases
Predictions are not ground-truth labels. Reviewed examples need appropriate rights and independent annotations before training. Evaluation must keep related captures of the same physical card together to avoid testing on near-duplicates of training images. Frozen benchmarks and promotion gates help compare model versions before a release. BAO does not automatically retrain after each scan or claim that every individual scan improves accuracy.