Closing the gap between
lab accuracy and field accuracy.

A team had trained a computer vision model to detect defects on a production line, validated against a held-out test set at 94% accuracy. When the model moved to the actual camera and lighting setup on the floor, accuracy dropped to roughly 61% — good enough to erode trust in the system, not good enough to rely on.

The gap wasn't a flaw in the model architecture. It was a mismatch between training data and deployment conditions. The training set was clean, well-lit studio images. The production environment had variable lighting across shifts, motion blur from the line's speed, and lens distortion from the mounted camera angle — none of which the model had ever seen during training.

The fix wasn't a bigger model. It was closing the gap between training conditions and deployment conditions:

• Collected real field data from the actual production camera and lighting setup, across different shifts and times of day, instead of relying solely on the clean studio dataset.
• Augmented training data with synthetic lighting variance, motion blur, and lens distortion matched to the real hardware, to extend coverage beyond what field data alone could capture quickly.
• Re-validated on held-out field footage, not just the original studio test set, so the reported accuracy number reflected what the model would actually see in production.
• Built a feedback loop for flagging low-confidence predictions on the line, so edge cases could be collected and folded into future training rounds instead of silently degrading accuracy over time.

This is a pattern-level breakdown of what our vision AI and edge deployment work is built to prevent. If a model's production accuracy is quietly worse than its test-set accuracy, it's almost always a training-versus-deployment mismatch — not a model problem.