The JIVISION assembly vision case revision record dated 2026-08-30 retains the stage results from that time. Under that project's verification data and configuration, the anomaly candidate count changed from 126 → 8. This describes a candidate count change after an existing processing path revision. It is not a general accuracy rate or a production quality metric.

Describe reference-image results for the actual samples

In the original record, 3 reference images each had 0 scratch candidates, and the on-site scratch sample produced a nonconforming result. These findings apply only to the images, hardware, parameters and software version used then. Reusing those reference images does not increase coverage of independent normal workpieces or defect samples.

Verify production rollout separately

The original public material retains requirements for independent production samples, metric definitions, continuous operation and acceptance of interfaces at the production site. Later work can organize samples by workpiece batch, defect, pose and imaging conditions, linking individual results to configurations. This is a later verification plan. It does not state that these tasks have been performed.

This review adds no new test conclusions

This review checks only the original publication record and public case. It adds no device tests, sample results or production acceptance. After later algorithm, reference-image or parameter changes, preserve earlier versions and results and verify the affected scope again. Update public descriptions when new evidence is available.

See the original assembly vision inspection case.