WM-811K is a public dataset of 811,457 wafer maps from real fabrication lines, each labelled by hand with one of nine defect categories (edge ring, scratch, donut, local cluster, and so on). The 27 maps below, three per category, are loaded into buildWaferMap and analyzeWaferMap with no special configuration. Each card names the dataset's label, then the pattern the library's classifier found on its own (stats.spatialPattern), with ✓ where they agree. Expand a card for its findings.

The dataset has no class for a local cluster, so Local cluster wafers are reported by what their cluster looks like, and a cluster on any wafer is still found as a cluster finding. Detection Analysis measures the classifier over the whole labelled set rather than these 27 maps.