MixedWM38 extends the WM-811K dataset with compound defect patterns — wafers labelled with two, three, or four co-occurring defect types simultaneously (e.g. Center+Edge-Ring+LOC). The dataset contains 38,015 wafer maps across 38 classes, encoded as 52×52 grids with 8 base defect types combined via multi-hot labels.

One wafer per class is loaded directly into buildWaferMap and analyzeWaferMap with no special configuration. Each card names the dataset's class, then what analyzeWaferMap found on it: the kinds of findings it raised (ring, quadrant, sector, cluster, edge arc) and, when one shape dominates, the spatial pattern it classified. A compound map is reported the way wafermap reports any wafer, as several findings, each with the dies it covers: expand a card and click one to see them. The pattern classifier names a single shape, so on a compound map it describes the dominant part at most; Detection Analysis measures it on the single-pattern WM-811K set it is designed for.