SF-Relate excludes spurious fourth-degree relatives detected by KING. (A) We show confusion matrices assessing the relatedness degree classification accuracy of KING (left) and SF-Relate (right), comparing with the output of PC-Relate as the ground truth. SF-Relate is performed in plaintext (i.e., without MHE), focusing on the evaluation of the bucket assignment. Unlike SF-Relate, KING classifies many unrelated samples as fourth-degree relatives. Most of these pairs involve the same outlier sample, which has many spurious relationships. (B) We verify that pairs involving the outlier do not exhibit IBD sharing patterns (left), evident in fourth-degree pairs from PC-Relate (right). Four example pairs are shown for both cases. For each pair, we compute the Hamming similarity between the two samples of genomic segments across the genome. Bright yellow bands represent likely IBD segments. The locations of the bands are randomly permuted to obscure their positions.