Not clustered: 19. Jay Kubeil — only 11 of 19 metrics available (needs ≥ 60%); 23. Paul Lawrence — only 8 of 19 metrics available (needs ≥ 60%); 25. John Harriott — only 8 of 19 metrics available (needs ≥ 60%); 26. Jason Lannstrom — only 11 of 19 metrics available (needs ≥ 60%); 27. Adrian Connor — only 8 of 19 metrics available (needs ≥ 60%); 28. Ward Gunn — only 8 of 19 metrics available (needs ≥ 60%); 29. Marcus De Vecchi — only 10 of 19 metrics available (needs ≥ 60%); 30. Pete Bolton — only 8 of 19 metrics available (needs ≥ 60%); 31. James Atkinson — only 3 of 19 metrics available (needs ≥ 60%); 32. Bruce Atkinson — only 1 of 19 metrics available (needs ≥ 60%); 33. Keith Lavers — only 2 of 19 metrics available (needs ≥ 60%); 34. Tushar Pokle — only 1 of 19 metrics available (needs ≥ 60%).
GlideComp groups the pilots by flying style, and not by score. It transforms the rank of every behavioural metric to a percentile inside the field. It then compares two pilots by the mean percentile gap over the metrics that both pilots have, and never fills in a missing value. Ward-linkage agglomeration forms the groups, and the best mean silhouette selects the number of groups. Each group carries the spread of the GAP ranks of its members, which shows where a style paid and where it did not.
k was searched from 2 to 6. The mean silhouette is 0.22 — a value near 0 means soft group boundaries, and a value near 1 means tight, well-separated groups.