Every player-season sits in a neighborhood of statistical twins — computed from the 64-d embedding. Pick any player-season and see theirs.
Loading the twin data…
Nodes are player-seasons; edge width is proportional to cosine similarity in the L2-normalized MTNN v5 64-d embedding (the player's own seasons excluded, top 8 shown). Twin Centrality = how many of the 12,966 player-seasons count this season among their 10 nearest neighbors. Era-Bridge Score = share of the 8 shown twins from a different decade. Every number computed from the 64-d embedding.
Four stages, one story: the era it belongs to, the archetype it plays like, the three stats that define it, and the twin it mirrors across time.
Flow widths are proportional to |z-score| share of the three defining stats in the 14-dim era-honest style vector. The era twin is the nearest other-decade career in the 64-d embedding. Twin coverage spans careers with 4+ charted seasons — seasons outside the pool show an honest empty state, never a placeholder.
Methods and provenance: methods.html#insights · Every figure is computed from the served vectors, never invented.