The Style Graph is Elara's core personalization layer. Where most recommendation systems learn from what shoppers purchase, the Style Graph learns from the full interaction — including what shoppers skip. A shopper who consistently passes over floral prints has told the system something precise about their aesthetic. That negative signal is weighted alongside positive signals like purchases and likes. The model is built per shopper and per brand. A shopper's Style Graph on Brand A is entirely separate from their Style Graph on Brand B — their taste data stays private to each brand relationship. The exception is the Elara iOS consumer app: shoppers who use the app build a cross-brand taste profile they can optionally share with brands they visit, making a first brand visit feel like a fourth. The Style Graph compounds. A single session with a complete outfit interaction gives the model enough signal to be useful on the next visit. By the third or fourth session, the model produces recommendations meaningfully better than a cold start. This compounding is why the Style Graph is described as a moat — taste data built up over months cannot be replicated by a competitor who enters a brand's store later.