Update README.md

README.md

@@ -23,9 +23,6 @@ in, one structured intent object out, across seven categories. The **1B** reache
 near-frontier accuracy at a fraction of the size and cost; the **0.5B** runs the
 same task at the lowest latency in the family.
 
-The work is in the data. We generate training conversations from a multi-agent
-marketplace simulation, with customers and assistants role-playing real payment
-interactions and adversarial actors mixed in, and we open-source it.
 
 **Models** · zeroproof-ecommerce-1b · zeroproof-ecommerce-0.5b
 **Dataset** · ecommerce-intent (train + eval)