The AI assistant's entity chips — the interactive references to cultures and other lab entities embedded in responses — now display correctly throughout the full conversation.
Two related issues were fixed: chips were dropped at the boundary of the final response segment during multi-block rendering, and on plates with large numbers of observations, entity selection was capped in a way that caused some cited cultures to render as raw identifiers depending on payload traversal order. Chips now precisely reflect what the assistant actually answered, even on observation-heavy plates with large reference payloads.
