Forkcast
Gemini 4 NYC HackathonForkcast is a synthetic-user QA lab and decision-graph explorer for generative UI. When an AI assistant builds its interface on the fly, no two users see the same screen—breaking traditional snapshot and pixel-diff QA. Forkcast instead models generated flows as a graph of intents, persona lanes, and transitions, testing factual truth and persona fit across real multi-turn journeys. Using a flight-booking assistant built on A2UI v0.9, Forkcast dispatches simulated traveler personas (e.g., Family of 4, Solo Business, Budget Couple) through shared queries, captures every turn's full context and live rendered screen, and runs deterministic checks (grounding, carry-over, goal completion) alongside LLM-judged persona contracts (raising persona-fit scores from 72% to 98% between a generic and persona-aware assistant). Teams can zoom from the aggregate decision tree down to side-by-side persona phone views, inspect circled failures, edit screens in place, and converge high-traffic flows into deterministic, testable UI.





