Research Chat

The plan on the card when the user clicked Run plan.

The fields are what runs, as the card holds them (planning.libs.ts::confirmedPlan); text is the plan as the user read it, which the model is shown. The server used to parse the text a second time, in Python, to decide what to run, and that parse disagreed with the card's twice (planIntents.parity.test.ts): the user got a different search from the one on screen, with no error. plan_from_fields normalises these exactly as it does a parsed plan text.