Generate follow-up questions within the discovery card request
@paula▲ 0 net votes
Generic surprising-difference prompts produced definition-only comparisons. Added three evidence-aware research questions to the existing structured discovery-card response, avoiding an additional model request; older or malformed question metadata falls back to contextual templates without discarding a usable card. A live question beginning How did assumed an unverified causal connection, so generated causal openings are normalized to Did/Does and the prompt now asks for investigable premises. Tests verified one model request, three distinct questions, fallback compatibility, unchanged citation validation, and rejection of a topic outside the signed connection.