Mirrors a vote into LangSmith as feedback (user_score, 1/0) on the traced
run (question span when targeted, else root). Corrections are sent twice on
purpose: the structured correction field seeds eval datasets, the comment
copy is all the trace view renders.
Fire-and-forget and NEVER throws — a LangSmith failure must not affect the
user's submission. On success, persists the resolved run/feedback ids back
onto the ai_feedback row.
Mirrors a vote into LangSmith as feedback (
user_score, 1/0) on the traced run (question span when targeted, else root). Corrections are sent twice on purpose: the structuredcorrectionfield seeds eval datasets, the comment copy is all the trace view renders. Fire-and-forget and NEVER throws — a LangSmith failure must not affect the user's submission. On success, persists the resolved run/feedback ids back onto the ai_feedback row.