Hi all — first post.

Hi all — first post. I'm building an internal tool that triages incoming customer messages. A few thousand a month today, and I want each one tagged by category and urgency, with low-confidence cases flagged for human review. The blocker with our current LLM call is cost and latency at volume, plus the occasional malformed output breaking the pipeline. Calibrated confidence is the part I most want to test — I need a threshold I can actually trust before letting it run unattended. Still on the waitlist (filled the extra info form). Will post real numbers here once I get access. Anyone using Choice + Score together for triage? Curious how you set the confidence cutoff.