Hi Team!
Hitting a wall with online evals and hoping someone's seen this. I've got an LLM-as-judge eval configured as Run continuously on new data — active, recurring, 100% sampling — but its run log is completely empty; it has never executed a single live run. The exact same evaluator with the exact same filters and column mappings, run as a One-Time Backfill, scores everything perfectly. So the evaluator, filters, and mappings are all correct — the continuous scheduler just never fires. Seeing it in two separate projects, so it's not project-specific. One clue: it's a trace-scope (multi-span-query) eval, and when I switch to a span-scope eval on the same traces, that does run live — so it looks specifically like trace-scope continuous evals never get queued (possibly waiting on a "trace finished" signal that never comes). Is continuous trace-scope eval actually supported/enabled for our space, or is there a setting or gate I'm missing?
