Comment by terekhindc

neat. when you say production-scale traces don't fit in context — does halo cluster failures first or just stream-summarize each trace?