Classify the overnight backlog

Every night you must label ~17 million documents — a sustained 200 requests/second — before the morning report. Nothing is waiting on a screen, so p95 latency up to 6 seconds is fine, and 74% label quality clears the bar for this triage pass. The board rewards raw throughput and the lowest cost per request. The lesson of this one: do not pay for a flagship model when a right-sized small model, batched hard with a warm cache, classifies the backlog for a fraction of the spend.

Loading the simulator…

The four axes are recomputed on our servers from the config you submit — the metrics you see are a faithful preview of the same model, so a forged score is impossible. Only players who choose a handle and opt in appear on the public board; we never show your email.