Artifacts
What each model built on each task, shown on the held-out cases it was graded on.
Streaming ETL load ramp
Each model built a pipeline that keeps payment tables and a fraud-alert stream fresh from a database's change stream while the input climbs 1.25× every 7 minutes from 1,000 events/s. How far each attempt's graded run, on the hidden test load and its faults, got before an output's p99 freshness broke its limit, and why it stopped.