Last week, our agent traced a flaky test to a dependency bump three weeks prior. It did this by writing its own SQL queries, scanning hundreds of millions of log lines across a dozen queries, and following a trail from job metadata to raw log output. The whole investigation took seconds. To do this, the agent needs context: not one log file, but every build, every test, every log line, across months of history. Every week, about 1.5 billion CI log lines and 700K jobs flow through our system. All...