Collect execution telemetry from agents and use it to build evaluation harnesses.

Soon you can unlock how this was done.

Behind this: the tool used · the method · what actually resulted.