FrontierHarness Eval – Visual ML Pipeline Editor

The Exo harness is the most spammed category of all spammed categories. Could this be used to perform some sort of ablation test too. E.g. what parts of Codex contribute most to the perf, and can they be recreated more minimally in something like Pi. What do you expect? "Best X" is the most interesting one in the benchmark, as it has the potential to complete tasks with fewer turns and cost.