Synthetic generation and its traps

Generator lab: L3.1 leaves 14 starved cells — 11 holding exactly one case and 3 holding nothing — so a perturbation generator is pointed at them. Its 50 candidates touch only 11 of the 14, because a perturbation can only vary a cell that had a seed: 30 land in the 7 deep cells and 0 in the 3 empty ones. Eleven candidates are repaired by their own perturbation and keep the seed's failing label; 7 of those collide with a real case key, so the 4 that survive dedup would have shipped. A reply-only difficulty screen keeps 14 of 50 and deletes 3 genuine failures it cannot read, while checking the label first keeps 17 and fills 6 cells. Fourteen written cases fill all fourteen.

This Evals Engineering lesson is part of a paid plan. Sign in with a plan that includes this course to read it.