Agent success falls from 100% to 32% as restoration depth goes 1 to 8.

ProgramDistill breaks the convention of handing an agent a written issue. It factors 26 fully functional web applications into features, mines 1,975 replay-verified behaviors, and auto-constructs 4,063 tasks with no human labeling, so the agent discovers the target behavior by driving the reference app. Across nine frontier agents, GPT-6 Astra and Claude Opus 5 reach 49.2% and 28.8% on cumulative workflows in full reconstruction; in partial reconstruction success drops from 100% to 64.0% and from 96% to 32% as depth rises from 1 to 8. The depth curve is the builder-usable number. Reliability collapses with the count of interdependent missing pieces, which argues for slicing agent work into shallow independently verifiable units.