Vision applications often reveal complexity with volume. Lessons arrive around tens of parts, then hundreds, then thousands. The interval between major discoveries stretches, but discovery does not end at installation. A lighting angle that was stable in a lab becomes unstable when new packaging stock changes specular behavior. A font that separated similar characters drifts after printer maintenance. These later lessons are improvement work, not proof the original job was careless. The practical implication is documentation. Training, maintenance notes, and saved boundary images let the next engineer recognize a known mode instead of retuning a limit that was already studied. Vision discovery simply lasts longer than it does for instruments that do not look at the part.