Most capture work is RPA bots, driven by AI agents, clicking through the source application's UI. Fix Jobs are the exception: jobs with no source-app interaction at all. They exist for two situations — data that was captured incorrectly, and after-the-fact requirement changes that don't justify reopening the application. The classic example is a formatting correction, like turning a code '123-456' into '123456'. They matter operationally because a full RPA rerun is very slow, so in the workflow Fix Jobs run after the primary capture pass to patch a result rather than regenerate it. It's a small piece of the factory, but it's what lets us stay responsive to change long after the bots have moved on.