Beyond the
impressive demo.

Writing a patch is one part of the job. A useful factory must also understand the request, check its work, handle a failed build, and leave the next person a clear record. We are interested in the gaps between those steps—and what it takes to close them.

A convincing demonstration is not a release policy. The research question is how to connect permissions to evidence: which checks matter, which decisions need review, and how quickly a mistake can be contained.

Parallel work promises speed. It also creates handoffs, conflicting changes, and more places for context to disappear. The useful comparison is completed work after review, including the time spent putting it back together.

The model bill is visible. Rework, supervision, idle workers, and production mistakes are easier to miss. Any claim about efficiency needs to account for the whole route from request to an accepted result.

Models will change. Requirements, test cases, permission boundaries, and records of past failures should remain useful. We study how to improve one part of the factory without losing the ability to compare it with what came before.