Nope. I ingested defects first.

Oops. You were asking if it changed stuff:

I just changed the order, ingesting the bias and flats first. This does not affect the directory structure in the db. I noticed that the biases and flats also used the same directory structure. But that never caused a problem.

I realized that I can't actually run with a current version of lsst_apps, because of a lsst.geom problem that I think go fixed? Figuring what I need to regit to be up to date to try to make sure we're working with the same stack.

No, you're probably right about the key point being that --mode=skip paths are blindly assumed to be relative to the calib directory.

lsst.geom got removed as a Stack package a few weeks ago (I can't pin down an exact date, but a Stack installed on May 16 doesn't have it). I'd recommend updating the Stack as a whole (through whichever update channel you have) rather than trying to cleverly pick individual packages.

I believe you've converged on what I (mostly) knew to be true: defect paths are assumed to be relative to the calib directory, so it may be simplest to cd into that directory before running calib ingestion, but it is not strictly necessary. The validity term has no effect for defects but cannot be left out.

I just added a note about this in the docs so we don't go around re-discovering it too many more times. Also relevant is DM-5467.

My issues were all addressed.