When we originally implemented the check for unused artifacts in our CI, we applied it to a diff between the PR and main to make sure that any newly added entries weren't missed, and because we anticipated retiring artifacts from the main papers/presentations page over time to keep them from getting too big (as they arguably have). Since then, we've introduced chronological archives of papers and presentations, so could rewrite this check to consider the whole file rather than just the diff.
This would have some potential benefits such as making it easier to run from the command-line while developing a diff. It also means that if the artifacts were changed outside of the PR process, they could still be caught in the future. I think the main downside might be slightly more redundant work, but I'm imagining the check is fast enough for our data sizes that it would be negligible.
When we originally implemented the check for unused artifacts in our CI, we applied it to a diff between the PR and main to make sure that any newly added entries weren't missed, and because we anticipated retiring artifacts from the main papers/presentations page over time to keep them from getting too big (as they arguably have). Since then, we've introduced chronological archives of papers and presentations, so could rewrite this check to consider the whole file rather than just the diff.
This would have some potential benefits such as making it easier to run from the command-line while developing a diff. It also means that if the artifacts were changed outside of the PR process, they could still be caught in the future. I think the main downside might be slightly more redundant work, but I'm imagining the check is fast enough for our data sizes that it would be negligible.