Why?

[numberZero]

This appears to be a weird self-modifying repository. What... are the reasons behind this?

Basically, I see several problems.

1. Storing benchmark results in a Git repository. Looks like abuse. No wait, this is abuse. Not big resource-wise but definitely a wrong way to use Git.
2. Keeping the harness separate of the primary repository. This is not a metadeployment collecting 100500 other repos, it only grabs two and with minetest#13034, would only need one.
3. Having code committing to a repository in that same repository. Follows from points 1 and 2 but even then, with these new shiny “fine-grained access tokens” that shouldn’t be necessary.

[rollerozxa]

1. Yes, this isn't necessary nowadays now that Github Actions is able to directly deploy to Github Pages. I elaborated on this in Deploy directly to Github Pages through Github Actions #5.
2. You can only deploy one Github Pages site per repository, and the main engine repo currently uses its Pages site to deploy the Lua API reference. It is admittedly a kludge but works fine enough.
3. As long as there is no unknown code that can be injected through unpinned dependencies, this is probably fine.