LLM Usage Policy isn't being adhered to

You are aware that the usage is from a guy that can stay at homel and code all this right? His job is Fluxer His job is the founder of fluxer His name is Hampus Obviously the dude is VERY knowledgeable in the coding he is doing. And to top if off... there is 3 devs on this project. So what makes you really think that there is AI usage in this? image

You do know that even though he pushed 1 million additions to the repository—it doesn't mean it's all code, right?

Hi, I'm a dev who writes code by hand and contributes to Fluxer. See https://github.com/fluxerapp/fluxer/pull/1129 ? It had 20k added lines of code in about 2 weeks of intermittent work, and that's me not working full time with some significant amount of time spent reading the codebase I wasn't familiar with. That diff is almost twice as big as my immediate-mode Rust GUI framework that I wrote over the course of more than a year. Turns out, Fluxer's codebase architecture just leads to big diffs, is all. I think that instead of just looking at the number of lines of code added, you should take into consideration the amount of these lines that come from generated files. Notably, all of the locales, which use a file format that points back to places in the source code and therefore are susceptible to any source code changes, and the massive OpenAPI JSON files, to cite the biggest culprits. I'd also note that GitHub seems to use a weird diff algorithm. When Hampus started committing to the repo again a few months ago, there was one huge commit that showed 10 million lines of code added, yet when cloning the repo and looking at that commit locally, it was only 3 million, and 2/3rds of that were from automatically generated files. (I analyzed that whole commit back then.) I'm not saying Hampus isn't currently using LLMs to generate anything (in fact first time translations are generated with AI and that's disclosed in the application), only that you can't just look at the number of lines of code on GitHub and judge based on that.

The 1 million figure comes from GitHub's contributors graph, which counts every line a commit adds. It doesn't subtract deletions, and a line rewritten five times is counted five times. About 967k lines were deleted over the same weeks, so the repository grew by about 112k lines. Close to half of the additions are the 34 translation catalogs, which a tool rewrites across every language whenever the source changes. #2750 was a translation review, and it alone is 192k of those lines, 91% of them generated catalog lines. Most of the rest is OpenAPI schemas regenerated by CI, lockfiles, the docs site migration, and files that were added and removed within the same window. Line counts aren't evidence of LLM use, and the policy says you can't publicly accuse people over suspected use (section 10). Rolling main back is a ridiculous request and it's not going to happen. It would throw away nearly a thousand commits, including hundreds of fixes already running in production and work from outside contributors, and you're asking for it without having looked at a single line of the code in question. If you find a change that's wrong, open an issue that points at it.