In the latest beta deployment, nixpkgs (>1.065.000 commits, about 1300 of which have been found to include AI stuff) doesn't get an F anymore but rather "only" an E, because 1300 commits is basically nothing against the >1.063.000 untainted commits. I like this direction a lot c:
This maximum level of score reduction can only be reached by repositories where 0–2% of all commits are slop-assisted/made, so that is really only an edge case.
As it currently stands, here's a little visualization of taint% vs. score multiplier:
Taint% Score multiplier
0-2% 0.025x
0-5% 0.1x
5-10% 0.33x
10-15% 0.6x
15-20% 0.8x
20-30% 1x
30-40% 1.15x
40-50% 1.25x
50-55% 1.5x
55-60% 2x
60-100% 3x
Keep in mind that source signals (found a file named AGENTS.md, for example) and ignorefile signals (found CLAUDE.md in the .gitignore, for example) are not affected by this multiplier. They always count the same no matter what.
flori_ava_star🔜MRMCD:~
@star@amazonawaws.com