Add LLM benchmarking proposal.

ideas/2026.md

@@ -300,3 +300,42 @@ Let's make more use of the Ryan(Tm) update bot!
 - add updateScripts
 - add tests to existing scripts that are failing
 - create a metric to analyze existing script pass/fail/update-time
+
+## Generative Nix: Surveying LLM Proficiency In NixOS
+
+Effort: small (90 hours)
+
+LLMs have gone from toys to industry-standard tooling but their ability to program or debug Nix hasn't really been studied; some folks have good luck with them and some don't, and so it'd be helpful to know where support for Nix in these LLMs stands.
+
+For this project, the mentee would:
+- Select a handful of commercial and open-source state-of-the-art LLMs
+- Select a handful of representative tasks using Nix, Nixpkgs, and NixOS
+- Select performance criteria
+- Conduct experiments to benchmark performance of each LLM against each task
+
+Deliverables:
+- Selection of benchmark tasks and criteria for LLM usage on Nix/NixOS/Nixpkgs
+- Blog post describing results of commerical and open-source state-of-the-art LLMs on those tests
+
+Skills required:
+- Basic experience with LLMs
+- Basic scripting knowledge
+- Access to commercial LLMs and open-source LLMs
+- Ability to design and conduct experiments
+- Ability to write and communicate clearly
+
+Skills suggested:
+- Basic knowledge of Nix
+- Basic knowledge of NixOS systems administration
+- Basic knowledge of Nixpkgs
+
+Possible Mentors:
+- [@crertel](https://github.com/crertel)
+
+Difficulty: medium
+
+References:
+- https://haskellforall.com/2026/02/my-experience-with-vibe-coding
+
+Prior efforts:
+- None, really.