mrpmorris/sparkrun-recipes ? reverse-engineered prompt

I want this repo to be a simple Python project that gives me a set of custom SparkRun recipes I can actually use, plus an easy way to benchmark them across different models.

Please set it up so the recipes live in a clear place, the available models are defined in one config file, and I can run quick command line scripts to download what I need, test one model, test all models, and compare the results. I want the benchmark output to be easy to read, with side by side comparisons and basic stats like timing and any simple quality score you can reasonably provide.

Keep the whole thing practical and lightweight, not over engineered. Make the scripts and defaults easy to understand so I can tweak prompts, models, and recipe settings without digging through a lot of code. If SparkRun has expected recipe conventions, follow those, and look up the current docs online if you need to.