Please attach which harness was used with every published entry in the data

I've been playing with Pi a lot recently with multiple agent configurations (which is just to say, presets with model + reasoning effort + subagents all baked in, much like what Amp does). Part of this also means naturally customising the prompts and tools for each agent to maximise each model's performance - and although benchmarks are always not super representative of real-world use, it's super helpful to go look at the harness that does the best job with each model. I was super excited to see ReactBench since I've been writing a lot of React with side-projects recently, and it'd be super helpful to understand which harness eventually gave the scores that you have published. NextJS evals and Artificial Analysis do this too.

Start by locating the data source and publishing path for ReactBench's published entries; the issue does not name specific files or tests. Trace how each entry is represented, then add the harness used for every published result and verify that the published data displays it consistently.