Wandr Benchmark: Evaluating Research Agents That Must Search Wide and Deep

Repo with benchmark tasks, evaluation harness, tech report:

https://github.com/perplexityai/wandr