Show HN: Sourcery – Benchmark for search APIs on the sources they return

I was picking a search API for a research agent and every comparison I could find was run by one of the vendors so I ran my own.

Most search benchmarks end up grading the answer the model produces, not the actual pages returned.

Open to feedback!