The exact sentence you cite is an example that I've tried everything I could to avoid that. In one of the original revisions it was "We can retrieve very unique documents no one else has", not "Search engines with their own indexes can retrieve documents no one else has".
> Besides the clear AI smell, this nonsensical claim also plainly contradicts the methodology's key evaluation claim that the quality of an engine's results should be measured against how much it overlaps with the reranked aggregate of the other engines. The benchmark thus seemingly values an engine's ability to "answer unanswerable questions" at zero.
It is fun to hear "AI smell" accusations about these specific sentences. I do use Claude to fix my grammar mistakes and give editorial opinions, mostly because I'm not a native English speaker. And as for these specific sentences, I even had a session where Claude fixed an obvious mistake in my writing: https://snipboard.io/rkm4LD.jpg
There is no contradiction to the methodology. We use ultimate engine to calculate IDCG, but that doesn't mean that the methodology favors the overlap. 5 relevant unique results would produce exactly the same nDCG as 5 relevant non-unique results.
The main part of the benchmark (the one that measures quality) doesn't care about result uniqueness (if it is not about full URL duplicates).
> Yeah? Care to cite anything for that?
The closest source is probably LRAT (https://arxiv.org/abs/2604.04949), it is cited in the benchmark, we used their trajectories in AgenticRare.