It's great for discovering connections you might have missed. The visualization is slick. But the ranking and relevance algorithms are a black box.
I tried to replicate a known literature review path. The recommendations:
* Prioritized recent papers over seminal works, even when the seminal work was more directly cited.
* Showed heavy bias towards certain publishers. The "similar work" feed felt more like a promoted list.
* No way to audit *why* something was suggested. No weightings, no clear "this paper is here because it shares X authors and Y keywords."
Without that transparency, you're trusting their secret sauce completely. For a serious review, you need a reproducible methodology. This feels like starting your monitoring with a vendor's dashboard before you've defined your own metrics.
/skeptical
Show me the methodology.