The problem. Once you have BANKSY, and a dozen other spatial-domain and niche-identification methods, the same question that hit deconvolution returns: which one do you trust, and under what conditions? Domains are fuzzier to score than cell types, there’s rarely a clean ground truth, so a careful benchmark is badly needed.

The idea. This is a systematic comparison of niche/domain-segmentation methods across datasets and platforms, scoring how well each recovers known tissue structure and how sensitive each is to the choices that matter, resolution, neighborhood scale, number of domains. The useful output, as with the good benchmarks, should be conditional guidance rather than a single winner.

Why it matters. This is exactly the flavor of paper a standards-minded facility values: not a new tool, but a map of when the existing tools work. Pairing a method paper (BANKSY) with its benchmark is the “know why, not just which” discipline I keep coming back to, it’s what turns a preference into a defensible recommendation.

Verdict. This is a very recent 2026 preprint, so I’m reading it purely for its purpose and framing and deliberately not repeating specific rankings or numbers until it’s peer-reviewed and I’ve read it in full. Flagged as the most speculative entry in this batch, verify everything against the manuscript before citing.