Validate a tool emulation against its output in native order

When checking a re-implementation (e.g. a Python emulation of an R marker-gene function) against the real tool's output, never re-sort or re-filter the reference table with your emulation's own assumptions. Read it in the order the tool returned it. I re-sorted real Seurat FindAllMarkers output by (p, -logFC) before comparing, and got a '169/170 match'. The tool's actual tie-break (by -abs(pct.1 - pct.2) since 5.1) was different, and a downstream claim was wrong. Also run the real tool on each variant you report (e.g. via a formula-override argument), not just the default, and compare list lengths as well as overlaps.