Default MS Excel charts in every other figure, terrific stuff

You might expect this means the peer review didn't work, which is completely true, but this figure is technically better than the preprint version

The Medical Detectives by Roueché?

The legal services point is particularly great a week after www.nytimes.com/2025/11/07/b...

It is, for whatever it's worth, interesting to read through the notebooks and really understand the gaps. But when the summary plainly says n=7 but every analysis assumes n=1000, the whole thing is a bit rough to get off the ground.

Those are some very small p-values for the single-cell data 🙃

Yes, the behavior of the overall vector magnitude, as well as the component magnitudes, are not very well-behaved. Not quite sure I understand what you mean here though.

Not necessarily that specific one, but something that combines stationarity + overdispersion, and cell type mixture models are a natural candidate. A null hypothesis is not very well-defined here but it would be helpful to know if this falls prey to the same pitfalls as the original 2018 software.

Very exciting work! If you don't mind me asking— what are the environments required by the disclosed RDS files? They throw errors and cannot be opened in R.

Yes, exactly. This is a very common theme in the performance of velocity benchmarks: every new paper has to treat the original papers as ground truth, so performance on their datasets is stable

Yes, the problem there is of course they're all suboptimal (and can be arbitrarily bad) by Johnson–Lindenstrauss, and neighborhoods + global structure get scrambled pretty badly. This is a generalization problem again— assuming that paper results are ~ok & user input is ~perfect.

Though if the distribution is nonuniform under a fully controlled null, that is a bit strange.

I would appreciate it. There are many ways to define a null hypothesis here, and it would be great to see whether a typical false positive (nondynamic simulation that produces velocity arrows) can be spotted by such a procedure.