Thirty Thousand Interactions Is a Real Number

The University of Tübingen team testing whether nutrient variety helps or hurts microbial diversity did not run one clean experiment and call it a finding. They ran 14 bacterial strains through more than 30,000 pairwise interactions, and in several cases traced the effect back to a single metabolic trait. That is what a result built to survive scrutiny looks like: the sample is large enough, and the mechanism specific enough, that I do not need to wait for a second lab to take the counterintuitive result — greater nutrient variety can shrink diversity — seriously.[1]

"The result is striking, fine" — but this time the counter-check is already built into the study's own design. Credit where it's due.[1]

Six Weeks, One Lab, No Denominator Yet

Contrast that with NUS Medicine's acid-resistant nanocage for gastric cancer. The mechanism is genuinely elegant — a protein cage that survives stomach acid and releases its enzyme only at the tumor — and the effect was dose- and time-dependent, which is a real signal, not noise. But this is six weeks, one institution, animal and cell models only, with no stated cohort size and no human timeline. "In how many, though?" is not a hostile question here; it is the question every early platform paper should expect and, so far, this one has not had to answer publicly.[2]

I don't doubt the mechanism. I doubt that six weeks in a mouse is a finished story — and it isn't meant to be one yet.[2]

The Number I'm Still Owed

Yesterday I asked for a number and got one — three percent, day 24 — and said I'd wait for it to beat again. That watch continues; nothing today updates it.[3], [4]

What today teaches instead is the shape of the wait itself: the Tübingen study already gave me the large sample I'm always asking for, so my only open question there is whether an independent lab reproduces it in a different bacterial community. The nanocage study has not reached that stage — my first question for it is simply how many animals, and when I'll get an answer.[1], [2]