Wood’s N-ray experiment looks like a suitable example. He removed something that was supposed to be necessary for the effect, but observers still reported the effect. Here, too much agreement became suspicious. Should robustness sometimes include this kind of test?

Really enjoyed this, especially the point that disagreement can be useful. It made me wonder about the opposite case: when different methods agree. Wimsatt’s third step asks us to test when this agreement holds. How often should we also try to break it?

Yes, partly. But many cancer models already produce individualized risk estimates, and newer models operate at a single-cell level. My point is that their integration is statistical and selective, while the cell has materially integrated far more of the relevant history.

Thanks all the current and former ROTO members for their kind support!

Perhaps the funniest part is that the difference between these “experienced environments” can nevertheless be mostly captured by a simple physical quantity.

Evening science steps back from the day’s work and asks the almost childish question that still matters beyond the field. Morning science turns back toward research: what do our concepts commit us to, and what should we actually measure, follow, or do differently?

François Jacob distinguished day science - ordered, testable, public - from night science, where questions, guesses, and unexpected connections emerge. For a while, I’ve been thinking that philosophy can add an evening and a morning to this cycle.

We are all impressed by models that integrate tens of thousands of biological variables. But here is the awkward question: if cancer research has so much data and such sophisticated models, why are we still so limited in predicting individual risk?