> Maybe I need to log in? There's some followup tweets: > I don't like to be harsh about a paper, but the fanfare around this one makes it necessary to respo

> Maybe I need to log in? There's some followup tweets: > I don't like to be harsh about a paper, but the fanfare around this one makes it necessary to respond. > There are many ways to go wrong when inferring past events from genomic data; the datasets are large and noisy, and computational methods get complicated. > Thus it's essential to demonstrate clearly that your results are not an artefact of data processing or methodology, or the consequence of overfitting to a particular aspect of the data. Simulations are a vital tool for this, and they need to show that your results are robust.. > to alternative scenarios that violate the assumptions of your model. They also need to show that the observed data are indeed consistent with the model you have inferred. Sadly, it's not clear that these requirements have been met in the paper published in Science.