Yeah. The higgs evidence is pretty convincing visually. Not so sure about LIGO. There is an extraordinary claim of noise reduction that requires extraordinary evidence and it's all obfuscated behind adaptive machine learning based filtering, and the statistical analysis on that is unparseable to a non-expert (that is worrisome). The pulsar timing network though is easily believable.

Luckily, there's pretty simple statistics that one can throw at that once the third detector comes online. Hopefully that comes in before we spend too much money on LISA.

It's basically this, from the article, but from astro:

> Particle physics does have situations where the hypothesis are not so data driven and they rely much more heavily on the theoretical edifice of quantum field theory and our simulation of the complicated detectors. In these cases, the statistical models are implicitly defined by simulators is actually a very hot topic that blends classical statistics with modern deep learning. We often say that the simulators don't have a tractable likelihood function. This applies to frequentist hypothesis testing, confidence intervals, and Bayesian inference. Confronting these challenging situations is what motivated simulation-based inference, which is applicable to a host of scientific disciplines.