LHC experiments hope to cast a wide net

The LHC experiments have done careful targetted analyses of their data, looking for expected variations on the standard model. You can't go all hog wild and try any old model, because you know that whatever the final version turns out to look like, it is going to have to make very similar predictions to the standard model for the usual interactions. In other words, the standard model works pretty well, and any replacement is going to have to do at least as well.

They've been running out of candidates. Not that there's a shortage of ideas, but there's a shortage of ideas proven to work as well as the standard model. Most fail that test, and there are so many things any new model has to compare against that the effort to validate isn't trivial.

One important feature of these kinds of searches is that you can make a prediction and then check the data. If I predict that you have two aces in your hand, it is a lot more impressive if I haven't seen a thing than if we're playing stud poker. If I predict that a process will produce an excess of muons flying off at 30 degrees from the beam line, and I find a small excess, we can cheer a bit.

On the other hand, if I don't know what I'm looking for and I see a small excess of muons at 30 degrees from the beamline, that means nothing much.

I was looking at dimuon mass plots for my thesis, and found about a 2.2-sigma peak at the D0 mass. That would have been wild--the D0 wasn't expected to decay into mu+ mu- at a rate we could have observed. But random bumps happen. (I was curious and looked at the events that made up the peak--enough were garbage to account for most of the peak). Not important.

My advisor used to say that "prior knowledge is worth 3 sigma."

So, the experiments are going to go for "we don't know what we're looking for" searches. I predict there will be plenty of 3-sigma deviations.

Since the monte-carlo simulations that are supposed to predict what the data should look like aren't perfect, they may wind up seeing points of disagreement that merely represent failures of prediction. It will be a big project, and "deep learning" isn't as much "learning" as people like to think. OK, you've got a bump there. What does it mean? Is it a bump from leptophilic Z' or a bump of Amativeness? Or is the simulation of the photon interactions in the TRD not quite right?

But at this point they don't have much choice.