[Already__Taken]

/Not a physicist in any way but;

The approach seems to be we observe a lot of behaviour, then build up models to explain that behaviour. If particles need to be made up to balance an equation then they get made up.

You project that model into some scenario to say "If this model is right, the following ... will happen" and observe. If you're right then work goes on for further evaluate "What about this scenario" and finally a big machine gets built to directly work out, from all that's prior, if a particle that needs X,Y,Z properties really does exist is should be seen in the measurements that are supposed create it.

So to answer your question; Both?

E.g You could take the model, put it in a computer. Simulate what the LHC does and look at that graph of kind of particles it makes. Then you build the LHC, actually smash some things together. Same graph?