The area of smashing particles is surrounded by detectors. Whatever is captured by detectors is a particle (I would say "by definition" for simplicity). It causes some reaction inside detector. For instance, it can be a trace of bubbles in Wilson camera. The shape and other parameters of this trace depend of the type of particle. It has been known from numerous experiments that there is only a limited number of trace types, and hence there is only limited number of particles that can be detected directly (i.e. each type of trace is associated with some particle type, like electron or proton. All electrons result in similar shapes of trace when crossing the Wilson camera). So, the first "class" of particles comprise of those instances that can be detected directly. Let's call the "real" particles.
There is also another "class" of particles that cannot be detected directly. Physicists think that they are particles too for two reasons:
- Physicists developed a model that explains why this particle decays into certain combination of "real" particles that can be detected directly.
- Quantitative predictions of this model are consistent with experimental data.
Examples of such particles are W and Z bosons, as well as number of so called "resonances" (presumably short-living particles) etc. And also Higgs boson of course.
In fact, collider is a kind of a black box: there are "real" particles on input and output, but we can only guess what happens inside. "Guessing" process is explained in principle by @zhermes.
This, in turn, brings us to another question:
Why some particles cannot be detected directly?
There might be a number of reasons for that:
a) our detectors are not good enough
b) particles are too short-living (which is in fact equivalent to reason (a))
c) for some "fundamental" reasons (usually quantum mechanical arguments are used)
Frankly speaking, I do not know for what reason, for instance, W and Z bozons cannot be detected directly.
And last (but not least): I admit that possibly those non-"real" particles are not in fact particles as per rigorous definition (field configurations that realise irreducible representations of Poincare group in Minkowsky space or something like that). I admit that they might be "shattered fragments" of "real" particles.
But modern physics is unable to establish that. Because our models were developed based on collider-type experiments (i.e. "real" particles in and out of the black box). And all models are based on "black box" logic - quantization. Do you know what quantization is? it is a set of numbers connecting inputs and ouptuts of the black box. Vicious circle. That's why these models are useless for identifying the properties of the isolated and not moving "real" particles (mass, charge ets.).