In a nutshell, it involves identifying a pool of explanation candidates to account for some state of affairs, then narrowing the list down by a principled set of criteria to the single best explanation, then inferring to it as the likely true explanation.