The essence of testing inflation can be condensed into a single sentence, namely
The simplest models of inflation predict power-law spectra of gaussian adiabatic scalar and tensor perturbations in their growing mode in a spatially-flat Universe.
This sentence contains 6 key predictions of the inflationary paradigm, which I've underlined, but also one crucial word, `simplest'. The trouble is that inflation is a paradigm rather than a model, and has many different realizations which can lead to a range of different predictions. From a straw poll of cosmologists, everyone agreed that there were at least several tens of different models, and I'd say there certainly aren't as many as a thousand, so a reasonable first guess is that at present there are around one hundred different models on the market, all consistent (at least more or less) with present observational data.
A valuable scientific theory is one which has sufficient predictive power that it can be subjected to observational tests which are capable of falsifying it. When the model survives such a test, it strengthens our view that the model is correct; in Bayesian terms, its likelihood is increased relative to models which are less capable of matching the data. It is useful to think of these models at three different levels:
Finally, note that since one can never completely rule out a small inflationary component added on to some rival structure formation model (e.g. a combined cosmic strings and inflation model ), in practice we are initially testing the paradigm of inflation as the sole origin of structure in the Universe.
It can also be helpful to make the admittedly rather narrow distinction between tests and supporting evidence . A test arises when there is a prediction which, if contradicted by observations, rules out the model, or at least greatly reduces its likelihood relative to a rival model. In this sense, the geometry of the Universe is not a test of inflation, because there exist inflation models predicting whatever geometry might be measured (including open and closed ones), and there is no rival regarded as giving a better explanation for any particular possible observation. By contrast, the oscillations in the microwave anisotropy power spectra (both temperature and polarization) do give rise to a test, as we will shortly see.
Supporting evidence arises with observational confirmation of a prediction which is regarded as characteristic, but which is not generic. A good example would be the observation of tensor perturbations with wavelengths exceeding the Hubble length, for which inflation would be by far the best available explanation; they do not give rise to a test because if they are not observed, then there are plentiful inflation models where such perturbations are predicted to be below any anticipated observational sensitivity.