Let's ask a different question: how do you know you're not twenty minutes old? That is, "you" didn't actually write the question, you just think you did because things were different in the past such that, well, somehow or other you exist now and have memories of writing a post, but there are parameters which changed and somehow spoiled stuff.
Pretty silly, right? But, frustrating, you can't completely rule this out with logic or science or anything else. It's just a variant of Descartes' evil demon, albeit possibly a non-intelligent demon (maybe there's just some process that generates pretend histories, and a flag flipped saying, "okay, stop generating pretend histories now, plug them into beings and start doing stuff for real").
So if you really want to know absolutely for sure, it is logically possible that we could be almost arbitrarily badly confused about almost everything. This does an absolutely terrible job of explaining all the regularities we see, but it can't completely be ruled out.
But if you're "just" doing science, you don't need absolute certainty. You ask: can we construct hypotheses that make predictions about what we observe, and do those hypotheses work on both observations we've already made and ones we haven't made yet? Can we find better hypotheses?
And here, of course we can do science on stuff that's already past because it doesn't matter why you failed to make the observation before. It doesn't need to have not happened; you just need to have not cheated by plugging the answer into the hypothesis before "testing" it. You could get new observations by making new things happen, or looking at consequences of what has happened previously. Doesn't matter at all, as long as it's honestly independent from the generation of the hypothesis.
Then the question is whether parameters can change in the past in a way that will confuse us. Possibly! The length of the day has, we think, changed throughout Earth's history. Based on physical effects we can observe now, it looks like the rotation of the earth is slowing slightly, and if we extrapolate the day should have been about 22 hours long during the Cambrian. If we check, by measuring relative changes in sedimentary patterns that occur on yearly, daily, and lunar-monthly cycles, it seems pretty much spot-on.
Now, note what we did: we took multiple different predictions and they agreed with each other. We didn't take one and trust it, we checked. If some parameter changed in the past, it would somehow have to have affected these different factors by the same amount, so that they'd agree.
Dating via radioisotopes is done much the same way. There are all sorts of radiogenic decay processes (for over a dozen, see Radiogenic Isotope Geology by Alan P. Dickin, Cambridge University Press (1995)) that agree with each other. So any "parameter change" would have to affect all radioactive decay the same way. Since decay of U235 and U238 is responsible for much of the internal heating of earth, if it had happened much much faster, the earth would have been a sphere of boiling lava unless thermal conductivity were also different. (And the sun wouldn't work properly either.) Furthermore, these methods agree with completely non-radioactive methods of relative time such as sediment accumulation, coral reef growth, sea-floor spreading, and so on. So not only can we perform scientific studies in the past, when it comes specifically to radioisotope dating, we can check and cross-check and cross-cross-check, and everything checks out. (To be perfectly clear: these are tests of the hypothesis!) Now, all measurements have some associated error, so if you want to tell to 0.1% how old something is, it can be quite a challenge. But if you want to tell if something is 5 million or 500 million years old it's really easy, and really robust (if you follow proper procedures and measure or avoid sources of error, as you will if you e.g. read Radiogenic Isotope Geology).
So, in conclusion: while it is logically possible for evil demons to trick us, or for multiple models to give similar answers, in practice with radioisotope dating there are no competing models which make any sense at all, and tons of hypothesis-testing has been performed and the standard hypothesis has done really, really well.