Your question is an area of incomplete evaluation in contemporary philosophy, and appears in both evolution and empiricism
There is an often unrecognized tension between empiricism and reasoning, which the problem of the “truth” of science and evolution drives to light. Classical logic, which is what most of us think of a reasoning, uses only two categories, true or false. It has a law that excludes any other state, the Law of the Excluded Middle, LEM. True can only be arrived at using certainty arguments, and there is no subjectivity or judgment involved.
In contrast, empiricism is based on four categories: Supported, counter-supported, currently indeterminate, and incoherent/unevaluable. All four categories violate the LEM, three involve explicit judgment calls, and the evidence that one uses to do empiricism is both subjective in nature (observations) and is generally treated as more reliable only if it is confirmed by a consensus of experts (expert is another subjective category). Intersubjective consensus of experts relative to data, analysis, and conclusions is how all science is done (peer review).
Empiricsm has a LOT of conflicts with classical reasoning. Verisimilitude is the particular conflict you are asking about
Analytically oriented critiques of empiricism point out that consensus of experts is difficult to distinguish from the bandwagon fallacy. And for outsiders accepting a consensus of experts, the justification of accepting science is an Argument from Authority. Also, the credibility arguments for accepting science PRESUME the empirical metric of utility, and are therefore circular. And finally, even the “accept as supported” is FAR short of the “true” that science claims as its goal.
The first formal attempt to demonstrate that science approaches truth that I know of was by Karl Popper, who tried to show that successive replacement of imperfect science theories thru the falsification process have increased Versimilitude. Popper’s proof was shown to be analytically incorrect, and that a falsification process cannot be shown to converge on the truth.
Popper is not the only one to fail. Repeated failures are implied in the sections 3.3-4 of this SEP entry on scientific progress: https://plato.stanford.edu/entries/scientific-progress/#TruInf The belief in scientific progress is widespread, but the entry does not cite any demonstrations that support this belief.
Evolution faces the same problems
There is a very close analogy between the evolution of science beliefs, and the evolution of species. Both are based on a pragmatic mechanism. And biological evolution faces the same problem as science.
A widely cited attack on naturalism and the trustworthiness of human reasoning, by Alvin Plantinga is based on the inability of evolutionary processes to generate truth, or a reliable mechanism within us to evaluate truth: https://en.wikipedia.org/wiki/Evolutionary_argument_against_naturalism The wiki article is dominated by critics of Plantinga, so for balance here is a supporter: https://www.biola.edu/blogs/good-book-blog/2015/plantinga-s-evolutionary-argument-against-naturalism
It is not just theists who note this failing of our evolutionarily based reasoning process. A very similar claim to Plantinga’s is made by Donald Hoffman, an atheist neuroscienctist, based algorithms he derived through evolutionary fitness modeling: https://social-epistemology.com/2019/12/05/do-we-see-icons-or-reality-a-review-of-donald-hoffmans-the-case-against-reality-brian-martin/
Draw an inference from these multiple demonstrations
Each of these separate failings can be and are often dismissed by science advocates who want to think science (and evolution) approach truth. But there is a common theme here, independently discovered by multiple thinkers. Synthetic (empirical) processes do not conform to the criteria of classical reasoning. They use a different methodology, where there are four logic states, and “true” is never achieved, only “useful”. Empiricism uses pragmatic “truth” not analytics/classical TRUTH, and is in Classical terms both illogical and fallacious.
My takeaway: pragmatic “truth” is the only truth available to us. Hence pragmatic utility (supported well enough to accept as a working hypothesis, for instance), with its subjectivity, and consensus of experts, etc should be taken as our best measure of goodness, not any absolutist “truth”. And yes, we HAVE no "certainty" that science, human reasoning faculties, or any empirical or evolutionary process will converge on "true", or even converge at all. But they generally converge, and do so to the benefit of those that rely upon them, so we can trust them despite the "logic" insufficiency.
