Take the evolution of temperature measurements through proxies (although that is not my field). If at some point it becomes known that some method of measurement is biased, then that implies a re-evaluation of past research.
Every time such adjustments happen, there is a meta-learning regarding the pace of adjustments and the severity, and also regarding which people andinstitutions prefer to stick to "old" or perhaps more conservative interpretations in some sense.
When you try to summarize the scientific consensus, you need to do so by literature because (1) the composition of the field is always changing to some degree and (2) you will never get all scientists to tell you their opinion at the same point in time.
But when you do this - summarize the literature - it is crucially important to adjust for historic biases, drift in methods adoption and interpretation and so on.
And this is just the measurements part: There is also the problem of predatory journals etc etc. - all of these can skew the "consensus" if you don't account for them.
So I'm pretty convinced that your argument of a pure, "individually verifiable" science is desirable, but has never existed and never will.