I suspect that with sufficient data, we could put together charts (curves) of the relation of time and the degree of skill acquisition at each point in time. I think this should be named the long-term skill acquisition curve.

I'm thinking about this as though it's similar to the microeconomics concept of a firm's Long-Run Average Total Cost curves — it's inferred from many smaller curves. Unlike with LRATC curves, I'm not explicitly thinking about this in a way that allows for both skill acquisition and loss (as, with the long-run nature of the LRATC, the inputs are able to be varied whenever one wishes to do so, but with skill acquisition you cannot will yourself to be suddenly more or less skilled — this is similar to the notion that there's a degree of rigidity in a given market).

I suspect that in gathering data about the skill, the data gathered addresses whether the skill was in fact acquired, retained, or lost, so it may be the case that the potential for skill loss is sufficiently addressed here — but I feel like this does not sufficiently address the potential for skill loss. In order to think about this potential, we likely need to think in terms of counterfactuals, especially as the data that we have is necessarily tied to whether the trick was attempted and omissions of records is not, yet, sufficient reason to infer that no attempts at the trick were made (otherwise we could assume that if we don't have data for a given time period, no attempts were made then, and we'd want to see the relation between these inferred periods of non-attempts and performance, especially looking for if we could infer that performance is behind where it otherwise would be). [This is a discussion about assessing causal relationships, and I have in mind the presentation related to this R package, Casual Impact.]