The common sense presumption that “skill and luck are opposites” is irrational. This means skills can't be totally or mostly based on luck. That's the opposite of the truth.
For example, when Warren Buffett chose to use the results of financial science to guide his stock picks, he did not have sufficient high-quality data and statistical analysis to justify his beliefs.
The CRSP database was not created until 1960. It still had a serious & under-recognized bias within small-cap stocks. The problem was not touched until the 1990s. The size effect was solved in papers only in the late 2010s. It was not until the 2020s that people became convinced that financial ratios did have out-of-sample effects through A.Y. Chen's published research. Of course, unpublished studies will be earlier, but not by much.
Financial ratios generated by data mining have out-of-sample persistent predicting abilities and weird formulas. These imply that some popular intuitive financial ratios generated by financial science are valid not primarily (or at all) because they are consistent with financial science or intuition, but rather because the financial data provide high-quality, usable information for predicting future stock returns. By the way, the so-called “signal vs. noise model is very inaccurate and the fact that people vastly overestimate predictability in this model is a topic for another day.
That said, the skill that led Buffett (or many other people) to pick what seems to be essentially the right criteria for stock picking before research based on CRSP comes largely from luck. As a result, we can infer that even the real skills that lead to sustainable predictability can be/are also mostly from luck, rather than being repeatable without luck.