And also on theory: if all stocks vary in valuation level between over and undervalued over time, the order of market cap gets shuffled somewhat away from the order you'd get on pure absolute intrinsic value. Undervalued things slide down the list, overvalued things move up while it lasts. The ones at the top by market cap will be, statistically, ones that got shuffled upwards in the list order due to transient overvaluation.

I don't think I made my point clear enough.

Let's start here. The order of market cap changes for a few reasons:
1. Some move up due to overvaluation.
2. Some move up due to higher earnings.
3. Some move down due to undervaluation.
4. Some move down due to lower earnings.
And likely all move around due to all 4 of the above.

But my point ISN'T about an index, let's say the S&P500, but more about the composition of the index. So, if you invest in the index, or perhaps a derivation of the index, maybe S&P500 minus the top 5 ("Next 495"), or maybe S&P500 minus the bottom 495 ("behemoth S&P"), or maybe equal weighted ("RSP"), or whatever. But when we say "small stocks outperform large stocks", we are referring to company size out of the entire universe of stocks, not just one of the indexes. If you look at "small stocks in an index", you have to account for the survivors bias. The ones that went bust, or close to bust, were removed from the index continuously so if you are looking only at index stocks then those are gone and you "lose" the data of counting them (as negative contributors to the overall return of "small stocks"). If you were to look at the TOTAL performance of all small stocks over the 31 years versus all large stocks over the 31 years, perhaps the large stock performance would look a little better than the small stocks ... mainly because a substantial percentage of the small stocks disappeared over the period (and essentially went to zero or were acquired at some level that stops its compounding at that point). I don't have sufficient data to say one way or another, and it remains a thought experiment for me. If anyone has data, and the wherewithal to massage it, it might be interesting to see the results and the methodology.