Or you might want to read the book Fooled by Randomness by Nassim Taleb.

Were folks fooled when we thought it worked? or when we thought it didn't work?

Seems like there are three main possibilities to explain these observations about screen performance:
- fooled by randomness
- post discovery or herd effect
- conditions at certain times favor a strategy and at other times they don't

YEY would seem not to be in the second camp, as it performed well for years after it was discovered (as I recall - have not used it or studied it).
I guess it could be the first, or the third.

For the first camp, we'd have to have randomness on the scale of years to explain the observations. I guess that is possible.
Usually, I'm mentally in the third camp. One could run ANOVA type statistical analyses on YEY performance using different possible factors.
For example whether fed policy/interest rates are rising, falling, or stable. Is performance correlated to employment numbers, other macro economic conditions.
Or to the dominance of mega-caps? Or even which party is in power? (I presume not).
These larger scale factors will come and go; interest rates will rise and fall, and mega-caps will not always be in favor. So who knows.

Coincidentally, I happened to be looking a few days ago at the Big_Jump screen which there was a thread about back in 2015.
Robbie and Ray were going at it, and Robbie posted this http://www.datahelper.com/mi/search.phtml?nofool=y... which showed a 42% return from 1987-2015.
What do you think it was from 2015 to the end of the pandemic? It crashed, 0% return through to the bottom of the pandemic in late March 2020.
And from then until now? It's back at about 46% CAGR, indistinguishable from the earlier period 1987-2015.

SiPro screens are much harder to interpret than something like YEY.
But my feeling is that screens can suddenly "work" again, that it's not just randomness.
I don't trust my feelings though, so a deep dive into screen performance would indeed be nice.

Mark