James,
I think that is true. One reason that P123 requires some out-of-sample performance before people are allowed to subscribe to a Designer Model, I guess.
I think your first post was also illustrative: “All That Glitters is not Gold.” There was some debate about that article as I recall. But I do not remember a single post saying: “They just didn’t backtest enough.”
The de Prado article does mention other methods besides walk-forward. And I do not think one should limit their ideas to just that short article. Even if you happen like walk-forward it is very time and/or computer resource intensive. There is no perfect method. I don’t think so anyway.
P123’s Designer Model method —essential paper trading, I guess—is just one example if someone does not want to use walk-forward. Maybe someone would want to paper trade even if they do use some sort of validation or a hold-out test sample.
Best,
Jim