How do you guard against data dredging?

There is no data length and depth sufficient for statisticians; we are talking about "acceptable". You reach that by comparing In and Out periods, rolling returns, rolling worst returns, examining what occurs if you withdraw the single best winners from the portfolio to see if the general return is still good and not dependent on outliers, you look at Sharpe, UI, drawdowns, recovery times, performance in shock/drop/recession/crashes/correction windows and compared to SPY and NDX, you examine to see how much overlap there is in the picks between strategies, worst year, worst drawdown, # of drawdowns, average time to recovery from all drawdowns of x%, average/median year, all start dates, with and without timing rules....

So far (and I'm not done) the only negative is more 10%+ drawdowns (but faster recovery from them than SPY and NDX). Most of the criteria are not that arcane or shocking, and I'm looking at "mounds of Toast" to see if they are lucky or there is a spectra of performance.

I think that much is a reflection of the change in the US economy in the last 50 years where tech has taken off; whether that is true going forward I leave to you (I believe it will be; the US, Sweden, and Switzerland had the best performing stock markets 1900-2000 and I expect capitalist tech-adopting economies to continue to outperform).

Equities beat bonds and gold over almost all 15 year periods, the US generally, and my own personal performance has matched my expectations. Crashes will occur so courage and patience is needed.

But I don't think the data result is merely randomness, no.