Taking TGMark's initial SI-Pro screen, using my typical 'crap' filters, doing a bit of (mostly irrelevant) simplification, stripping out the derived fields to enable direct SI-Pro screening & out-sourcing the VL1700 replacement to the S&P500...

Offhand I don't think this screen is well suited to the S&P 500, as that group is so picked over that dividend yield is lower than it might otherwise be. Many/most decent companies with decent yields are further down the list where they are more likely to be a pricing anomaly.

I took yours and did what I think (?) is a closer approximation of my original post.
After your crap filters, simple market cap top 1700 as a quickie approximation of the VL universe.
Then 50-HTD-80 and 0.4% friction as I tested it.
gtr1.net: GTR1 Backtester - ROE_500

A side effect is that the market cap field I used limits the backtest to 2012 and later, but that's easily fixed with the appropriate pref() if you like.

The only thing I added (because I couldn't easily do so in my owntest) is creating an ROE figure for firms with negative book value.
Positive earnings on negative book is actually a good thing on average ... it means the business needs no net assets at all to be able to earn money. So I plug in an arbitrary ROE of 100 for those firms. Losing money on negative book value is of course a bad thing.
If you're using this URL in future, note that my version ends at end 2024.

My GTR1-fu is a bit rusty, apologies for any goofs.

The VL universe and this universe behave pretty similarly. (VL .5% higher, within statistical noise)
The backtest CAGR with SIpro as above is around 2.8%/year lower than my backtest with VL once I match the date range.
The strange thing is that the VL test outperforms its 1700 universe by 2.8% in backtest, but the SIpro one lags its universe by 1.5%.

None of that is a very big deal, since the goal isn't really maximizing CAGR, but maximizing yield while merely maintaining a merely decent overall return over time.

Jim