My most recent backtest experiments are working with RUNDOWNS...
for example, number 348
step 1, Add 111 (348+111=459, 459 becomes the first number in the rundown)
step 2, add 111 (459+111=570, 570 becomes the Second number in the rundown)
step 3, add 111 (570+111=681)
step 4, add 111 (681+111=792)
step 5, add 111 (792+111=903)
step 6, add 111 (903+111=1014, DROP the 1,000's column... becomes 014)
step 7, add 111 (014+111=125)
step 8, add 111 (125+111=236)
step 9, add 111 (236+111=347)
step 10, add 111 (347+111=458)
this gives you 10 combos... 459, 570, 681, 792, 903, 014, 125, 236, 347 and 458 to play for the NEXT draw...
Now, even the BEST performer over the entire 9,145 drawings of the PA lottery still leaves you OUT money, but not as much as sticking to any single number).
One interesting observation is that the hits are moved around for each rundown number (111 vs 222 vs 333 etc).... I'm wondering if some sort of rundown calendar could be developed that would give some indication as to which ones are "due"... The best single performers so far are 500 (125 straight hits in 9145 draws) and 444 (427 BOX hits in 9145 drawings).
This is more of a "why-less" system, as I have no idea WHY it has a "still-losing-but-better-than-random" hit rate (someone else on LP mentioned it), maybe because you play 10 numbers rather than 1? no, because the 10 best performing numbers played together racked up almost 200 straight hits... Almost (but not quite) breaking EVEN in the big picture (when counting boxes).
I'm working on the concept of a calendar to track hits by rundown number (optimizing for straights) but this will be a MAJOR project. Yet another project for the overflowing back burner...