He he, there's no harm in trying eka nga. This is my idea, best is start with your "invented" start-up positions one at a time as the moves progress, and on each instance make a FEN for each variation, collect and make an EPD out of them, then use at least 5 strong but unique chess engine (hindi clones) to auto-analyze/auto-annotate the EPD to show the scores - EPD testing yan. Up to what ply or depth, it's up to you. Be sure you allow the engine to think a little bit longer to get the right score.
Or, make a FEN for your starting position, and make the engine with multipv-option analyze it say multi-pv = 2 to 4 for a start. Observe the scores on each variation. In this way, you will check if the move is safe to follow or not. then follow what I mentioned previously. Ginagawa ko yan noong araw to add some variations on my opening book manually, pero noong lumakas na yung mga engines, I only play the lines with the top best score and nothing else to be on the safe side. But I see that the engine scores on the first moves (<6 ply) doesn't really matter as there's not much data to assume that both colors have the advantage at this point. Pero creating unusual lines is always a risk, but if you learn how to manipulate the lines perfectly to your advantage, why not! Kaya, you need to verify it using multiple strong engines to make sure it's safe to use.
Nowadays, you can do this automatically w/out even using a pen. Kung kilala mo so Ferdinand Mosca, yung pinoy na gumawa ng Deuterium chess engine. May application siyang ganyan and he's been helping testers on this area. May isa pang pinoy, si Ed Apostol ng Twisted Logic/Hannibal chess engine - magaling din yan sa ganyang process. Ang kilala ko lang dyan ay si Ed at tinulungan ko yan noon dahil wala siyang magandang C/C++ compiler sa bahay he he - sa Mindoro.
Sa computerchess AI naman, yung strategy na alam ko ay to find the weakness of the "known" opponent - anti-chess strategy. We use stats to find it and do some thousands of simulation tests till we get it right. We use different books for each opponent + black and white book. Kung unknown yung opponent, we just use the best book we have and just rely on our engine and hardware. But that was a long time ago. A lot has change, simula ng lumabas yang Monte Carlo Tree search algorithm atbp. he he.
Masakit sa ulo kapag na-addict ako ulit dito he he. Dumugo yung ilong ko dahil sa HBP kaya nag-retire na ako dyan.