chess-2 VS StockNemo

Yes, that is true. One can use NNs in either evaluation or heuristics or both - to my understanding, I haven't researched this that much. But yes what you say would fall under the heuristics category that mainly drives move ordering, better looking moves first. As the search is ultimately a depth first in basically every engine. If you are curious, allow me the shameless plug, you can check out my engine @ https://github.com/phaul/chess-2 I'm happy to answer you anything else about the engine, I love talking about this stuff :)

If you decide to implement all rules ( en-passant / castle etc ) I recommend implementing a simple recursive function known by chess engines as perft. (https://www.chessprogramming.org/Perft#:\~:text=Perft%20is%20mostly%20for%20debugging,functions%20work%20correctly%20or%20not.) Running perft on different positions to a certain depth and then comparing your numbers against perft in stockfish etc is really helpful shaking bugs out.For my pet project engine I wrote some shell script that is supposed to debug and find positions where the engine has a bug in the move generator https://github.com/phaul/chess-2/blob/main/debug.sh

https://github.com/TheBlackPlague/StockNemo, eval function is 76 lines of code. and simpleeval https://github.com/vondele/Stockfish/tree/simpleEval a modified version of stockfish which just uses piece value as an eval function (1, 3, 3, 5, 9)