Things in RPGs that you used to like but don’t anymore

[RandomPeasant]

Particularly for monster rules if the monsters are built using points or to a specific challenge rating or similar, since something can be significantly more or less useful to a PC than to a monster you are likely to meet once.

Having monster creation rules is pretty much an afterthought. Every edition of D&D that has every existed has approached the problem of "how you do you get new monsters" with "buy more monster books", and that has worked out at least as well as any monster creation system I've seen a game of non-trivial complexity. The real issue is PC-like NPCs, and that's an inherently harder problem because you can't just cheat. A Gargoyle is not supposed to be a PC. It doesn't matter if its Stone Form ability is different from PCs abilities, because it is not presented as being the same. But once you start talking about differences between PC Wizards and NPC Wizards, you're on thin ice very quickly. There's a real chance that what you want is not any kind of mechanical change, but some kind of easy-to-use online character generator. If you could just say "give me an NPC Wizard at X power level" and have a website spit out a useable stat block, that would be fine for 99% of use-cases and require few-to-no sacrifices in terms of transparency.

If you do want a mechanical alternative, the easy thing to do is not to create some bespoke NPC-generating system, but to identify some adequately PC-feeling chunk of a PC and design it so it can be stapled to some stats on its own. If there's a Wizard sub-class or lifepath or whatever that gives some mechanically simple Wizard-y abilities that you can rapidly turn into an expendable NPC, that solves your problem without needing to try to marry distinct mechanics into a single in-world concept.

But the idea that you can get adequate results while having NPCs and PCs work fundamentally differently is something I am very suspicious of.

For building significant NPCs, I'm more concerned about the first, and not at all about the second.

Your "what stuff costs" system is (or at least should be) closely tied to your system for assessing power levels. That's something you care a great deal about when creating an NPC, particularly one PCs expect to fight.

Being able to reliably do something, especially if you were good at it, seemed, in my testing, less fun than betting extreme success against extreme failure.

I don't think that's an accurate assessment of how a d20 RNG works. At the extreme end, a d20 allows more consistent results than dicepools (though equivalent to something like 2d10). If you have a +10 bonus on a d20 roll, you will succeed on a DC 10 check 100% of the time (barring epicycles like nat-1 auto-fails). Conversely, if you roll a dicepool of five six-sided dice that hit on 4+, you will fail at a task requiring one hit a bit under 5% of the time (again, barring some epicycle like allowing people to default). Flat RNGs simply do different things than dicepools, and are suited to different sorts of games. In a game where hard-core badasses can expect to take on armies of normals, a flat RNG is better because you can push people off the RNG without needing any extra mechanical work. In a game where "the cops showed up" is supposed to be a lose condition for even high-power characters, a dicepool is better because it limits how powerful characters can be.