I like class and subclass abilities that:
A) are reasonably straightforward without being overly fiddly or confusing (ie: user friendly); B) are powerful enough to be worth keeping track of but not game breakingly powerful (so: worthwhile in a balanced way); and C) are evocative of the character, class, or subclass's narrative. I also really like abilities that D) lend themself to creative play or have outside the obvious usage, but don't expect every ability to do so (at least not within a 5e+ framework). A good class or subclass feature fulfills at least the first three criteria reasonably well, though being weak on one point can be made up for on others.
So a pair of examples that come to mind for me from the Hunter Ranger subclass (because their names are, for little good reason, linked):
-A feature I mostly Like: Hunter's Prey. Extra damage to an injured enemy or an extra attack against an enemy next to another. Either option you can pick with it easily satisfies criterion A of being easy enough to understand, track, impliment, etc. and criterion B of being worthwhile but balanced, since both options are pretty powerful, especially at 3rd level, but they are also likely to only work in a few rounds per combat if you're lucky. The abilities themselves are great for criterion C, with Colossus Slayer going right to the whole hunter thing and Horde Breaker always making me think of Aragorn outnumbered by the Uruk-Hai, but the concept that you can switch options on a short rest makes no sense. This switching also makes it more complicated, and frankly wastes half the ability because unless you know you're to be up against a lot of melee weenies you probably never pick Horde Breaker (mileage may vary if your DM is always clumping enemies together, the point is that almost every combat has Colossus Slayer opportunities and many have no Horde Breaker ones). The switching is better perhaps on this front of seeing both get use than the old 5e version where you just picked one forever, but this improvement has come at the cost of making no damned ludo-narrative sense. My main design suggestion would be to simply allow either to be used on a per turn basis, or just make them separate abilities.
-A feature I mostly Dislike: Superior Hunter's Prey: Throw your Hunter's Mark damage on another creature within 30 feet of the one you hit. This is simple enough (at least if you use a battlemap) though I've got to deduct a few points for having a name which would seem to link it to the other ability when they have no relationship whatsoever (the 5e version had them linked for some more logical reason if I recall). But in terms of power, throwing an extra d6 damage around at level 11 and up is not going to amount for much. Honestly if there is a saving grace it is that since the extra damage just targets "a creature" and is automatic there is a little outside the box play to be had here, whether it is helping your allied barbarian maintain rage, waking someone up, or forcing an concentration check, so I guess I'd give it some points on Criterion D. But on Criterion C it really falls apart for me. Why is the spell which normally only augment's my Ranger's martial skills going off and doing damage of its own against creatures they didn't even hit who were 30 feet away. "It's just magic" is really just not good enough when there are so many abilities going to the "just magic" well, rather than logically flowing from the narrative of the class or subclass. If the class or subclass is some flavor of magician that is one thing, but this is a thematically pretty non-magical subclass.
So yeah, that's basically how I approach particular features working or not working from a design perspective. A subclass with a suite of features that "work" in this manner (and at least some sparking creative play) and a compelling concept people would want to play is basically a good one. A class that does so and can support at least a half-dozen or so subclasses that do so is a good class. It should also fit and support the setting it is designed for if it's setting specific, or not make any outlandish demands of the setting if it's setting neutral.
Overall classes and subclasses should be rigid enough to be evocative and aid in players creating compelling chacter concepts, while also being flexible enough to accomodate many PC concepts, so there is some fine-tuning to make the whole more than the sum of its parts, but I do believe making sure all the specific constituent features are good is where 2024 D&D subclasses seem to struggle so I have focused on how I approach specific features.
