One more time, then I'll let it be.
I don't want the proposed assassin's auto-kill class feature (described in detail in another thread) and the proposed swordmage's big-effects-with-no-save class features to be published because, as described, they do not appear reasonably balanced. It's not an assumption. It's an observation and judgment.
I mean, there are already spells that do this exact thing--where they have a spell attack roll, rather than forcing a saving throw. And, mathematically, it should be pretty much the same thing either way: someone rolls a d20 and adds bonuses, which are usually soft-capped. It's not
perfectly equivalent since (for example) it's usually easier to gain advantage on your own rolls than to force disadvantage on an enemy's rolls, but plenty of features apply symmetrically. Since, yes, attack rolls tend to be easier to juice up, that would warrant keeping the numbers more modest on the proposed stuff.
But...well, you seem to be fixating on "this ONE SPECIFIC proposal, which can be read as being unbalanced-as-stated," rather than considering the whole thing in bulk. Again, I've mentioned my own proposal, which requires a
sequence of successful attack rolls in order to access the powerful spells. That is, in principle, actually a
harder requirement than saving throws, because even if you have a 90% chance to succeed on any given attack, and you need 4 hits in order to cast a particular spell, you only have (0.9^4) = 0.6561, a 65.61% chance to succeed. If you have a more normal ~65% chance to succeed on any given attack, your chance to pull off the overall spell drops to only 17.85%. And most enemies don't have more than an 80% chance to pass a save! That's the whole reason I made that proposal, it actually
leverages the difficulty of iterative probability to make big spells costly but justifiably powerful, while smaller spells are nearly guaranteed to work but far less powerful as a result.
So, ignoring the thing that you found innately problematic (though I disagree with that assessment), would you find
this proposal close enough to the realm of reasonable that, if genuinely iterated on and playtested and not simply yeeted onto the page with reckless abandon, it might actually be fine?