There was a lot of tweaking, finding edge cases, and rewriting that went into this, so I'm putting the bottom line up front:
Legendary Resistance (X/Day). If the creature suffers a failed saving throw, it may attempt to resist the effects. If all of the effects of the failed saving throw are instantaneous (such as failing to save for half damage), the creature may attempt the saving throw again immediately, otherwise the creature may attempt the saving throw again at the end of the next creature's turn in initiative. This new attempt ignores any effects which would cause the saving throw to automatically fail.
Let's lay out our problems:
- Given a number of guaranteed failures, PCs aren't even trying to force attempts
- Spell effects are binary, everything or nothing
- Baiting out legendary resistances (hereafter LRs) with weaker spells (I'll argue this is a feature and not a bug, read on)
I notice two more:
- All of the above leads to an encounter being done and dusted by the time a caster can finally land the crippling spell
- LRs are as effective against high DCs as they are against low DCs
And our goals:
- Allow tactics blunted by LRs to still be at least partially viable
- Viable across wide range of CRs, and with or without LAs
- Avoid the 'save or suck' problem
Immediately I will argue that "weaker spells to draw off LRs" is not a bug, but a feature; it is a mind game in its own right. Caster has limited spell slots, dragon has limited LRs. Both have the option to take short term risks in return for long term success, and it's a balancing act: does the caster use big spells that make LRs more likely, or small spells that don't have as great an effect? does the dragon use LRs often and risk running out, or rarely and risk suffering bigger effects and losing the fight with unspent LRs? This already resembles some problems in game theory, TCGs have this same mind game with creature removal (which is where I started to really grok the concept), poker has the choice of whether (and how much) to raise, etc., and if you're facing something big and intelligent enough to have LRs, it'd be a little jarring if you didn't see this kind of mind game.
One other effect I've seen that limits the effect of control effects on a legendary creature without completely nullifying them is this: allow it to work normally, but for a fraction of its normal duration. I encountered this in one of the Mystery Dungeon handheld games, where control effects still worked but had greatly reduced durations, but I also encountered this effect a lot as well in Dungeon World, as results on 7-9 often stipulate that you get what you want but only in the here and now: Zuko Style's effect lasts only a moment, Ordering a Hireling makes them do it now but come back with demands later, Parley requires you to give some concrete assurance which may backbite you in the future in order to get what you want right now...
So, with all that on the table, we come to the proposal that opened this answer.
The intent of the rewriting of the first clause is:
- to allow an effect that would have succeeded if not for a burned LR to affect the creature for a very short time, eliminating some of the determinism.
- to favor the kind of control spells that give your allies an opening, rather than do the deed themselves. A canny team of wizard and rogue, for example, can see the rogue readying their action to attack after the wizard lands a hold monster, in order to benefit from the advantage (and the Sneak Attack). A barbarian might not get sneak attack, but at least still gets advantage. Still runs the risk of the target outright succeeding without needing to use LR, but still allows for awesome party combos that make everyone feel awesome.
- for the creature's use of LR to represent a continuous struggle against the effect until breaking free, preserving the necessity to decide at the time of failing the save whether to burn the LR and try again next round. They can always choose to stop resisting and let it go, but if they do let it go, they're stuck with it.
- to allow the relative difficulty of resisting the spell to still be a factor in how hard it is to resist. Before this change, the save DC still affects whether the target needs to use one, but if they do, it only takes one regardless of the save DC. Under this change, if the save DC is high, successfully resisting it takes more LR (and turns) than if the save DC was lower.
- to still provide that possibility, however slim and remote, that the first spell you land will eventually burn through all of the target's LRs if all of their reattempts fail. It's extremely unlikely to happen, it takes at least as many turns as the target has LRs, and it encourages the same early toying with lower level spells to draw off some LRs, but the possibility makes leading with your biggest guns no longer a guarantee of disappointment.
- to preserve LR's effectiveness against non-control spells like fireball.
- to reflect the way I've seen them narrated anyway; failure, yet a moment later, success regardless.
The second clause's purpose is to preserve the original LR's ability to circumvent effects that cause automatic failures.
So, yes, it's a straight nerf to the LR ability, and is likely to get eaten up much more quickly; to counter that, give them more uses per day. But how many more? Let's do the math.
If we want it to be roughly as effective, it should be able to last through the same number of spells at a given save chance. If that's 50%, you can expect six "safe" saving throws before the change: three natural successes and three failures countered by LR. Six may sound like a lot, but if anything, we might want to bump it up to compensate for the small window of opportunity granted by the brief effect until LR succeeds.
Given that 50% save success rate with this change, that's a 1/2 probability of spending none, 1/4 of spending 1, 1/8 of spending 2, etc.
A quick trip to Cross Validated's question on how many coin tosses before a heads to confirm the math, and the expected number of LR charges spent on a single spell comes out to 2 (under RAW LR it was 0.5). So although none of my trial runs have had a fight last that long, that would indicate that quadrupling the charges from 3 to 12 would make it effective against the same expected number of spells (though, again, you may want more than that to compensate for the moment of vulnerability with the new version).