A lot of people think positive feedback loops are bad and negative feedback loops are good, other people think negative feedback loops are bad and positive feedback loops are good, but those people are all wrong; as usual I'm right about everything and I'm going to share my wisdom.
A positive feedback loop is when being ahead makes it easier to get even farther ahead. It's present in most games where you control units and win by destroying enemy units (Chess, Prismata, Starcraft), since losing units also makes it harder to protect the units you have left. The general argument against these mechanics is that they cause a small early mistake to quickly snowball into an insurmountable disadvantage such that the player who makes it will lose even if they play better than their opponent for the rest of the match. This can lead to understandable feeings of frustration.
Lunarch Studios, the developers of Prismata, argue that this isn't a problem because players can just resign when they see they're in an impossible situation. And that's a valid argument, for Prismata.
In Prismata you're not disincentivized to resign because the game is very blunder-resistant after the first few turns - that is, it's hard to make a critical mistake once into the meat of the game - and so you know the odds of your opponent slipping up and letting you come back are essentially nil. But Chess is all the about the tactics; it's far more common for someone with a material and/or positional lead to mess up and lose, which makes players who are behind feel like they have to play on incase this happens.
The only reason Prismata even regularly gets past the first 8 turns or so is because it's hard to tell who's winning. In Chess you can easily measure material advantage; positional advantage exists but is less important. In Prismata you have to count not only the Drone difference and attack difference, but the tech difference (which is un-quantifiable), differences in the lifespan of your attackers and the click abilities you have access to, and sometimes even the supply difference; and so players usually aren't certain that they're behind at all until they're severely behind. This means you almost never actually play with the sure belief that you're behind, nullifying all the psychological downsides.
Some people might argue that Go has a similar problem even though it doesn't technically have a feedback loop because decisions later in the game are usually less important (outside of life and death). (In Go it's harder to tell who's winning than in Chess, but not as hard as in Prismata.) But Go has two solutions up its sleeve to avoid the frustration:
The scoring. Even if you fall behind and aren't able to catch up because skill has become less efficient, your disadvantage doesn't snowball into a total loss; there's a non-binary score at the end and you can see that your efforts are still reflected. Even if a loss by 1 point and a loss by 100 points are treated the same for the sake of rating adjustment and tournament results they're very different for player psychology.
While Go is very blunder-prone (much more than Chess), you can actually take advantage of this to solve the effect, because Go has a much stronger ability than most games to choose how "risky" you want to be. So if you can only win if the opponent blunders, you don't just wait around for it because they'll play safe. What you do is start a crazy battle - try to kill a group that probably can't be killed or something - and if you succeed, you got your comeback, and if you fail, you've fallen even further behind from trying and you can now resign sure that there's no chance of winning. You're never incentivized to prolong a game with the feeling that nothing you do matters.
I don't think anyone who goes into Go without a preconceived belief that it has this problem would feel frustrated by it.
And there are still arguably some downsides to a positive feedback loop. The winning player often knows they've won before the losing player knows they've lost, and has to play for a little while knowing the outcome. But this is a small problem because it's far less upsetting to play knowing you've won than to play knowing you've lost.
Making the game effectively over before it's technically over creates some trolling possibilities as well; some people will grief their opponents by going AFK or making moves to drag it out once they know they've lost, which can't happen in a game that mechanically ends as soon as that happens. But this is also a small problem since very few people do this and there's usually a limit to how long it can go.
But it gets worse...
Some people who have a sufficiently strong hatred of positive feedback loops advocate doing the opposite thing and making the game get harder on whoever's in the lead. And worse, this philosophy seems to be much more popular - while fewer games implement it, I've never actually heard anyone criticize it. So I'm going to be the first to do so in a noble attempt to save the game industry. The negative feedback loop is significantly worse for a couple of reasons.
Strong negative feedback loops can make the early game feel like wasted effort in the same way positive feedback loops do the late game. The big difference is that this time it's not escapable through resignation because it happens before the outcome is known. The players are forced to suffer through it, every match.
And not only that, but this may actually increase the winrate of the stronger player. The weaker player who was graced by this effect and took the upper hand will now have a harder time keeping that upper hand. In a positive-feedback game like Prismata, a weaker player who gains an advantage early will go on to win unless they get severely outplayed in the following turns. This is part of how Prismata achieves its valuable performance variance.
Are comebacks even a good thing, anyway?
I'd even question the idea that comebacks are a thing we should be designing for. Sure, they feel great for the player who comes back. But does anyone ever talk about how it feels to have someone come back against you?
As someone who's played some games designed to make that happen, let me tell you, it feels pretty shit.
In Dragon Ball FighterZ it feels just as disappointing as any other game to lose by a landslide. But this game has no feedback loop (at least not much); comebacks are always possible through outplaying the opponent. And despite how great it feels to be on your last character with low health and proceed to wipe the opponent's team, it feels significantly worse to be on the losing end of that than it does to just lose a match from the start.
And that's in a game where the comeback is entirely emergent and can only happen through the underdog legitimately outplaying you. When the game artificially rewards the underdog or creates its comebacks through randomness, it feels even worse. People just learn to stop seeing the game as legitimate competition to protect their emotions.
That last thing I said is something I didn't come up with until I said it. Looking back on my experience with Spellweaver it wasn't really considered evidence of superiority to beat someone. It wasn't embarrassing to lose to a weaker player (and correspondingly it didn't feel particularly rewarding to beat the #1 on ladder), you never reviewed your games and asked what you did wrong, and it was basically as if we weren't saying "I beat DragonBoy328", so much as "My deck beat DragonBoy328's deck".
Go was the polar opposite. Every match of Go I played - and this might be partly because of the Asian culture assiciations and the Samurai music I often listened to while I played - felt like an honorable duel, as personal as it was epic, just my intellect against theirs.
I think comebacks are a mixed bag in general, and I like when games allow for them, but sacrificing other design goals to create them is folly.
In conclusion, my position is that games should have no feedback loop when possible; a positive feedback loop is acceptable if the game's design doesn't force players to suffer through it. The important thing is to always either reward the player for skill, or don't disincentivize them to resign when you won't.