OK, let's clear some misunderstandings:

1. Responses regarding act 3: issues we are talking about concern early-mid game. The fact that the game is unbalanced and gradually gets from easy to very easy is completely different issue. This is typical for Larian, in DOS2 we had the same thing.

2. The problem with misses is that initially everyone misses (including these 20 goblins attacking you), it's slow and just boring. Therefore I switched to Karmic and surprisingly got better experience. I hit nearly always but they also hit me nearly always, everything was faster, more substance. But it broke low tier DC spells and generally felt "not right", breaking the underlying system which was certainly not designed for that. Therefore an idea to fix that mode (which is an option anyway). Also note that the fix actually does not make hits more likely (as KD currently do), it makes them exactly as likely as they should be, by ruling out any "local anomalies" in RNG.

3. This is not about making the game easier. Exactly opposite - I wrote clearly "difficulty is equal to zero", that means the game is already easy (zero is rather small magnitude number...). One of culprits is almighty and all-tempting F5/F8 allowing you to basically circumvent anything the game throws at you. Imagine Dark Souls with quicksave/load even in the middle of a boss fight. Would it still be Dark Souls? My proposal limits it a bit. You can still save mid-fight and reload, but most of the time you will get same outcomes because this RNG is semi-deterministic and its state is saved. (It does not fix stupid AI and other problems but these are a matter for different discussion).

4. It does not alter balance, because basic properties (mean and variance) are same as in standard RNG. You can tell the difference with advanced statistical methods (do not use it to generate your crypto keys!!!), but for video game it is more than sufficient.

5. The effect is primarily psychological, the "bad luck" factor is eliminated over short term, equal to 20 uses of particular RNG instance.

6. Finally I'd like to point out once more: when I say this RNG works in cycles, that absolutely does NOT mean there is always the same sequence of outcomes which repeats. The sequence is randomly shuffled. You always get all possible numbers 1 .. 20 in random order. That means there are exactly 20! = 2432902008176640000 possibilities for a cycle (as Google says) which is quite a lot uncertainty -- less than in 64-bit linear congruential RNG with maximum period, but anyway that's a lot. More than age of the universe in seconds for instance. Practically we may assume it never repeats.