Walkthroughs for custom maps.

I wanted to ask hardcore map enthusiasts, and not just them, but everyone – what do you think about "increasing" the difficulty of battles through RNG?

I can consider myself an expert on this topic, at least in the practical field, not in calculations.
As Marmot rightly pointed out, this is a partially philosophical question and depends on the player's needs: what do they expect, a playthrough on the first attempt or catching a favorable set of circumstances?
For me personally, battles with well-thought-out tactics are more interesting than simply killing everyone on the first attempt. There is no limit to the number of replays for me.
I can say that there are individual maps that I have played through, and I have previously made reviews of them in this thread, which are based on the principle of RNG, in particular:
1) the presence of resistance in the enemy unit + 3 artifacts for resistance, and +/- in the same battle, RNG with unicorns / stacks of fairies or sorcerers (both yours and the AI's),
I can name even more twisted models on which battles are built:
1) on Xylo, there was a battle in which it was critically important for me that the AI, after over 100 rounds in the battle (and before that, there were also random events), would start summoning WATER elementals to spend mana, and not earth or air. Since water elementals are two-hex units, and I was sitting behind a force field with an Archdevil (for speed) and a Ghost (for -2 mana per turn). I couldn't explain to myself the principle of summoning a specific type of elemental, because by the time they are summoned, I and the AI have the same armies and the same buffs/debuffs.
In general, only in every 8-10th battle did it summon water elementals, so that the battle could be won.
2) catching unreal combinations right away in one battle (I also described this before): in a battle with the enemy - Solmyr, you need to catch your own "resistance" to his chain lightning spell in the first two of his consecutive casts. In the battle, there are about 120 Archdevils, and about 10-15 die from one chain lightning. At the same time, frenzy is applied to your Archdevils, and again, in the first round, you need to definitely catch double morale (under frenzy), ideally with luck :). Otherwise, the army of Archdevils will die in the first 3-4 rounds, and there is no resurrection.
another similar example: on one of the maps, to complete the battle against Lacus, you need to simultaneously catch double fear from the opponent's phoenixes on your azure dragons, and under frenzy, deal double damage with Archdevils on morale.
another example: when attacking a Castle, kill the opponent's sorcerers with archers, while scaring another specific stack of the opponent, so that the third stack moves to the location of the killed sorcerers, and the AI does not make sacrifices.
3) RNG with the passage of dynamic maps (management of the AI hero's movement) - where they will go, in which direction, where they will spend the night, what we need, etc. A separate topic, see reviews of Woxin maps.
4) RNG of how a 30-minute battle will go, and how much mana the AI will have left (0 or 1-2 for teleporting from the Castle when the catapult is destroyed).
or an attempt to conduct a battle for 20 minutes, and then, without amulets, try to make sure that the AI does not run away. It is quite a challenge to plan the last 2-3 rounds of the battle in such a way.
In short, I think everyone can come up with many examples.
There are individual maps that are designed for the random occurrence of certain factors. This is what makes Heroes interesting, in my opinion!
Coming up with such a task for the player is not easy. But completing it is interesting!
Such variability, in my opinion, shows that the map author understands complexity and introduces an element of interest.