- agility: from +2 to defense to +1 to defense (weakened 3 times, considering the weakening of defense itself)

With +2 if he runs 6, this is +12 defense. Unacceptable for any tier 2 creature.

- cavalry charge: from +5% to +3.33%

The effect of the ability equals 1 attack point bonus per square, just as in ToTe.

vampires: torpor is much stronger than retaliatory attack (you are almost not affected, and -1/-1 to attack and defense is nothing), respectively, the first upgrade for vampires is not needed

Torpor duration is reduced. I don't remember how much. From the top of my head, I think the duration is about 1.5 turns.
Also, you don't want to get retaliated against with vampires because they die very fast. I had a very good replay that shows just how vampires thrive when they take hits. I hope I haven't deleted it.

rakshasas: whirlwind is much stronger than retaliatory attack, respectively, the first upgrade for rakshasas is not needed

Not true. Every glass cannon build will pick the first upgrade.

Rakasha +2 initiative proposal is totally overpowered in any mod. 10 base initiative + mini artifact bonuses + haste spell combined with dash ability can bring them up to a 30 initiative value.

Do not look at upgrades just as upgrades but how they fit into hero build.
I don't mind criticism as far as it has ground. We may find something we have missed. Sometimes I do not have the time or willpower to think it through as I am fending off 4 different communication channels.
==================================================================

What I do not like about vanilla Heroes V is that skill paths are unrelated and in some parts just random.
I know you look a lot at numbers, but creature numbers do not make sense when you combine them with hero builds.

From the defense example you gave, it seems all the perks are balanced enough, which is fine.
What I do not like is that the same Heroes V idea of similar perk distribution is taken.
The idea behind MMH55 is that one does not need to know everything to make a decent build of his own.

Basically, all skills and perk trees focus on one idea. For example:
Offense tree -> unit damage
Defense tree -> unit defense
Combat tree -> hero damage and utility
Light magic tree -> increased light magic efficiency
Shatters -> magic suppression

War Machines -> More efficiency out of War machines depending on which tree you pick.
The only inconsistencies are in skills where no skills could be properly found to fit (mostly faction native skills).

Here are some specific examples of what I do not like:
- "Suppress light" perk in Light Magic tree - The tree purpose is to empower light magic efficiency. Why does it have a counter perk inside?
- "Storm wind" in Light - This perk has nothing to do with light. Why is it inside?
- "Mana burst" in Destruction - yes, the text claims to be a destruction perk, but the effect has nothing to do with utilizing Destruction school spells in any way.
- "Chain strike" in blood rage - this skill has nothing to do with blood rage ability. It is not even a stronghold native thing, like for example the "Goblin cart".

There are endless examples like the one above.

Regarding WGE skill wheel:
- I noticed that Magic schools and Shatters have common perks. Why?
- Also, why do Shatters have creature-icon perks? Does this give a population of random units? Why?