Auto-translated
Okay, based on the significant problems identified:
Rakshasa: upgrade < alt, increase initiative from 8 to 10, plus possibly swap speed (7 for upgrade, 6 for alt)
Lizards: upgrade > alt, increase alt's chance to hit from 25% to 100%
Crossbowmen: upgrade < alt, reduce alt's damage from 2-8 to 2-6, increase upgrade's range from 3 to 4
Demons: upgrade < alt, increase upgrade's hit points from 13 to 16, increase explosion damage by 2-3 times (or more)
Succubi: upgrade < alt, reduce alt's initiative from 9 to 7
Skeletons: upgrade < alt, increase upgrade's damage from 1-2 to 1-3
Vampires: upgrade < alt, change alt's stun factor from 1 instead of 3 (approximately 30% chance instead of 70%), increase upgrade's hit points from 35 to 45
Dancers: upgrade < alt, reduce alt's damage from 4-6 to 2-5
Lukari: upgrade < alt, remove direct hit or enhanced arrow from alt, plus possibly swap initiative (11 for upgrade, 10 for alt)
Shields: upgrade > alt, increase alt's damage from 1-2 to 2-2
Bonecrushers: upgrade < alt, increase upgrade's damage from 3-8 to 6-8
Fighters: upgrade < alt, increase upgrade's hit points from 12 to 15
Shamans: upgrade < alt, increase upgrade's mana by 16
Roughly, without using calculators. It's not a fact that these are the best changes. But it's almost certain that these are appropriate changes (in the right places and in the right direction).
What's good here: minimal number of edits (1-2 per pair, small number of pairs), no excessively sharp changes in stats (all within 20-30%), no insignificant bonuses that don't change much (like +2 defense), no creativity/arbitrariness in adding new abilities, and an ability is only removed once (where it is clearly redundant), in most cases, the general logic of the units is preserved (characteristic aspects are not averaged), and the key characteristic of unit perception - speed - is not changed (with the exception of the swap for Rakshasa). It's not a fact that these are the best criteria for selecting changes, but the very fact of their existence reduces arbitrariness (subjective desires) and adds to the consistency of the systematic approach.
Next, a little creativity regarding frankly failed tiers:
Wyverns (expensive, weak, and without blood): increase growth from 2 to 3, or make them available on week 1 (lower the threshold)
Necrodrakes (very expensive and weak): significantly (by 2 times?) reduce the overall cost of building dwellings, both in general and for mercury
You can also remember units that are subjectively worse and have weak abilities. For example, the weak flame of dwarven dragons (increase it by 2 times). Such things, although not as obvious as the imbalances above, have a pleasant side effect - bad abilities become better. At least, it can be considered in a vote. Plus, you can raise other subjective questions, for example, in the Nightmare pair, is there an imbalance between upgrade and alt? And if there is a large number of votes confirming the imbalance, only then come up with an adjustment (for example, reduce the alt's aura damage).
Furthermore, you can look at races. We see that the Elves have had their 2nd and 3rd tiers weakened; you can increase the growth of fairies by 20% (as an option). Or, since the 4th tier of Demons has been weakened, you can make it available 3 days earlier (lower the threshold). These are not independent changes (there is no point in simply strengthening fairies), but rather consequences of the previous ones.