1. Drastic shifts in numbers
I'd like to nerf those numbers even further, and replace the combat mechanics in the process. Instead of +2 for each improvement, I'd like it to be +1.
Give every actor the following offensive skills: Melee, Missile, Magic
And give every actor the following defensive skills: Fire, Water, Earth, Air, Life, Death, Blunt, Slashing, Piercing.
Further, reduce the skill baseline for mundane units to 1, keep the skill baseline for heroes at 10, and go crazy with the monsters.
Make every point in every skill equal a roll with a die, preferably a die with just two sides; side "success" and side "fail".
Continue to use opposed "rolls" (current is "rndAttack-rndDefence=damage"), but of a different kind that involves more steps, like this:
An attack consists of two rounds of opposed rolls.
The first is an opposed roll using the skill relevant to the method of attack (melee, missile or magic). The difference between the number of success rolls of the attacker and the target, are the number of hits inflicted. If the difference is positive the attacks are made on the target as intended, if the difference is negative the hits are inflicted on the attacker instead, assuming the target has the appropriate method of counter-attacking (a peasant can't counter an attack with a spell). Otherwise a negative difference is ignored.
The second round of opposed rolls is to determine the damage inflicted by the hits, if any, and is rolled against one of the victim's defence skills (according to the type of attack the hitting unit made). The side inflicting the hits rolls a number of dice equal to the number of hits inflicted. The victim rolls a number of dice equal to its relevant defence skill. As before, a positive difference is the hitpoints of damage inflicted on the hit unit, but unlike before, a negative difference is always ignored.
A system like this is vastly easier to balance. It also has several other neat advantages over the current mess, of course.
The exploits are just an extension of the badly designed game mechanics and the lack of balance. Depending on your starting resources, another exploit might be tech'ing to bowmen parties and using 2-4 units of those to steamroll everything. It's often faster & even more overkill than the early ball of summoned doom.
3. Too many ideas
I neither agree nor disagree exactly. I think the problem is that whomever designed the game didn't do any designing beyond a lit of bullet points, and never actually got around to try playing the game.
The unit designer and the tactical combat are a total waste right now, but only because the game mechanics don't justify them. If, for example, you used the game mechanics I just suggested and gave equipment multiple skill modifiers, both positive and negative, both unit designer and tactical combat would suddenly be justified. The former would be useful and the latter would be fun.
4. Stack sizes
I have a couple of suggestions regarding this as well, though it's really more an extension of "too many ideas" thing that I prefer calling "bad design".
First, increase the minimum size of trainable units to 4. Or better yet, go metric and make the size categories 5, 10, 15 and 20. It's way easier to remember. Anyway, increase the minimum size of trainable units.
Second, differentiate between army stacks and adventure stacks, based on the size of the units they contain. Army stacks are stacks with unit sizes greater than one. The rest are adventure stacks.
Third, make different rules for what each type of stack can do. Adventure stacks shouldn't be able to attack army stacks and cities. Army stacks shouldn't be able to attack adventure stacks, monster stacks or go questing.
Fourth, to make sure heroes can join army stacks, introduce the Vanishing Retinue Extraordinaré. By which I mean: throw up the unit designer every time a new unit size is researched, and ask the player to design a retinue unit for each of his heroes. Once he's done that, conjure up a gratis retinue of the player's current maximum unit size-1 and make it part of the hero unit, whenever the hero is in an army stack. Likewise, make the retinue vanish without a trace whenever the hero isn't in an army stack (or a city).
These are 4 fairly simple ideas and as far as I can tell they solve the stack problem, the quest problem and the hero problem.
5. Tactical Battle Process
If you combine my various suggestions up to now with appropriate terrain modifiers (as in: related to the actual terrain and in the form of +/-dice rather than a percentage), a pre-battle deployment phase (preferably one that remembers the last deployment), a 500% animation speed option, and spells redesigned to work with the new combat mechanics, I believe it would end up being fairly enjoyable - with the caveat that every unit under the sun needs at least 1 special ability, and that specials for non-trainable units have to be unique. I don't mind if all guys with shields can shield-bash (or whatever), but I do mind if 5 different monsters all have the "hurl boulder" special.
Incidentally, by redesigning the spells along the lines of the combat mechanics I've proposed, I mean stuff like giving a fireball spell +5 magic skill for every size category of the target, or give a melting touch spell +1 damage dice for every hit, or making a spell of Brittleness modify (bonuses and penalties) the target's defensive skills instead of its hitpoints.. Or whatever. It's Yet Another Reason to use such a system instead of the current one: it allows a whole lot more fun stuff.
6. No city damage
Total agreement here. Cities should take damage when they're conquered. Moreover, they should suffer various culture-based penalties. A bunch of goody-two-shoes Kingdom guys presumably wouldn't be very happy & productive under the iron-fisted occupation of some malicious Empire guy.
7. No expansion penalties
Agreed. At the very least I think captured cities should be cut off from the global resource pool unless/until they have a caravan route to a player-founded city. But giving captured cities size-dependent piles of negative modifiers that has to be off-set in specific ways, I think, is a very good idea. Force us to raze cities and make the world hate genocidal tyrants, so we're afraid to do it.
The only expansion penalty to speak of right now, is the built-in deficiencies of the Sins tree thingy (wutsit called, dammit?) and that's a usability penalty, not a gameplay penalty so... Please remove it, yes? Thank you.
8. Random monster spawn
Monster spawns would - in my opinion - be cooler if they couldn't be attacked by army stacks, didn't spawn so often, couldn't attack cities, and were much more prone to eating resource buildings & caravans. It would be a great reason to invest in a couple of adventurer stacks, and keep those stacks relevant throughout the game.
This is the one thing I have total faith Stardock will get right, eventually. I just hope - but currently strongly doubt - it won't be done at the fatal expense of the gameplay.
10. No diplomatic consequences
Yes well, the AI is a bit on the stupid side right now. As for what I'd like to see, once diplomacy becomes relevant because half the game mechanics have been redesigned from the ground up, the game has undergone at least a single balance pass, and the AI has learned to play at least a little bit... Are things like global consequences to player activities, and unintended consequences from mixing bloodlines. I'd love to see the world uniting against me when I'm an ass, and I'd love for some Enemy ruler to try to claim my throne through some combination of the right blood ties and diplomatic capital. Either would be a hugely satisfying way of losing the game.
Edit: @ Damon_Lundy contrary to what you apparently believe, collateral damage becomes more of a factor the lower the level of technology of the involved parties. When your warfare tech is at the melee level, the best weapons are the ones that stop the other guy from getting into melee. Without technology, those are weapons like diseases and wildfires, and what made them so good isn't just that they are indiscriminate. Their greatest advantage was that nobody had the technology to control them. Pond-based bucket-lines & blood-letting are, at best, ineffectual in the face of such things.