I'll have to look at the capture logic again.  If i recall at warrank 8 it subtracts from the ally portal count, so if its 2-2 portals the AI thinks it is 1-2 and tries to capture an enemy portal to make it 2-1 in its favor (really it is making it 3-1 in its favor).  If the AI never gets the portal in its favor it will probably ignore all the other flags.  Maybe I'll randomly send it off to the closest enemy flag instead of a portal.

Reg snipe build may over focus on snipe and do less team stuff.  I've seen him TP to defend things though, so could have just been distracted by other actions.

1) there is no tracking on who has what minion currently following.  The condition is to send a minion: randomly send to an ally that does not have the same level idol as the one being cast.   miriyaka had an idea to track the minion, but it didn't work when i tried to impliment it.

2) Minotaurs run 10-25 priority and demolisher run 15-30, so for the most part demolishers are bought over minotaurs if it has the money.

3) I can probably add a function to check who is the high gold and only send that one back.  Its generally not an issue for normal AI's as only one usually has the money.

4) Experience is not a super high priority item, so it may have bought other items over it.  I can up the priority if its a good upgrade to always get.  Sedna is also a poor shopper.  With her heals she rarely shops and when she does it is usually only a couple items and back out.

5) I bet the silence in the middle of nowhere was based on a health check.  I don't think i put a check for enemies on that part, so it blasts silence when it hits a common retreat level.

Very few builds have custom artifact levels, so they are mostly from the global.  All of them used to be 100, but i have pulled most of them down to 0.

Bracelet of Rage 100

Bulwark of the Ages 100

All Father's Ring 150

Since the other two are cheaper than all father's they will normally be bought before all father's.  The AI always buys the highest priority item it can afford.

I think pounce is fire at will down to a certain mana percent, then it holds it for an interrupt.

The sniper build does tend to stop and snipe whenever it has mana and there is a target in range, so if its a team full of them it may be slow to capture flags.  i set the mana threshold at 650, anything below that and it should do more team stuff until it gets mana and a target gets in its range, then it is back to sniping.

I think bman? said the UI couldn't pass values to the sim?   something made it difficult or impossible.

Yes, the UI lua cannot touch any actual units.  The engine UI obviously can, as it sends orders you give units to the sim, but lua cannot issue any orders or get handles to any units on the UI side.  This is to prevent a whole boatload of exploits (both of the cheating and desyncing variety).

Re: tracking healer assignment, this should be very easy.  It's just a matter of testing code to find what works.   Stick these utility functions in AIGlobals or AIUtilities or wherever, and see if you can get them to work.  If they cause an error, and you can't figure it out, let me know.  This is a fully solvable problem, it just needs the right approach.

The conditions in ShouldAssignHealer might need to be expanded to include GENERAL category checks and subsequent healer type comparisons, etc.  It should be easily expandable to do healer-swapping between two generals.

AssignHealer also needs to set some kind of flag on the assigning AI or demigod unit, so that it knows where its healers are.  You might already be doing this - I haven't looked - so I didn't implement that.

This should also be adding OnKilled callbacks to the assignee demigod that tells it when its healer is freed up.  This isn't difficult at all, I'll just need to do some review before I can tell you how to do that.

lua\common\Items\generalitems.lua

In a loop for the number of healers spawned, the second healer spawned is given the guard command to another hero.  Selection process for those heroes is: find all demigods on our team.  of those find ones that do not own the idol we have.  of those randomly assign the spawned minion to them.

On teleport all minions are warped to thier owner and assigned guard command to thier owner.  That is done in \lua\sim\ability.lua, function WarpMinions(unit).   I think i forgot to update the logic there to see which allies own the idol, so it is just a random send of the first healer unit it finds.  Since it is at the health stone even if it only has one unit it will summon a new one shortly, so it is safe to send off its only healer if it has one.

Both are in UseItemActions, and I think both can be hooked nicely to handle healer assignment and re-assignment.

I'll work in this in my spare time; for now, just change both of the categories.HERO to categories.HERO + categories.ASSASSIN, so it's not dumping healers on generals.  You'll also want to modify the WarpMinion method so that it counts healers before it sends them off (just do a table.getn(minions) instead of adding in the loop), because not all teleports are to health stones.  It's better for the AI general to keep one healer if they have it, and then send off the other whenever they summon it, which is already handled in the summoning function right now.

Edit: Actually, looks like the summoning function will have to be modified too.  This is already getting complicated - there's simply no fire-and-forget way to handle healer assignment, and they'll have to be tracked somehow.

Hedgie's build (heal_silence) should not fire Pounce at will at all. Please take a closer look at the build. It gets:

'HSednaPounce02',                     # 16

'HSednaPounce03',                     # 17

'HSednaPounce04',                     # 18

Practically, it means those heal_silence Sednas just waste mana by Pouncing at will for 400 damage for 80-90% of match's duration (not to mention many matches end before levels 16-18 are reached)! They all have Nature's Reckoning. Why waste mana if you can AA+NR with better results?!

ty - good for a test. 

I think I like your reg ai specific coding here.  Snipe reg made for some fun and I think having him as a teammate or enemy should be fun.  I'll try a few all reg tests soon along with a few just adding a reg on a team tests and seeing how things go.

Oh - and side thought - if you want to add it to the coding (if you can), any ai reg on any team should purchase cur1 and fs1. Regs in MP are generally burdened with the citadel upgrades as they are somewhat effective at a distance.  Anyway, add that bit (any ai reg is the primary for early cit upgrades) to the mod if you get a chance.  Thanks!

Will test this build Monday. 

I don't think the AI is intended to be played at those settings. The priorities for artifacts are set higher then fs1 probably because in a normal game fs1 will be available a lot sooner then an artifact to buy, you started with gold on rich + nightmare ai. this nets the ai a ridiculous amount of starting gold and its no surprise they all went for artifacts.

On cataract they should not have gone artifact shopping until their side owns the shop, so it was probably just the standard 20 second shopping to start followed by capturing the first flag (random selection of 3 closest flags with lowest threat).

Once they owned the artifact shop if they were near the shop and had the money then they would buy artifacts. 

I set mage slayer, girdle of giants, bulwark, and all fathers as globally ok to buy.  I also added godplate as an ok item to buy globally.   I added these back to address a complaint of the nightmare AI dropping off midgame.  Nightmare AI's generally have 10-20k to spare before warrank 8, so figured mage slayer is better than banded armor.  If there are specific builds or classes that these items should not be purchased for we can zero out the priority in the specific build.   If any of these is globally bad I can remove its overall priority as well.  

If we want to force builds into having more cash for upgrades I guess we can zero out the global items they should purchase.  This would weaken their hard/nightmare game as they can probably afford upgrades and items on those difficulties.

Have them get Journey Man treads instead of mage slayer.  That might be a little imbalanced against nightmare AI since they could buy them right away, but everyone's better with J-Mans.