FIRAXIS: Bug reports coming from MT V testing

[DeepO]

First of all, admins, I know the strat forum is not really a place to post bug reports, but as they are a direct result of the testing of the extreme scenario "Last stand at the Alamo" (see the thread 'testing the Alamo' for further info), I hope you'll not move it out of here. It would be nice to keep these threads close to each other.

Most of these bugs are of the AI, or could mean possible improvements / complaints about how the AI handles things. Saves are available on request, but as I have to wade through 50-odd saves to find them again, not too many requests please. I'll update when I find more of these, as I'm fairly certain more will surface. This is such an extreme situation, and as we are keeping a close look on the AI to make sure it stays playable (e.g. not being overrun by 600MIs and 300MAs in one turn when you only have 40 MI and 10 MA to hold them off), several weird things are noticed.

1. Stealth bombers / fighters are shot down way to often
In the two first tests, all of my stealth fighters and bombers were shot down on their first attempt to bomb jet-protected territory. This is not a simple bad-luck-with-the-RNG, I'm talking about 4 fighters and 2 bombers being shot in the first turn, twice (in the two first betas). The third time, when the fighters were modded to have extra bombard power, they survived, but I still lost the bombers. This is way too often compared to what intercept chances would suggest.

It might be that there is a hidden mechanic at work, but as this is undocumented, and highly annoying please look into this. I have heard the same complaint from other testers (Vel), but haven't seen any other reports of other games on this.

2. transports on convoys will attack subs in their way
I had an instance where a transport and her destroyer encountered a sub in their path on auto movement. It was already a questionable decision of the destroyer to attack the sub, and yes, the destroyer got killed in the process (regular destroyer against a elite sub, IIRC). This may be no bug, but any human would move the destroyer on tile up/down, instead of proceeding on the original path.

However, right after that, my now 2HP sub got attacked by the fully loaded transport! Through shear luck, the transport survived, but there is no way an attack of a 1 attack vessel against a 4 defense sub is rectified, and certainly not when the transport is loaded with what I expect to be a combination of MAs and MIs. The same RNG God that let me keep my sub was on the AI side this time, and the transport succeeded... only to be destroyed by a closeby ship in the next turn.

3. AI artillery is not used on offense, even if there are plenty around
In the setup, the Russians are given plenty of RAs. As we all know by now, these will not attack the human player, and this in itself can be considered a serious flaw. However, as Vel was making sure there were enough RAs so that all the defensive roles would be fulfilled, the Russians had additional RAs fortified idle in the middle of their territory, exactly where Vel had placed them. These were not used offensive, and were not even used defensively, while there was ample opportunity.

4. 4HP elite units will retreat from enemy territory, 1HP units will attack when next to human
Another thing that is strange to say the least: in the game a few of my fortresses got overrun by German MAs / MIs. When bombarded so they will lose 1 HP, they will retreat no matter the possible uses. I saw 4 HP MAs retreating en mass, while there was a 1 HP MI city within reach. In itself this is an easily exploitable AI tactic, made only worse because if you have a unit next to a damaged MA, it will attack, against all odds. I saw a bunch of 1 HP MAs attack a MI fortified on a fortress on a hill. No way they will win this.

5. units will load onto transports en masse
In the quest to get into impossible situations, the MT V gives each AI some 200-odd MAs and MIs, over 20 transports, and let the human fend this off with his initial 40-some MIs and navy. Risk based analysis would suggest that when you see the human navy is the stronger one, movement over land is preferred instead of coming by sea. But, the Russians in this game will invariably load all that they can onto transports that take an estimated 15-20 turns to reach the human coast, while going over land (attacking a chokepoint) would only mean a movement of 3-4 turns. Our solution was of course to diminish the amount of transports the Russians are given, but it should decide for itself that the land based approach is preferred instead of risking 20 turns over sea when the human navy is stronger then their own.


More to come when I spot it, I saw some very strange movements of AI troops as well (like the time when one of my conquered fotresses got mysteriously abandoned, a troop of 7 MIs just left, leaving 1 MI and 1 MA behind. I don't know where they were heading, so can't post it as a point were the AI can improve)

DeepO

[vulture]

[SIZE=1]
1. Stealth bombers / fighters are shot down way to often
In the two first tests, all of my stealth fighters and bombers were shot down on their first attempt to bomb jet-protected territory. This is not a simple bad-luck-with-the-RNG, I'm talking about 4 fighters and 2 bombers being shot in the first turn, twice (in the two first betas). The third time, when the fighters were modded to have extra bombard power, they survived, but I still lost the bombers. This is way too often compared to what intercept chances would suggest.

It might be that there is a hidden mechanic at work, but as this is undocumented, and highly annoying please look into this. I have heard the same complaint from other testers (Vel), but haven't seen any other reports of other games on this.

DeepO

Since I know you've done the 'steal plans' thing, can you tell us how many AI interceptors there were in the area where you were running the missions? The odds are clearly not compatible with a 0.05 probability of each stealth mission being intercepted (to intercept the first 12 missions in a row is a probability of around 1 in 10,000,000,000,000,000 - so we can rule that out if the RNG is working correctly). So perhaps the intercept chance is the chance for each aircraft on air superiority to intercept a stealth mission. If you are attacking an area that has (for example) 10 jet fighters defending it, each with a 5% intercept chance, then there is a 40% chance of an interception occuring (10 is also too small a number to reliably get the results you report - 10 jet fighters would have a 0.09% chance of intercepting 6 attacks in a row, and a roughly 16 in 1 million chance of doing it two games in a row). If there is something like 15-20 jet fighters in the area, then that would fit in with this explanation, more or less.

[DeepO]

Vulture, that could be the hidden game mechanic I was talking about... Indeed, there were a lot of jets there, but not 20 of them. More like 10... I'll look into that when I get home. Plus, quite a few of the jets were bombarding, the AI can't have them both bombarding and set to air superiority, can it?

Then again... the range of air superiority is half the bombard range, I thought, which could indeed mean there were 20 jets defending on the squares I was attacking. (coming from multiple cities) If it works not by having a 5% chance for each fighter calculated after another(1-0.95^20 = 64%), but as a simple addition (20*5%=100%) it would explain the results.

If it does work like that, it should be changed anyhow. I can live with a 64% chance that my stealth is shot down when crossing enemy territory protected by 20 jets, but find a 100% not realisticly. So now Firaxis is not only given the problem, but the solution as well

DeepO

[DeepO]

6. Drafting button disappears from City view when cycling through cities
When you draft something from the city view, and then either hit your arrow keys, or click on the arrows with your mouse to proceed to the next city, the draft button does not reappear. You have to exit the city view, enter it again, and there it is again. Minor update problem, but can be annoying when you want to draft in all your cities.

DeepO

[DeepO]

Re: the stealth bombers being shot down. I stole their maps again tonight, and indeed, there are more then 20 fighters in the vicinity. It is impossible to tell how much of these are set to air superiority, but if, from the 30 odd jets in Moskow, 20 are set to air superiority while the AI knows (I suspect them to have a spy) that I only have 2 bombers, I'm posting that one as bug 7. A little bit of reason would be welcome...

Further, Re: bug 2; the transports attacking subs: I've seen it again, in a slightly different version this time. Now I took out the defending battleship with a sub, but my sub still was in the original path of the transport. Instead of going around me, they attacked the sub. Result: 8 dead MAs

DeepO

[Jaybe]

The 'blood lust' is on the Russian AI. They think they have you outnumbered so bad, they don't mind losing the transport cargo.

[DeepO]

If only they were Russian... it was a Chinese transport. But then again, the Chinese sure can get bloodlust too

DeepO

[DeepO]

7. use leaders immediately, not at the end of the turn
Another thing I've noticed a few times already, but is very apparant in MT V: when elite MAs win a battle and generate a GL, these stay in place until all the other units have fought. Only then will they be moved to the nearest city (together with a defender). This might be done because it would be otherwise be too much, but as you start off with some 15 elite attacks in MT V, there is a big chance that the AI could generate more then one leader in one turn. Now, this is not possible, as they already have a leader.

It would be super cool if a just generated leader would go to a city, build an army, fill it up, and use that army on the front, all in one turn. It would add diversity to the game. Now the army gets build, but all available MAs have already fighted, and therefore can not be instantly loaded into the army...

DeepO

[DeepO]

8. In some cases, moving the carrier will not automatically move the airplanes on it too
See attached pic. Yeah that's right, Aeson has his own private jet, and not just one of them, but a whole squadron

Additional information: this is in 3950 BC, and I haven't touched the jets yet. They were on air superiority when Vel gave the save to me. I just moved the USS Aeson's pride (carrier) together with the other ships in the attackforce, they are now in the upper left corner of the screenshot. Normally the jets move fine, I suspect it has something to with not having touched them already.

DeepO

[Velociryx]

Hmmm...this is an interesting one....I wonder...if you put them on air superiority again this turn, will they "catch up" with the carrier on their own, or crash for lack of fuel, or just ....stay there?

I'm also wondering if you have to "rebase them" to the carrier to make them behave again....very interesting one, that....

-=Vel=-

[DeepO]

Vel, if you wonder very hard, I might be tempted to reload... now I just rebased them. The problem is, of course, that once rebased I have to wait another turn before they will control the skys again, but, no worries, I counted that I'm still 9 tiles away from the nearest city, and as I can see their carrier (I damaged it a little) I shouldn't be bombed.

At any rate, it is annoying, even if it doesn't happen much.

DeepO

[Velociryx]

Nahhh, don't let it interrupt the flow of your game or anything....just something to keep a curious eye out for as you're going along with it.

-=Vel=-

[DeepO]

Here's another one. This time, I have two jets floating in the sky, I'll set them to air superiority and see what will happen. One reload isn't that bad, if I can time it so I won't have to do all the unit movements again, it's fine.

On second thought: I'll use one of them to bomb something, I wonder where they will return to: the sea square, the carrier, or just crash? If it crashes CivIII, this is not a simple annoyance, but a major bug.

DeepO

[DeepO]

And another. This time, it is highly annoying, as I am sure that the ships that it was supposed to protect will be bombed. To make sure that they were protected, I first of all set these fighters to air superiority before moving my carrier... with the same result. Please don't tell me that throughout the game I have to wake the F15s, move my carrier, and set them on air sup again, the turns are already taking a lot of time without this

DeepO

[DeepO]

Vel, the next turn all the F15s stayed in their place, they didn't run out of fuel (what would have surprised me anyway, that was a CivII thing).

Further, the one I used to bomb (one of my own squares, nothing else was in range) just went back to its place on the sea, I guess I have a floating air field now. I can't say whether the air sup one is really doing something or not, for that someone else should attack me, or I should wait until the end of the turn and see whether it gets activated or not. I don't feel like that right now, I want to play this game!

Oh, BTW, don't get me wrong in that I'm only posting negative things on the AI here, this scenario also brings a lot of previously unseen AI tactics on top: the AI will use massive bombing, use transports very effectively, and is a pain in the *ss on the high seas. It's just a shame that we should set it up for them in a special scenario, AIs should be able to use these in normal games to greater effect as well.

DeepO

[DeepO]

Turn 2: If I now move my carrier, even with the F15s on air up., every goes like it should be. A very strange problem indeed.

DeepO

[DeepO]

9. Transports should see the risk of staying on the sea too long, and drop their troops asap
I take back what I just said about the effective use of transports The AI do, in general, but I just saw something else. The first time I went through the turn, one transport moved out of a city, and dropped its load of troops (7 MAs and one longbowman(!)) on a hill in the same turn. I thought: great! They are circumventing my line of defense, and instead of going through the chokepoint they go around it. Clever.

As I had to reload for the F15s to be positioned on my carrier instead of on a floating airfield, I couldn't help myself and possitioned a conscript MI on the hill, so the transport could use the grassland aside of it to drop their troops. This scenario is already hard enough as it is, a little cheating while testing should be possible

Now, the problem is that that transport did not drop its troops again in the same turn, but decided to go on for another empty hill on the other side of the sea... passing my attackforce of 5 destroyers, a few subs, a BS and a Aegis. And I'm pretty sure it is exploitable: If I would take my conscript MI, and move it to that other hill, I'm sure it would return back to the original, passing my attackforce again.

Sure, they are protected by a regular BS, but how long do they think this will last? No, when deciding where to drop off troops, the distance a transport has to move on the sea should be taken into account. I can understand that MAs on hill will be better protected then on grasland, but now they will get lost on sea, without even tying up any of my land troops. And it is way too easy to exploit this: if I put a warrior on every hill and mountain top, except for one very distant mountain, I can set up lines of destroyers towards that lone mountain, which will bombard the transport and their buddy on the way for free, so I can pick them off at the end of the line by any single destroyer I like. Where to land is important, but so is how you get to there...

10. stack AI transports with their defense in larger packs on purpose, instead of by coincidence
In MT V, the transports are divided so that large stacks of these will enter the human coastal area at the same time, making it very hard to kill them off without the use of nukes. However, this should not be done by the scenario builder, but by the AI itself, they should see the advantage of moving a lot of them together, like humans do.

In MT V, you can see lots of them moving in packs: some stacks of transports, followed closely by other small stacks, followed again by other stacks (think 2 transports - 4 transports - 2 transports, all spaced 2 tiles apart). Again this is primarily done by unit spacing in the scenario, and not a 'conscious' decision of the AI. But it should be easily adjustable to build it into the AI...

These packs create maybe the largest threat of all: you can't nuke them all, as you don't have that many nukes available, and they are very hard to kill because by the time you spot them, there isn't enough time to gather you navy and kill them all. Some of these will get through.

11. Adjust the defense of transport according to the needs
Closely related to the previous points. Now it is fixed to, whenever there are enough ships (BS, destroyers, whatever) one of these is coupled with a transport, and they will move together. But this is not always ideal: sometimes, certainly when facing an able navy, it pays to have 2 defensive ships on top of a transport. Sometimes, when there is little navy defending, 3 transports accompanied by 2 destroyers woud be better. The fixed coupling gives again rise to exploits, and makes it easy for the human to intercept AI transports: you know they will come in two's, so having 2 destroyers waiting you can pick them off after bombardment.

DeepO

[DeepO]

12. When AI transports lose their buddy, they should head straight for land (any land) and drop of their troops instead of continuing
Quite self explanatory, this one. It used to be so in the last patch that they just waited until another buddy came along, but now they will continue on their way. Unless they are hit, then they will return home.

Besides, why do hit transports head home so often, when they are less then a move away from the enemy territory? Again, this could take several moves, passing hostile attack forces... first of all drop the troops off, after that the AI may try to get his transport at full health again.

DeepO

[dawidge]

Originally posted by DeepO
12. When AI transports lose their buddy, they should head straight for land (any land) and drop of their troops instead of continuing

Besides, why do hit transports head home so often, when they are less then a move away from the enemy territory? Again, this could take several moves, passing hostile attack forces... first of all drop the troops off, after that the AI may try to get his transport at full health again.
DeepO

Empty transports are expendable. Loaded transports are priceless, but serve a single purpose: point A to point B. Do not pass go, do not collect $200. Drop your load, then go home. If you live, you get to do it again (and again...). If you don't, I'll just have to rush another one to take your place for the next wave.

Stack those transports. In Babylon and On, I had 6 transports with two destroyer escorts conducting the majority of my landings. I could guarantee that I'd soak up two attacks against the convoy before a transport got damaged. While we're on the topic, A transport should never be the preferred defender when a capital ship is in the same sector. That wounded destroyer exists to die saving the transports.

I have never seen the AI use a ferry system between landmasses. If the landmasses are less than 5 sectors apart with conveniently located port cities, then funneling units between landmasses is trivial. If more than 5 sectors, it just takes another transport in the middle for the troops to "load" onto when the first one runs out of mobility. The AI needs to understand about redirecting mobility costs. Again in Babylon... I had a stack of three transports in Chicago. Chicago-Satsuma was 5 sectors. I had a heap of MA and armies in America and needed them in Japan, but had only 5 airports in America. Simply pumping them through the ferry system delivered 24 MA every other turn to the front line (+ 5/turn, plus whetever I was producing on the other landmasses). If the AI had this problem, it would try sailing around the Japanese/German continent to land the troops on the front instead of ferrying them to a railhead. It might take 8 hazardous turns to get a transport to the front, but the troops were able to fight on that turn because I could ferry them. Lucky for me that the French had destroyed all the Japanese subs, but my transports sailed right past a Japanese BB each turn, only occasionally taking ZOC damage.

If the AI had a better understanding of how to use transports effectively, I would have lost the game. Instead, it would drop off 8 troops (a rare occurrence) where it needed 20. More often than not, it would only drop 2-3 units, and those would be delivered onto very hostile shores. The AI's use of transports is an inefficient method of disbanding those troops.

[DeepO]

Indeed Dawidge, I'm glad you approve. Now If only Soren would post a tiny message saying that he has seen this thread, and takes it into consideration...

More of these will come, I think that with a couple of extra rules, the naval combat of the AIs would greatly improve.

DeepO

[DeepO]

13. AI will always go for the least defended tile, which gives rise to exploits as you can move your least defended tile
This one, I alluded on in one of the higher posts, and it is known for some time now. But in this scenario it becomes very visible, and it is very hard to not exploit it... as you have two chokepoints, shifting troops from one side to the other will mean that the now better protected chokepoint will see all it's attackers leaving for the other one. As they arrive there, it is only natural to concentrate some of your defenders there, which will prompt the AI to move back to the first... and again, and again.

Brinoch posted a tactic to gain more out of this yo-yo by bombing the AI troops on their way around the inland sea, which is a fine tactic, but shouldn't be possible to this extend. Where the line between exploit and tactic lays here is unclear, if only the normal troops were moving, it would be very good, but as the yo-yo troops are also here, it is less admirable... At any rate the AI should not be fooled like this.

I am unsure how to solve it. But, similar to the transports that risk too much being at sea for turns and turns when they could have a (less favorable) landing spot just one turn away, this maybe could be improved by taking the turns it will take an army to reach its target into account. After all, if an army has a certain strategic goal (with its 'desirability' factor, not all goals are equal), surely the suddenness is also a factor: if an army can attack a goal in one turn, that should be better then attacking a sligthly easier goal in 5 turns. But the problem of course is the use of slightly: it will be a balancing problem to make sure that in all situations this extra 'turns moving towards the goal' factor works, and doesn't destroy the strategic importance factor already in place.

DeepO

[alexman]

Good job DeepO

Even though Firaxis would argue that most of these points are not bugs but AI enhancement requests, I posted a link to this thread in the 1.29f bugs. I hope you don't mind.

Perhaps Soren will take a look here and tweak the AI for PtW.

[DeepO]

Thanks alexman. And indeed, most are improvements, not bugs. But that doesn't mean that changing them wouldn't make civ even more fun to play

DeepO

[Catt]

Originally posted by dawidge


The AI's use of transports is an inefficient method of disbanding those troops.

[DeepO]

14. Funny things happen in the save game popup
I know from experience that no programmer wants to mess with popup screens, as these do not add anything to the core program but are simply things you have to spend time on, but as they are so basic it makes it even more unforgiving when something goes wrong.

I've got two things I've noticed lately. First can be seen in the attached screenshot: I just loaded the final alamo .sav, and, in windows explorer I made a folder where to put the saves of this game. By coincidence, these two (the .sav and the folder) have the same name. After I set up everything for the first turn, I wanted to enter the folder, and save my game there for future reference. However, the game doesn't let me do that, and thinks I want to overwrite the folder with the .sav... not very nice.

The second is less important: If I do not use the deafults saves, and select a .sav already in a folder (let's say "Alamo 3950 BC.sav") which name I want to change, there is a problem with the selecting of characters: If I select the "5" in "Alamo 3950 BC.sav" to change it into a "0", the game will select "90" and replace it with "0". The new file name becomes "Alamo 300 BC.sav". This is just a wrong counting of what is selected, I don't know it by heart and it will probably depend on which compiler is used (I'm mostly familiar with Visual C++), but my guess is that the GetStartSelect() method is 0-based, where the code presumes it is 1-based. Minor detail, but as I was posting bugs, I thought I should post everything I saw.

DeepO

[DeepO]

15. default popup closing buttons
While we're at the popups, it can be tedious to have to close most popups with your mouse, it would be good if the 'ok' buttons got a default Enter to close it, while the 'cancel' would get a Esc attached as the default key. Another minor thing, but searching for those buttons with your mouse all the time when your hand is already on the Enter key can be too much energy wasted, especially when you are, like me, partly paralyzed in your right (mouse) arm... I have to concentrate each time I move my hand from the keyboard to the mouse, and this concentration can distract from the game

DeepO

[DeepO]

Regarding the second part of 14... I was wrong, and am sorry for that. It seems to be a problem with my mouse selecting, If you start very close to a character, it will take automatically the one in front. I don't have this problem in other applications, but it is not something a programmer can help. If I very precisely select something, there is no problem, it is just that when I'm a bit careless I see it. It seems the margins are a bit too small as opposed to the general OpenFileDialiag from windows.

The first 14 thing still stands though, I had to rename the original file before I could select the folder.

DeepO

[DeepO]

Hmmm... it maybe not a simple 0-based, 1-based thing with the selecting of text in the save game popup, but there is definately something strange going on... for instance simply clicking in the name will always select one character... oh well, never mind.

DeepO

[DeepO]

Oh, and I found why the carriers will move without their fighters (bug 8): The fighters aren't loaded on the carriers. If you give any other order then Load to the F15s, they will stay in their place. After load, you can immediately wake them, set them to air sup, and they will move correctly.

So the title of bug 8 should change to: Airplanes should auto load onto carriers when in the middle of the sea when a new save game is opened

DeepO

[Velociryx]

16. Stacks of two or more units can be set to "Fortify All" even if individual members IN the stack have no movement remaining

This probably falls under the heading of "minor exploit," but in the context of the Alamo scenario, it can make or break you. Example: Turn one, I pillaged all my tier one forts, and retreated back to "The Wire" (tier two fortlines). The units I used for pillaging had no movement points remaining, and so could not fortify on their own, but because they were stacked with other units, right clicking on the stack allowed them to fortify anyway, using the "Fortify All" command.

-=Vel=-