AGA9: Always War

Sullla AGA9: SPGT23 "Always War"
v.62
Noble Difficulty
Japan - Tokugawa vs. AI Team of Alexander/Mansa Musa/Montezuma
Great Plains Map Script

Note: Tokugawa had his original Aggressive / Organized trait pairing for this game, which he would retain until the Warlords expansion shuffled all the leader traits around. I also went a little bit insane doing the Always War testing, making this my longest report by a good margin.

I dove right into this game as soon as it was announced; Always War was one of my favorite variants in Civ3, and I knew it was only a matter of time until we would be running this stress test on the wartime performance of the Civ4 AI. Judging by the number of people who tried their hand at this game and reported, we can safely say that Civ4 Always War is just as difficult and addictive as it was in Civ3. For all of the problems I came across (and there were many), I still take that as a major positive conclusion. Now our task will be to get into the nuts and bolts of the AI to make it more competitive in these situations while still retaining that sense of fun for the player. I'm confident that we can do both.

Here was my start. I don't think I actually changed anything from the situation Sirian gave everyone in the save file, although I thought for a while (probably close to 15 minutes) before ending that opening turn. Having played several Always War games in Civ3, I well knew that the player is most vulnerable in the first 20-30 or so turns of the game. With that in mind, I made the easy decision not to move the starting warrior out exploring. If he were to be even 4 tiles away from the capital when contact was made, an AI warrior could walk right in and end my game. I didn't want that to happen! He was staying put right where he was. I debated changing the research path, but I checked the tech tree and saw that Hunting is the cheapest way to get to Archery, so I left it unchanged. In Always War, there is no time to go chasing a religion or start out with a worker tech. In Civ3, I would start an Always War game by researching Bronze Working (for spearmen) first if I didn't have it initially, even over Pottery for an early granary. I applied the same principle here to go for Archery as fast as possible. I had better have archers when the AI shows up with warriors, or this game could end REAL fast! (I see from some other reports that this did in fact happen to a number of people.) I left the initial build on barracks because I wanted to build a scout first; I figured I would put 4 turns of work into a barracks, then build a scout to start seeing the map, and the scout would finish right around when I discovered Archery and could have Kyoto go right onto work on one of those. OK, that was the plan. Now time to advance to 3960BC.

I hit spacebar a couple times. Hunting discovered in 3840BC -> Archery, as planned. The barracks work in Kyoto is swapped over to a scout. I get the message that Buddhism has been FIDL [founded in a distant land] in 3640BC (no loss, as I have no time to chase after an early religion). Kyoto's borders expand in 3600BC and I rearrange the tiles somewhat to use the plains "buffalo" (cows) instead of the grasslands one. I also finished Archery at this point and began researching Animal Husbandry, as that seemed to be the most important tech to pursue next. Now that I had Archery, I also dialed the tech rate back a little bit, to 90% I think, to start accumulating some money on hand for emergency upgrades. I figured it was better to delay some techs for 1 or 2 turns and have some upgrading cash on hand than to get the techs sooner but be caught in a potentially disastrous situation without any money! Those upgrades already saved my bacon in SPGT21 with the 5 City Challenge, so I knew they could potentially be critical here in Always War.

I get my scout out in 3480BC, then start on an archer. (This was the turn that the AI won a domination victory when I first fired up the SPGT23 save. After Aeson posted his fix, I started again from 4000BC but made exactly the same moves both times.) And it's not long before the AI civs show up with one of their scouts - the game's afoot!

3320BC - that's turn 17, so not a lot of cushion. The AI indeed is very aggressive with its scouts here in Civ4, definitely more so than in Civ3 where one of the Always War strategies was to turtle and avoid meeting AI civs for as long as possible. On archipelago maps, you could do this for a very LONG time, which I always thought was pretty stupid and defeated the whole purpose of Always War... but I'm getting off track here. In any case, gonna be some surprised faces in the Civ3 community if they try something like that in Civ4 with these aggressive scouts. The Greek scout was no threat, of course, but a Greek warrior showed up only 2 turns later in 3240BC. Now I was very glad indeed I had saved some money for potential upgrades and kept my warrior fortified in Kyoto for the whole game! Here was the situation super-early on:

As soon as I spotted the Greek warrior, I had dropped science to 0% in order to upgrade my warrior to an archer. You can see I'm just going to make it with one turn to spare (the upgrade is 30g). I guess I had pushed things a little too much just by squeezing in the scout before the archer! In any case, I did upgrade the warrior to an archer in 3160BC with the Greek warrior 1 tile away from Kyoto, and on the interturn he successfully defended the city from attack. Not much of a chance for the warrior to win that battle, but still WHEW! In 3080BC, Kyoto built its first actual archer and I was in pretty good shape. Meanwhile, my scout had chanced upon the Malinese capital and came across something interesting:

Warriors could be upgraded to archers at this time; this was later removed and warriors given the 25% defensive bonus inside cities to compensate.

Timbuktu is EMPTY! And I cannot take it with my scout, because scouts can't attack. Argh! Needless to say though, 3160BC is 21 turns into the game, which is more than enough time for a human player to get a warrior down here and walk into the empty city. The AI definitely should not be leaving itself open like this. Sending out the initial warrior to explore is fine - but then the second build should probably be a defensive unit that STAYS in the capital to protect against an incredibly early human rush. Right now, the AI is so scout-happy that it's leaving itself too open to attack in the extremely early pre-3000BC era. Now on the higher difficulties it gets enough free units that this can't happen, but I think there's room for improvement here on Noble as well. I'll go back and check things more thoroughly in debug mode (Ctrl-Z) a bit later, but this would appear to be a problem. At the very least, when the AI is at war with another civ, it shouldn't leave its capital so exposed even this early in the game! If I had popped a free warrior from a hut with this scout... Game Over for Mansa Musa in 3160BC.

I had survived the initial Greek rush, but they still had some other units in the area to throw at me. I made a note in 3080BC that I successfully baited a Greek warrior into attacking an archer with a warrior across a river. The AI seems to have little knowledge on how to fight a war on the tactical level; I can fully accept that we're probably unlikely to get it to succeed on the strategic level, but tactically there should be room for improvement. At the very least, the AI units shouldn't hurl themselves across a river at a player unit if they can simply walk across the river and make the same attack. More on this later, as I found other examples of AI tactical blunders.

After going scout -> archer in Kyoto, it was 2 turns away from growing to size 4, so I put two more turns of work into a barracks and then started a worker in 3000BC when the city hit size 4. Yes, I did not get out an early barracks in this game; there were simply too many other things that needed to be done first. I would much prefer to have 2 unpromoted archers than 1 archer with 4XP! An archer ate another Alex warrior snack in 2960BC, and that appeared to be the end of his initial rush of units (I didn't see any more Greek units for a while).

Judaism was founded in 2680BC; Hinduism had already been founded in 3120BC. Each of the AI civs got one religion, which may have been chance or it may reflect the fact that each civ pursued its own religion, then backed off after it got one of its own. I'm going to have to look at some of the savegame files I took using debug mode in order to see which one it was. So while there's some good news here about no one running the trifecta or getting HinJewism, there remains a serious problem with the AIs. They are TOO hungry for religions in the early game. With all of them at war with me, they should have put more focus on military techs like Bronze Working or Horseback Riding. Instead, even when at war, they are still going after the religions like crazy. Ideally, the AI should be able to adapt to player aggression by pursuing a more militaristic research path themselves. If they were beelining for Monotheism to get to Judaism for a THIRD religion, already having the other two... well, it's not really a good sign. Still, I'm not going to make any final judgements until I take a look and see exactly what it was that they were doing.

More AI tactical blundering: look where this scout ended his turn.

Scouting out my territory is all well and good, but this unit did not have to end its turn next to Kyoto. Unless that Greek scout is on a suicide mission, he should have moved differently. I can't remember which tile he started from, but I distinctly made a note that the scout did not have to end his turn next to the capital. Even across the river might have made my archers at least think twice about attacking versus not attacking! As it was... *crunch* *crunch* mmmm, yummy snack for my archers.

Meanwhile, I had finished the worker in Kyoto and went back to clean up the long-overdue barracks in the capital. After finishing Agriculture, I researched Bronze Working so that I could place my first settler in a location that would grab copper (no forest chops to be had in this starting position!) After popping out another archer, Kyoto finally went to work on a settler in 2040BC. This would be extremely slow for a normal game - but Always War is not a normal game, and if you play it like a normal game, you're going to run into trouble at some point. The AI had been extremely quiet for a while; were they building granaries or city walls or settlers? I didn't know, but they weren't attacking me. I was secure enough that I could send out some archers poking around as well; as I knew from Civ3's Epic Fourteen, it's possible to mess up the AI pretty badly by sending pillagers into their territory extremely early. In the multiplayer environment, I believe this is referred to as the "choke" - not trying to kill your opponent, just stunt his growth so that the eventual outcome is never in doubt.

Let us see what happens when I apply the choke to the Civ4 AI.

First up is another case of poor tactics from the AI. I've sent out this archer, who has followed in the footsteps of my scout and entered Greek territory. On the minimap, the red dot at the top right corner in my scout, the red dot in the titled box is of course this archer, and the red dot in the west is a second archer heading down to investigate Mali. I saw a Greek archer coming and decided to fortify my archer on this forested hill to see if he would attack me. If the unit declined battle, I would have had to shadow it as it moved towards Kyoto in order to be safe. Sure enough, however, the Greek archer charged up the hill to attack my unit (which got a +100% defensive bonus) and was slaughtered. Umm... yeah. Tactically, we can do better with the AI here. I'm not asking for the AI to comprehend the full strategic situation, but it should know well enough NOT to give battle under this kind of condition. There was no chance for this Greek archer to win. With the incoming threat to my territory neutralized, I was able to move the untouched (and promoted!) archer deeper into Greek territory on my Epic 14-style mission.

Kyoto finished its first settler in 1760BC and went onto a second worker (needed to go with the new city). I was able to continue building these non-military units because my territory was at least temporarily safe from attack. The interesting stuff continued to take place in the AI lands, where my archers were moving in to investigate and disrupt whatever the AI was planning. In 1680BC, the same archer in the above picture penetrated to Athens and I got my first look at that AI capital:

There's little to criticize about what the AI has done so far here. Athens is correctly protected behind city walls, and there are three archers protecting it (a good number that doesn't cross the line into defensive overkill for this stage in the game). Obviously I can't attack with my archer and have any chance of capturing the capital. I wasn't about to attack, of course, but merely stuck my archer on a forested hill across a river (uh oh, you can see where this is going) and continued to watch what Athens was doing. The fact that there was an enemy unit only 1 tile away from the AI capital seemed to mess with what the Greeks were doing - but I will get more into that in a minute.

Meanwhile, I had set up my second city, in a spot down by the Malinese capital:

I picked out Osaka's location as soon as I discovered Bronze Working and the copper there was revealed. It's a little hard to see from this shot, but the city has both copper and horses (the only horses even remotely close to the start position) in its first-tile radius. After culture expansion, it also would pick up a plains cow for food; this is why I had it work on an obelisk at zero food to start (I figured the best way for it to grow was to expand its borders and pick up the plains cow as soon as possible). Osaka would serve as my forward base against Mali, and it would allow my to keep close tabs on what Mansa Musa was doing. All in all a great location for a city - if I could hold it. I would backfill Tokyo (my third city) between the two, but I was afraid at this point that Mali would get to those horses if I didn't go there ASAP. That turned out to be a phantom fear, but I think the logic was correct all the same. Kyoto, having finished a second worker, is now back on archers to shore up my stretched defenses.

In 1480BC, my scout ended its turn next to an Aztec archer in the fog, thus ending its life. It had revealed most of the map for me and done a great job, but I was unable to penetrate into Aztec territory, since it was the furthest away from me. This would have significant implications down the road. My archer watching the Greeks had meanwhile observed them release a settler/archer pair to go found a second city; I decided to shadow this pair and see what happened. Only a turn later, the game offered up a critical turning point:

The Greek settler/archer pair is on flat ground (no terrain bonuses) and I have a Drill II archer that can attack down a hill at them. I knew right away that this would be one of the most critical moments of the entire game. There was no way I could pass up the opportunity to attack in this situation, but I was in one of those situations where a Big Huge Dice Roll (TM) was going to determine the complexion of what followed for ages to come. Either Alex gets his second city and I lose my choke unit... or, Alex loses his settler, I gain a free worker, and I get an even more promoted unit choking off his capital city. Let's roll the dice!

As you might guess, since the odds were in my favor, my archer won. Alex's progress will be set back for thousands of years to come, as I continue to choke his capital, and all that work put into a settler went for absolutely nothing. Hey, I actually gained a free worker, which was a nice boost! MAJOR turning point in my Always War game. I had a unit in the right place at the right time, and it just worked out. No need to go adjusting any AI programming over this particular event.

Facing no real heat from the AI civs, I continued to build up my own civilization in relative security. Kyoto finished its second settler in 1320BC and went back onto archer production for a while; I was still in the process of hooking up my copper, so archers remained my best unit. Tokyo was founded in 1280BC and began building a barracks (the first build in almost all of my new cities). In reality, all that each city really needed was a barracks and maybe a granary or library later on, then units forever. Since I lacked a religion, I decided to have Kyoto build Stonehenge for culture and never research Calendar to obsolete it. I had the capital start working on that in 1200BC, after finishing yet more archers.

Here is my choke on Mali. Mansa Musa was by far the easiest to choke, since he was so close to my city of Osaka and easily in range of my road network. This archer simply sat on a hill tile across the river and prevented the AI from doing anything whatsoever. Periodically, the AI would attack with a skirmisher (across the river, onto a hill) but they would never attack multiple times in a single turn, and so my archer outside Timbuktu just kept getting more and more and more experience without ever being truly threatened. With no tile improvements, Timbuktu could not get more than 5 shields/turn, so Mansa Musa was going nowhere fast. At some point, the AI needs to throw everything it has at a unit in order to prevent this nonsense from taking place. Determining when and where it is proper to do this, however, will be tricky. All I can say is that Mansa Musa had more than enough offensive power to kill this archer, yet continued to trickle his forces at it one at a time, which was never going to do the job. And so he sat there, and sat there, and sat there...

And here is the choke on Alex. Again, he has enough archers to clear out this unit, but he continues to send them one at a time, once every 10 turns, attacking across a river and up a forested hill. That's NEVER going to get the job done! And thus Alex and Mansa Musa were stillborn, helpless infants trapped in their cradles. However, Monty was still left unaccounted for (he was too far away for me to try this kind of endeavor), and he still had quite a bit of fight left in him. The game was in good shape, but it was far from won at this point, especially in this 3-on-1 situation. I'll add the next part of the story later today (hopefully, if I get time...)

In my last post, I had successfully managed to fend off the ultra-early attack from Alexander and reached a relatively strong position with three cities by 1000BC. Even better, I had two archer units choking the capitals of Greece and Mali, preventing them from building any tile improvements or threatening my own civ in any serious way. Monty's Aztecs though remained unhindered by me, and I still had trouble ahead of me before this game would be finished. First however, a map from 1000BC:

That shows the position I had reached by the above date. Kyoto was taking advantage of the lull in the fighting to slip Stonehenge in between military builds. Osaka and Tokyo were still getting started, but would be strong once they grew a little. From the lack of the fog of war, it should also be clear where my units were located outside the AI capitals. The one problem was that I had actually grown too complacent, something that is always a danger in Always War. I got caught in the trap of building too much infrastructure all at once. Instead of building some axes and horse archers right away after hooking up the resources, I paused to build some more buildings first, and that proved to be a costly mistake.

Up until 750BC, I had seen virtually nothing from the Aztecs. Just a single scout early on in the game. Well, evidently they had been holding some forces back, because when they came after me for the first time, Monty brought some heat! Two Aztec archers and a Jag showed up together in 750BC. This was the first time I had seen a Jag, so I had to look up in the Civilopedia to find the right counter unit... hmm, looks like axes will work best, even though they themselves are also melee units. Unfortunately, my one copper source was far out in a difficult to defend location. Anyway, the first Aztec archer suicided itself by attacking one of my archers fortified on a hill tile. But there were still two more in the area, and I was forced to retreat back into Osaka for safety. Several dicey turns followed, in which I traded archer for archer and a horse archer for a Jag, but none of my tiles were pillaged. The bad news? That was only Monty's first wave of attackers, and they had taken out most of my defenses. Here comes the second wave!

I've already killed a Jag and 4 archers from the Aztecs at this point, and here comes two more and another one of those horrible Jags. My notes from this point read "Gonna get pillaged... Can't stop it..." I only have two archers in Osaka in the above picture, and I had to strip Tokyo of its defender in order to get the two in there. Clearly, I can't risk attacking the two archers outside the city, so my horses and copper there are toast. And then there's that Jag coming up to the city as well... Yikes, this is NOT looking good here, folks. At a loss for what to do, I kept my two archers bottled up in Osaka and hit the return key.

Cows and copper go "poof" and up marchers the Jag. Oh dear, this is NOT good! Two archers at full strength in Osaka, one with City Garrison I, the other with City Garrison II. But if the Jag attacks down a hill, there's a pretty good chance he'll win the battle, and then the other archer would have to defend against BOTH Aztec archers... well, I don't like those odds. I fully expected to lose Osaka here and to have to take it back. I crossed my fingers and ended the turn.

The blow... does not fall.

Rather than attack the city, Monty's units decide to venture further into my territory to pillage. Ha! Another serious mistake for the AI, although this time it was more strategic than tactical. Clearly, even if it was a coin flip situation, they SHOULD have attacked with those three units. Monty was never going to get a better chance than this to attack me, Jag + 2 archers against 2 archer defenders. Instead, by passing up this opportunity in order to continue on a pillaging raid, the units ended up floundering about and accomplishing nothing. Another Horse Archer (one I managed to build on the very turn my horses were pillaged!) redlined the Jag, and an archer took it out. Then I managed to kill one of the remaining archers with one of my own archers, and the last one suicided himself against a City Garrison II archer, for no apparent reason. The AI did itself in yet again with its extremely poor tactical management of battle. I've made this observation several times before, but I'll say it again: the AI is very good at streaming units into the player's territory, pillaging, and causing disruption, but pathetically bad at capturing cities. At some point in time, this needs to be addressed or we'll be left with an AI that has its fangs pulled somewhat. In an ordinary game, the player should be able to turtle up in his or her cities and survive even if tile improvements are lost. Even here in Always War, by wasting a few turns trying to pillage, I got the time I needed to reinforce Osaka and prevent its fall. I hope Soren can find a way to make the AI stronger at taking cities.

That was the bad news. The good news here is that the Aztecs actually came after me with a stack of enemies! Yes, rather than dribble them in one at a time, Monty really did bring all his forces at me all at once. 2 Jags and about 6 archers total, which was more than my civ was prepared to deal with. I was in deep trouble for a while, and if the Aztecs hadn't been at the end of an enormously long supply line, things might have gotten serious. At the very least, if Alex and Mansa Musa had been in the fight, it would have been extremely dangerous. Unfortunately, after this initial attack, Monty sent little dribbles of units after me in the typical AI fashion. So while I don't know quite what prompted that initial attack, if we can reproduce it and get the AI to fight that way all the time, we'll be moving in the right direction.

During the desperate moments when I expected Osaka to be lost, I had founded Edo to give me a new source of copper. It also pulled in gold and wheat without overlapping with Kyoto, so was by no means a wasted city. Once I managed to mine the copper in Edo, I decided to "get serious" and stop building so much infrastructure; I built about 3-4 axes and put two each in the north and south just to be safe against Jags. With Combat I and Shock, the axes won most of the time against Jags, and I felt MUCH more secure than when I had been throwing axes at the Jags. If you've got no resources, that is one SCARY unit! You can see I was researching Machinery, which would get me crossbows and (eventually) samurai. If I were going to do this over again, I would have headed for Construction first and built the Oracle for an expensive tech like Machinery. Ah well, I guess I'm still learning how expensive all the techs are in Civ4 and which ones to go for sooner. Rob went for the early cats + horse archers and turned in what will probably be the fastest finish (not that this was a competiton game!) I planned to conquer the world with samurai all along, so I'd just be focusing on building up my own civ for the samurai crusade until that point.

As a result of the AI periodically sending units after my choke archers, I ended up with a lot of experience on those units. This guy in particular was impressive:

Drill IV Archer, baby! I don't care if Mali has Skirmishers, there's not going to be much of them left after they get hit by 4-7 first strikes! From what I've seen, the Archer class units seem to be most effective on offense if they pursue the Drill line over promotions over the Combat ones. Of course, getting Combat I opens the route to unit-specific promotions like Shock and Formation, so it's another tradeoff there as well. If Soren got one thing right with Civ4, it's this promotion system - the thing is simply fantastic! It also creates a real attachment to some of the "elite" units; this guy was one of my favorites. Now is there any chance we can get a rename feature for Civ4, especially for units with a lot of experience like this one?

After Monty's big attack, things were quiet for a while. Aztec units continued to trickle in, but I had the forces on hand to deal with them - and with Greece and Mali fully under my thumb, there was no threat from that quarter. I continued to mix military and infrastructure and did not let my guard down again. I founded another city, Satsuma, in my back lines to open up access to iron (and eventually samurai). With some extra units on hand, I even established the rudiments of a multiplayer "sentinel net" to give me some warning of incoming AI units, though I'm sure Fried or one of our MP pros could have run circles around my clumsy handling.

Basically still holding down the fort at this point, nothing too crazy going on. I want to found one more backline city in the southwest where I can get a mixture of floodplains and gold resources, then the next few cities will expand towards the AI civs in order to have some forward bases for the later assaults. Both Alex and Mansa Musa STILL don't have a single tile improvement in place! In the AD years! It's Epic Fourteen all over again, and I don't even have hoplites on hand here. That's not to say Alex didn't WANT to start laying down tile improvements; he had gone to the trouble of building not one, but two workers in his capital:

It must have taken him forever to build each one too, in a city with no improvements whatsoever. Now Alex COULD have sent out the workers anyway and covered them with one of the archers from the city, but he did not do so. I'm not saying I expect the AI to be able to do that - it's probably beyond what we can have it do - but I don't think it would be unreasonable to have the AI attack with multiple units in a row, using the logic of "we might take some losses, but this archer is going DOWN!" (preferably not attacking across the river as well). At the very least, it shouldn't be this easy to competely paralyze an entire AI civ with just one archer by standing outside the capital doing nothing!

Left alone, the stalemate outside Athens likely would have continued indefinitely, with Alex continuing to suicide his archers against mine to no avail. In the end, however, it was Montezuma who rode to the rescue of Alex and got him back on his feet:

Two Jags and a spear headed up to Athens with the clear intent of wiping out this archer unit. Well - score one for Soren's AI programming! This was the most intelligent move I had yet seen from any AI in this game. A squad of units was clearly working together to free up Monty's brother AI civ. To my archer's credit, he took out the first Jag that attacked and redlined the spear, but eventually fell in battle. RIP, brave soul. Alex was now able to move those workers out of his capital and begin making some progress - but he was several thousands of years behind the development curve on his civilization now. I did eventually see some Greek swords, but the danger was long past and Alex was never a threat again. Samurai can do evil things to swords...

At about the same time as my archer up in Greece was taken out by Monty's little task force, the archer I had choking Mali was taken out as well, but what I can only presume was an insanely lucky dice roll from a Malinese skirmisher. (Seriously, did all seven of my first strikes miss or something?!) In any case, Mansa Musa was now free as well - but he was close enough that I didn't intend to keep it that way for long. Since I now had discovered crossbows, I built a couple and sent them down to put Mali back under my thumb again.

"Hey guys, remember me?"

Mansa Musa managed to get out one tile improvement, a pasture on his cows. He never finished farming that floodplains, as the worker retreated back into the city. I pillaged the pasture in 500AD, and that was that - he was stuck back at his old 5 shield/turn again. With two crossbows now applying the choke, I wasn't about to go anywhere. While Alex may have temporarily escaped, Mansa Musa was never going to be able to do anything. He was a complete non-factor the entire game; I guess he contributed research to the AI team, but he sure didn't do much else.

By looking at the icons in the bottom right of the above screen, you can see I had just discovered Code of Laws and founded Confucianism in 425AD. Yes, that's the right date; I can only presume that my choking had drastically slowed the research of the AI civs as both Alex and Mansa Musa lacked any tile improvements at all - much less developed cottages! Tokyo became the Confucian Holy City, and since I had a Great Prophet lying around as a result of Stonehenge, I built the Confucian Shrine on the very same turn. The culture wasn't too necessary with Stonehenge still providing free obelisks, but the extra happiness really helped out.

I was now researching Civil Service, which would enable samurai and allow me to begin my crusade that would finish the AI civs. Short of some kind of drastic mistake on my part, the game was basically decided in my favor at this point. What remained was to play out the endgame, see how well the AI civs could defend their capitals, and see just how fast I could manage to win this thing in decisive fashion.

By 500AD, I was in pretty good shape. Mali was still little more than my puppet, Greece was just beginning to build itself into a proper civ after thousands of years of stunted growth, and the Aztecs by themselves were not dangerous enough to be a threat. With samurai on the horizon, I was beginning preparations for a campaign to sweep them all from the face of the earth. But first, I still had to research a few more techs and set up some forward bases from which to prepare the attacks.

Nara has been founded since the last picture of my civ; it extends my lines out a little further towards Greece, although I wanted another city closer to their borders before preparing a final attack. I saw that some players expanded in this direction first, but aside from the wines there were no resources I wanted in this direction, and so I pushed south for the copper and horses instead and only got around to founding this site late in the game. Evidently the AI wasn't pushing me too hard at this point, as I have 0 military units in production! Kyoto has finally gotten around to its library after ages of having more pressing needs, and I'm slipping in some forges and courthouses elsewhere. Civil Service will get me samurai, but it's going to be a LONG time getting there. I guess I should have maybe built the Oracle and researched up to Construction instead, thus killing two birds with one stone and saving me some time. Live and learn!

In fighting off routine Aztec units a bit later, I noticed something completely bizarre. What is going on here?

Why is Montezuma sending an AXEMAN at me?! For several turns, the Aztecs had two axes coming after me, at the same time that a Jag was also wandering around. Why were they building axes if they had Jags available? More to the point, HOW were they building axes if Jags were available? I thought that the unique unit always took the place of its non-unique equivalent. This may be a bug due to the fact that Jags are resourceless units, so if the Aztecs hooked up copper/iron, maybe that gave them the ability to build axes (?) If so, it's a small bug that can probably be fixed. Regardless of what was happening, Monty shouldn't be building a 5-strength axe when his 6-strength Jag is available with exactly the same stats!

The Jaguar Warrior was originally an absolute monster: a resourceless axe replacement with 6 strength instead of 5 strength though it did lose the +50% bonus against melee units. This was way too strong for rushing in Multiplayer games, especially with Montezuma also being Aggressive for free Combat I promotions, and the Jaguar was shifted into a sword replacement which toned them down to a reasonable and if anything underpowered status.

A couple of turns later, I'm about to place another city that will fill in a gap in my territory. I have the site all picked out to fit in with my other cities, only to get the message that I can't settle. What?

Of course, the problem is that I'm trying to settle on desert here, which is not allowed in this particular version. I ended up moving one tile west and founding there, and it wasn't a huge deal, but still - the location wasn't as good and it created overlap with my other cities. Not being able to settle on desert or ice - wow, what an UNFUN addition to Civ4. I forsee much griping from the community if this is left unchanged. Soren, I know that this change was made because the AI was settling in unintelligent locations, but in this case I think the cure has become worse than the disease. There are so many different occasions when I would want to settle a city in desert or on ice when fresh water is not present: to grab a key resource, to use for military operations, for use as a fishing city, maybe even to fill a hole in my borders. This just comes across to me as something that adds a lot of irritation for relatively little benefit. My strong recommendation is that this be changed back in a future version, and a different fix found for the AI's problem with settling junk cities.

Thankfully this change was reverted in future versions; settlers can place cities anywhere the player wants.

I discovered Civil Service in 760AD and set the research path to Math -> Construction (cats!) -> Engineering (better roads). Those would be the only techs I still needed; everything else would be unnecessary. I revolted into Hereditary Rule... err, Bureaucracy, yes, that's what it's called now, Bureaucracy. My civ switched over to full production on the samurai crusade:

We shall teach them the way of Bushido! Shortly after taking this picture, I hooked up my newly acquired Stone resource and had Kyoto start work on the Pyramids. I was not close to Feudalism on the tech tree, so I decided to build the wonder and take the Police State civic for the XP boost for my units. As far as the government civics column now goes... It's had its teeth pulled completely. Several versions ago, it offered two very strong civics, Hereditary Rule's boost in the capital city and Representation's free specialist in each city. Vassalage was in that column too, giving them player a number of different strong choices. Now Vassalage and Hereditary Rule have been moved over to the Legal column, with Hereditary Rule pushed back to Bureaucracy. Representation was changed too, although probably for the better. The new Hereditary Rule... I can see where it would be potentially useful, but it just doesn't seem to be that good. I thought we had a better balance in the earlier versions, where the player had to consider whether to go for (expensive) Monarchy early on for the boost in the capital, at the cost of having to delay a lot of other research paths. Now, I just see no real reason to research Monarchy aside from moving on from there to Feudalism. I never researched Monarchy AT ALL in this entire game, and that would have been unthinkable before. The (old) Hereditary Rule and Vassalage together pose a problem, but why can't we just move Vassalage back to the Government column where it was before? That would seem to work better to me. Outside of the Religion column, I'm not seeing any early civics worth changing to currently, and that's a crying shame in my opinion. Moving the capital bonus from Monarchy back to Civil Service appears to me to be a step backwards, making the tech choices in the early game less interesting. My two cents.

This was played just after a major civics change: Hereditary Rule civic used to have the familiar Bureaucracy function, 50% more production and commerce in the capital city, while Bureaucracy unlocked cash-rushing. The problem was that this was too powerful when the player only had a handful of cities, making a Hereditary Rule beeline the best choice in almost every situation. Soren therefore moved the civic's benefit back to Bureaucracy, by which time there would be more cities on the map and pumping up the capital would be less impactful, with cash-rushing moving all the way back to Universal Suffrage. I was griping about the change here but Soren was correct and this swap needed to be made.

As far as the game goes, I discovered Construction in 900AD and proceeded onward to Engineering. Osaka, the closest city to Mali, began producing a steady stream of cats. I also saw at this point that Monty had captured the barb city of Cherokee up near the Greek border, so I took it from him (it was guarded only by a Jag) and thus gained a forward base myself near Alex. By 1000AD, I had moved a sizable stack into Mali's territory and only needed another turn or two to bring down the walls. In 1020AD, my Samurai sliced apart Mansa Musa's skirmishers while barely breaking a sweat and Mali was destroyed. (I swear I took some screenshots of this, but I couldn't find them later in Civ4's screenshot folder. I wonder what happened?)

The next task was to close in on the remaining two AI civs. I had a gang of four workers in the south laying down two road segments a turn from the Malinese border to Azteca in order to speed my army along (had a lot of practice in doing that from Epic 22, should anyone remember that). Meanwhile, as I closed in on Alex, I discovered that he had produced another settler.

"Oh no, not again! Why does this keep happening to me?"

Poor Alex. Two times he builds a settler, and on both occasions I have units in place to take it out immediately. Just to be safe, however, I shadowed this unit until I could get a crossbow and a samurai in place to kill off the sword defender, so Alex did have a second city for about 3 turns. That didn't exactly cause any serious problems.

Thanks to all the roads I built, my samurai force actually reached Azteca slightly before my second batch of units reached Greece. Teotihuacan, Monty's only city aside from his capital, was reached and captured in 1110AD. Four cats brought the walls down pretty fast. After that, it was basically over for Monty:

Montezuma has actually done a pretty decent job of building up the land around his capital, since he was left alone up to this point. The one big exception? Forested hills around the capital! I keep seeing this again and again in my games; it's like the AI is putting up a gigantic "attack from THIS tile" sign. The AI would do better anyway if it replaced these forests with mines: more shields from the tile, the forest chop shield bonus, and an inability for attackers to use these tiles for defense. Otherwise the AI workers seem to be doing a pretty good job, but there's definite room for improvement in that regard.

Not surprisingly, I wiped out Monty on the next turn... or did I?

I'm confused here. Montezuma has no cities left, but still has a score and some units wandering around on the map. Is this supposed to happen? I killed off the Jag in this picture, then captured an Aztec worker that was still lurking around, but couldn't find any other Aztec units around. I actually changed a bunch of build orders in my cities to scouts, thinking that there was another Aztec city on the map that I would have to find, but no, they had no other cities. When I ended the turn and 1070AD began, the score for the Aztecs went to 0 and it properly showed that they were gone. This was all very surprising, since the score for Mali went to 0 immediately upon taking their capital (I did not have to wait another turn). Did this effect happen because Monty still had some units alive when I killed off his civ (?) In any case, it was very confusing and should be cleared up. Even if we're going to leave the units for a civ alive after all their cities are gone (which would be OK), the score needs to go to 0 if all cities and settlers are gone. Otherwise, it's much too confusing as to whether or not the civ is still alive somewhere on the map.

At the same time that I was finishing off Montezuma, my forces were moving in for the kill on Alex as well. On the turn after I destroyed the Aztecs, I got the message that I had triggered a Domination win.

Was that 75% of the map? It sure didn't feel like it... For good measure, I finished off the Greeks anyway on that turn. Always War never feels right without winning a Conquest victory anyway.

I killed Mali in 1020AD, Azteca in 1160AD, triggered Domination in 1170AD, and conquered Greece in 1180AD. It took a while for my forces to get started, but once they did it was a pretty quick sweep across all three AI civs. I'm also pleased that I was able to coordinate my forces well enough to finish off Alex and Monty only two turns apart at the end of extremely long supply lines. That kind of timing isn't easy! Also included in the picture is a Combat IV/Shock Axeman, which was the unit that got the most experience for me in this Always War game. I was waiting to get him to Combat V and then upgrade to a samurai, but that never happened. Oh well.

Overall impressions... Always War remains just as much fun as ever here in Civ4. It is significantly more challenging than in Civ3, however, especially if the AIs are teamed together so that they can share research (I don't know whether that will be the usual way that people will choose to play Always War, but it's a nice additional option to have). Without the bombardment units of Civ3, 1-turn barracks healing, and those ridiculous armies in Conquests, the absurd kill ratios from Civ3 against the AI are not likely to be seen again. On the plus side, however, the promotions system and the combat engine are both better than what existed in Civ3, and I forsee much fun ahead for the community with this variant. But - no more Deity Always War games. That never should have been possible in the first place, and I'm glad that Civ4 will be fixing some of the gross exploits that players were pulling off to win such games in Civ3. Hopefully Soren can look at the evidence piled up in this SPGT and get a better handle on how to improve at least the tactical side of the Civ4 AI.

Next up: going through the savegame files I took in debug mode to investigate exactly what the AI was up to at different points in time.

While playing this group test, I specifically took a number of different save files at various points in time to see what the AI was up to later using debug mode (Ctrl-Z). After finishing, I went back and opened up these saves from 1400BC, 675BC, 25AD, 500AD, and 920AD to take a look at what was REALLY going on. Not all of what I found is necessarily pretty, but it needs to be reported to improve the wartime performance of the AI civs.

1400BC

Here's Mali's capital fairly early on in the game. I only need to post the picture of Timbuktu here because Athens and Tenochtitlan looked exactly the same at this point in time, only with different terrain and different religions in each. The good news is that the city is well defended, with four skirmishers and city walls. The bad news about Timbuktu is... well, almost everything else. This city needs worker improvement in the worst way, but Mansa Musa hasn't even tried to build a worker yet. Please note that this shot is from BEFORE I had any kind of units choking the AI civs; on the minimap, you can see a red dot of the unit coming down towards Mali to do just that, but nothing had taken place up to this point. Until Mansa Musa gets some tile improvements going, he can never grow past size 5 and never get more than 4 shields/turn. The AI civs had also researched a LOT of techs by this point (more on that in a minute) - they had far more techs than I did at this point - yet there has been no attempt to build a granary, library, or even a barracks! The only thing that I can deduce is that Timbuktu has been doing nothing but build units for the entire game thus far. Tenochtitlan was in the same boat, as best I could tell. Athens had produced a settler - but I managed to kill it with my archer in the right place at the right time, thus making that effort for naught. This is pretty ugly, but worse is still to come.

Here's a shot of Azteca, because something interesting is taking place. Alex had sent a wave of attackers after me very early in the game, around 3000BC, then his production of units dried up as he stopped to produce that settler. Mansa Musa's pitiful 4 shields/turn production had been so slow that he had only been able to defend his capital with units, nothing more. But Monty has managed to build up a little army here, with five archers and a Jag present. This is the group that would cause me so many problems a couple dozen turns later; I estimate they must have left their borders around 1000BC judging by when they reached me. So while I don't know exactly why this occurred, if the AI can be made to form little "squads" of units and attack with them instead of individual units, we'll be moving in the right direction. Unfortunately, while Monty's territory looks good on the outside, his capital also has nothing but city walls in it...

The AI tech tree in 1400BC. There's little to criticize here as the AIs have picked up an impressive list of techs, including several important military technologies. The only thing that makes no sense is their baffling skip of Agriculture; none of the AI civs had workers out yet, but this still seems very strange. They would go back and research this after Horseback Riding, but still - huh?

675BC

These next pictures come from when the Aztec attack on my civilization was in progress. Unfortunately, I had a settler unit active when taking these pictures so the food and shields are accidentally displayed on the map (sorry!) Here are three archers and a Jag closing in on Osaka; they all have the AI function "AI Attack", which is probably why they are all heading to the same area. I would like to think that this is a planned attack; however, it is more likely that all of the units left individually at roughly the same time rather than exibiting any kind of group behavior. If these units could be made to form a real "squad" instead of walking in individually two and three tiles apart, the AI would be oh so much stronger as a result. As it was, I almost lost Osaka to this group. If those four units had all moved in at the same time? Not a chance that the city survives.

And here's the other component of the attack, another Jag. But what is he doing way up here at Kyoto, separate from the other Aztec units? I'm pretty sure it's because he has the "AI Attack City" task assigned to him, while the other Aztec units have the "AI Attack" function instead. This is a shame, since if this Jag were in the south instead of the north, Monty would have been a lot more successful. I barely managed to fend off these scattered assaults as it was! So long as the AI continues to treat every unit individually, it will always be stuck in this tactical quagmire. If the AI can be made to attack using groups of units, its performance will take a quantum leap forwards in terms of strategy. I don't know if Soren will be able to do this, but if he can pull it off, Civ4's AI will be enormously more successful at fighting wars. Ideally, Monty would be able to see that he has 5 units heading for Japanese territory, so he would form a group of some kind out of them and send them all at ONE specific target - AND have them get there at about the same time. It does no good even if the AI can pick out a target if all the forces dribble in one and two at a time! Again, this may be unworkable, but this situation illustrates where improvement could stand to be made.

Moving on Tenochtitlan, it looks... basically the same it did in 1400BC. Monty has produced his first worker at this point, who has just finished a quarry on the stone. I guess he went to that resource first because the AI civs still don't have Agriculture to hook up the corn! The lack of city improvements is becoming more disturbing here; all units all the time is beginning to take its toll. Remember, the Aztecs were never influenced by any kind of choking unit in this game at all, so they serve as the control group against which Mali and Greece can be compared. I'd like to say I planned that, but it was simply a coincidence that took place over the course of my game.

25AD

Monty continues to make progress in his lands; with no Japanese units present to interfere with his workers, he has begun to improve his land a bit. There are two workers out, and he even managed to get a second city built. But he has still slowed himself very badly by building all this military; I've managed to get five cities built at this point, and that's with constant attacks upon my land. Anyway, there is something positive to report from this picture. I found that these "AI Reserve" units hang back in the AI civ's lands and do things like protect workers. This is a good sign, the AI building some offensive units but holding them back to use as counter-attackers against a player invasion. This is definitely a plus! Now, what would be nice is if Greece and Mali, the civs being choked at this point, could have built some units with this "AI Reserve" function and then sent out their workers under guard. That would have forced me to attack with my archers or allow their civs to start building tile improvements. Instead, however, they continued to build "AI Attack" units to try and clear out my archers, wasting them in one at a time attacks that did nothing of any consequence. Maybe it wouldn't be practical to use the "AI Reserve" for that kind of function, but it's a possibility.

Here's Athens in 25AD. There are no tile improvements at all still, but since Athens has better land than Timbuktu, Greece isn't in quite such bad shape as Mali. There are TWO workers bottled up inside the capital, along with four archers, all being "held" there by a single archer on the tile NW of the city. There certainly seems to be enough units on hand to either 1) kill my unit by attacking with more than one unit at a time or 2) send out the workers and protect them with archers. Either would have been a better option than sitting here making no progress for eons at a time. Athens also still has no city improvements other than walls. Yeah, that's really starting to catch up to the AI civs now, and it will become more and more pronounced as time goes on.

The AI tech tree. What's wrong with this picture? Hey, you guys are at war, right? Why are you heading for Literature and Drama while ignoring Metal Casting, Machinery, and Feudalism? Hello? I know the AI isn't always going to make the smartest tech choices in the world, but it CAN react to circumstances in the game, right? I would certainly hope that when the AI civs are at war, they would tend to pursue vitally important military techs like Feudalism just a little bit more. But judging from the Casual Wednesday game that Sirian and Aeson and I played two weeks ago, that is not the case with the AI civs. Needless to say, they need to recognize at SOME point that going up the top of the tree just is not going to cut it in an Always War game!

500AD

Something interesting is going on here. A barb city has popped up on the map, and Greece and Azteca have shifted their ENTIRE ARMIES off pursuit of me in favor of attacking this barb city. Stuff like this pulls back the veil of fighting real opponents and reveals the AI to be nothing more than lines of code. Puppet strings, anyone? It's not that there's an exploitable strategy here or anything like that, but it's bizarre watching the AI units change track with one mind and all go off in pursuit of the same objective. The AI also continues to exhibit extreme tactical stupidity here. All of the units circled have the same function assigned to them, "AI Attack". Yet they continue to stream at this city one at a time instead of forming a group; there are two Greek archers and two Jags (one is just off the bottom of the screen) all heading for the same place but arriving on different turns. Even with the AI's ridiculous +30% against barbarians (which is the same the player gets on Chieftain... I still maintain they should get the player's Noble bonus against barbs, which is a more reasonable +10%), those Greek archers stand no chance of winning and are just serving as free promotions for the barb archers. The Jags are no smarter, but since they have 6 strength, they will eventually be successful and end up taking the city (which I will then take from the Aztecs, since they have no way to reinforce it). The need for the AI to form attack "squads" practically leaps off the screen in situations like this. Sending one unit at a time is even worse in Civ4 than in Civ3, since defenders can just get stronger and stronger with promotions!

920AD

This brings us back to our original civ, Mali. Let's see how well Mansa Musa was managing at this point:

Even worse than in 1400BC! Timbuktu is actually smaller (size 4) and producing fewer shields (3/turn) now than it was over 2000 years earlier. And still with no a single city improvement? Wow. Mansa Musa managed to get a pasture on the cows for a couple of turns after my initial archer was destroyed, but I quickly returned with crossbows and pillaged it. So now he is back to no tile improvements, no city improvements, and a completely helpless civ. Mansa Musa was strangled in his cradle. Destroying him in 1020AD was practically a mercy killing.

But I had been controlling Mali for practically the entire game, leaving them little room to get anything done, so that wasn't the fairest example. The Aztecs, on the other hand, had been left completely alone up to this point, aside from the fact that I was at war with them. They SHOULD be doing much better, as I have never moved a unit into their territory to this date. Let's take a look:

Things certainly do look much better. There are three workers, a number of tile improvements, quite a few Jags running around, etc. The two Jags at the top of the screen are headed out on attack missions, the rest are part of the "AI Reserve" designed to defend Aztec territory. It's strange that there aren't more Aztec cities (Monty has not been under pressure from me and has had plenty of room to expand) and there are a disturbing number of forests that have not been chopped on hills, but otherwise things appear pretty good. The AI clearly needs work at learning to expand while at war (this is one area in which the Civ3 AI was superior; it ALWAYS wanted to expand), but on the surface the Aztecs appear to be doing OK.

Under the hood, however...

...the Aztecs are exposed as paper tigers. There are no city improvements in Tenochtitan, here in 920AD no less. None. In fact, aside from city walls, no AI city ever built any city improvements at any point in time! Not even any barracks!!! No wonder all these SPGT23 games seemed to get easier as time passed - the AI civs did nothing but build endless units in every single game from start until the end of time! And as time passed, their initial spectacular research effort fell apart without the support of libraries, their production began to lag without forges, their cities were unable to grow past a certain point without granaries, aqueducts, and markets, and the player inevitably gained the upper hand just so long as he or she could manage to survive to that point. In Civ3, the AI often neglected building cultural improvements in Always War games, to the point that entire civs would have only the palace for culture. This is Civ3's problem on crack. The AI isn't building ANYTHING but units, ever. There are so many problems with this that I don't even know where to begin. And my guess is that this is taking place in every game whenever an AI goes to war. This is clearly a BIG problem; Soren's wartime AI code is basically broken right now. I'm sorry to be the harbringer of bad news, but better this be found out now than later.

One of the incredible things about working on Civ4 in pre-release testing was the direct, immediate pipeline back to the developers. Soren took the time to read all of these reports that we were posting in the Frankenstein forums and he constantly responded to the feedback that we were providing him. The amount of tweaks and alterations to the civics alone that took place during testing were too many to count. These old reports should make it clear that Civ4 didn't emerge sui generis as a fantastic strategy game, it worked its way there over time through iteration after iteration. Future Civilization games would have benefited enormously from the same process and simply didn't get it.

One last picture, this from the AI tech tree again. In addition to what's shown, the AI civs also still have not researched Metal Casting. Yet they have Drama and Music? The AI got the free Great Artist at Drama, but didn't end up using it; I never saw it in any of their cities (even in debug mode), so I have no clue what they did with it. It was probably merged into one of the capitals. Still, the total avoidance of the military parts of the tech tree in this Always War game is very troubling. I look at this and I say to myself, "is the AI really that clueless?" Apparently so. So the problem is not that the AI is over-pursuing religions (it left Code of Laws and Philosophy alone), but that it continues to favor the peaceful top of the tree over the metals line at the bottom over and over again. If the AI is that predictable in its tech path (as it seems to be), we have a problem there as well. I can conclusively say that this was not a very smart path to take in this particular game.

Surprisingly, for all these flaws, the AI still put up a competitive fight in SPGT23. If some of these problems can be corrected, we may have something special on our hands here.