Sullla AGA18: SPGT38 "Land Always War"
Rome - Julius Caesar (Expansive/Organized)
Standard Lakes map, Noble Difficulty, 7 AI Opponents
v.80
This was my final Single Player report before Civ4 released, finished in early October about three weeks before launch, though there were some additional Multiplayer games that followed which are covered on the next page.
Out of the three SPGT games that Sirian posted for v.80, I decided that I would take a swing at the land version of the always war game. I've simply played too much Civ4 by this point to get a fair assessment of Chieftain/Warlord (beyond "this is easy") and to be honest I've done Always War on an archipelago map before, and the ship moving is a PIA. 
Since there were few takers on the land game, I think it worked out rather well. Aeson played the game too, and did so in a drastically different playstyle than mine, which also worked nicely.
4000BC - found on the starting tile, research set to Hunting -> Archery, build order goes Warrior -> Scout -> Archer for safety. Yes, this would set my development way back - but you just can't take a lot of chances early on in Always War! I would need an archer in the capital to be safe, so that's what I went after first. Everything else could wait for that.
A couple turns later Elizabeth's scout shows up, and I'm already at war (3760BC).
My warrior in the picture grabbed that hut on the next turn and got experience from it, which shifting my thinking a little bit. I decided to give the warrior the City Raider I and II promotions and go out to try and cause to mischief among the AI civs. Elizabeth's scout foolishly moved next to that promoted warrior in 3680BC and was taken out with ease. I'll talk more about the scouts later, but for now I should mention that the AI doesn't try to avoid your own units with their scouts much at all when at war, making them easy pickings. Free experience point for my warrior, basically.
Met Isabella in 3520BC. My warrior was in forest next to hers, and I was curious to see if she would decide to attack my warrior in the forest (in the last AW game I played ages ago, the AI units consistently suicided themselves in situations like this). To my surprise, the Spanish warrior did NOT attack my warrior, but moved off to the north to keep exploring. Wow. Way to go, AI! That was by far the smarter move, rather than trying to attack my warrior in the forest, where I could take advantage of the +50% defense. I shadowed the warrior for a little while, trying to see what it was doing, but once I realized it was just exploring, I pushed further south to try and find Spain.
Finally found Spanish borders (with the scout I had just built actually, not the warrior) and started moving in for the possible early kill. Back home, I finished building an archer at size 3, and began a worker so that my development curve could finally take off. After researching Hunting and Archery, I went Mysticism -> Polytheism, but was unfortunately beaten to Hinduism by 4 turns. Obviously it was there for the taking, but I took the safe path to Archery first and thus got beaten out. Most of the game I would end up playing without a religion as a result. Research into Animal Husbandry and other worker techs.
So my warrior gets down to Madrid, and I expect to have good odds to attack the city, since they only have warriors defending and my warrior is loaded up with City Raider II promotions. But... that's not the case! 
The odds are badly against me. Badly! My City Raider promotions only get through the cultural defense of Madrid (it has Buddhism and thus tons of culture, argh) and then I still have to deal with another +50% of bonuses from fortification and the warrior city defense bonus. While it might have hurt me here, this is a great development on the whole. The early warrior rush is basically impossible now, and that's a good thing. Civ was never meant to be about warrior rushes; you can take people out with chariots or axes, but it shouldn't be getting done with warriors. Very positive development. As for what I did here, I wasn't about to suicide my warrior on a basically hopeless attack, so I stuck him on the forested hill next to Madrid and decided to watch what Spain did next.
Also note the clams bug outside Madrid above. And this was a LAKES map! 
Warrior rushes were a serious issue in some of the earlier builds of Civ4. Soren's late addition of the 25% city defense bonus for defending warriors essentially shut down that whole concept, outside of the Incan quechuas anyway.
For the next few turns, Spain produced a worker and then had it sit in its capital doing nothing due to the presence of my warrior. More accurately, it would move the worker out and then move it right back into the city when I threatened it. I could have kept doing this indefinitely until Spain managed to get some real units out to threaten my warrior. Soren, the AI could use some help defending against a choke like this. Right now, it simply holes up in its capital and stagnates. The AI should send out its worker with an escort guard, and eventually train some of its best units available (even if it's archers) to use in a stack to take out the choking unit. Obviously you'd have to be careful with this, but right now it's simply too easy to choke the AI. Sit outside their capital with a unit, keep their workers stuck inside the city, and their expansion gets slowed drastically. It's a problem.
So after a little while of doing this, I decide to try something different. I realize that I can move my warrior so as to get it BETWEEN the Spanish worker and Madrid. What will it do then?
The answer is, "it moves right next to my warrior and I capture it". 
I don't know if there's time or interest in improving the "AI worker runaway" coding, but this was very funny. With two movement, the worker could find some way to outmaneuver my warrior, but instead it just moved right next to it. There's plenty of room for improvement here... That extra worker proved to be very helpful, once I escorted it back to my borders with my scout.
Izzy built a settler in 1920BC, but it sat in the capital for a long time; I suppose that Spain was waiting to build escorts for it (which WAS smart). While I was waiting, I met a few more of the AI civs, built my second city, connected copper and horses, etc. Finally, in 1440BC the settler moves out - but take a look at this:
Izzy has 3 archers in Madrid (good). The settler is being escorted with two units (also good). But... the two escorts are both WARRIORS! 
Clearly, Spain should have escorted the settler with an ARCHER, especially if there were three of them already in Madrid. Look, having two escorts for a settler is rarely a bad idea. But warriors should not be guarding settlers! We experienced this over and over again in this week's Casual Wednesday game, the AI sending out settlers with only warriors to guard them. It was incredibly easy to take over those cities once they expanded to size 2 so that they wouldn't be auto-razed. While it would slow down the AI's expansion a bit in some cases, it would be a good idea to tell the AI to escort its settlers with archers - get rid of the whole 2 warriors as escort idea. That's better than 1 warrior as an escort - but it's still not good. 1 archer is cheaper than 2 warriors - and actually provides better defense in most cases!
Just to finish this story, my warrior shadowed this settler group until it founded the city of Barcelona. Barb attacks had given my warrior enough experience to get to City Raider III, so I figured I would have a good chance to grab the city. But the reality was that even with the +75% bonus, the combat odds were barely in my favor; the defending warriors in Barcelona were getting +50% from fortification and the built-in city defense bonus. That made the odds 3.5 against 3, which were barely in my favor, and... I lost. City Raider III warrior lost to an unpromoted warrior in a city with no culture. And it wasn't a fluke result either!
One thing's for sure: the warrior rush is dead dead DEAD! 
Other issues I noticed early on... when Tokugawa researched Monotheism he swapped to Organized Religion despite not having a religion. Considering the upkeep costs associated with Organized Religion, that was a very weedy move! The AI should not swap to ANY Religious civics until it gets its hands on a religion! Otherwise, it's simply bleeding cash away for no purpose. Organized Religion is the worst culprit due to the cost, but an AI could also switch to Theocracy when it has no religion, and then it would NEVER get one - wow, that would be pretty stupid. I hope there's time to look at something like this, or at least jot it down for attention later...
I had 4 cities up by 500BC, and was making good progress on some wonders thanks to the stone that I had. Rome had already built Stonehenge and was now working on the Pyramids (I needed Stonehenge badly because I had no religion still). To this point I still had seen very little offensive action on the part of the AI civs; they seemed content to sit back and build up their own civs rather than come after me. We might as well not have been at war for all that I was seeing.
One other issue about this map: Lakes is a very difficult script for Always War! The fact that it's all land makes it tough enough, but since the world also wraps around (unlike Inland Sea, for example) there are no backlines or safe grounds to work with. You have fronts on all sides until you expand far enough to get backline cities in the center of your territory. It's difficult having no oceans to serve as safe spots on the flanks... Also, research was simply awful on this map for ages. No water tiles, no trading possible in Always War, and I couldn't start to build cottages until I had enough production to be safe. Even with cottages, Rome is not Financial and so it took forever for them to start yielding real revenue. A lot of long research times in this game, in other words.
One of the next things that I noticed was the AI's total inattention to scouts. I was moving my scout around in enemy territory, ending his turn next to AI units over and over again, and they blithely continue to ignore him. Now this is probably better than allowing humans to bait defenders out of cities with suicide scouts... but it's still not optimal. I scouted out quite a bit of the AI territory before an archer finally did him in. This is probably not important enough to get attention considering where we are now, but it's still not great. You'd think the AI would do something about a scout wandering through its territory when you're at war!
Meanwhile, Persia captured a barb city right on my border, but had only a warrior remaining to defend it. That made for a very easy victory for me:
I mention this because Khoisan provided some insight into the way that the AI was functioning (or at least the way I believe the AI functions). As I mentioned in the previous post, I had seen very little offensive action from the AI civs despite this being an Always War game. When I took Khoisan, I had only a single chariot in range of it, for a city on the edges of my empire. Within a short period of time - before I could build an archer and get it to Khoisan for defense - the AI showed up with some of its own units. Spain sent a chariot after Khoisan and I was left with a coinflip battle for control of the city, which I lost. Khoisan was razed. I would rebuild Khoisan on the same spot, which would reveal more insight into the AI.
I had ages more of inactivity from the AI after losing Khoisan, but after I rebuilt the city again as New Khoisan, the AI suddenly returned and paid me a visit. Take a look:
Two pairs of axes, two from Persia and two from England, both beelining directly for New Khoisan. Since I had only an archer and a chariot in the city, I was clearly in trouble! City is size one too, so I can't whip out another defender. Well, I expect to lose the city, let's see what happens. The Persian axe pair moves next to the city and attacks... and my archer wins BOTH combats and promotes! He only had City Garrison I, so that was quite surprising. Meanwhile, the English axe waits a turn for his partner axe to catch up (as I noted in the picture), so that I get the turn I need to bring in reinforcements and contain the threat. Whew. It was NOT a stupid decision for the English axe to wait for his partner, it just happened to be bad in this case because the English attack was not timed with the Persian one. Even with only 4 units, the attack was quite effective, as I lost an archer and and axe of my own in taking them out.
Now, as for what was going on under the hood here. I was going to talk about how the AI did a great job of sending out groups of 4 or 5, but never quite managed to put a Stack of Doom (SoD) together. But... Soren posted in another thread yesterday that he had deliberately limited the AI to stacks no larger than six. The clear and easy recommendation is that the AI needs to form larger stacks, like 10-12 units. Now it should not ALWAYS do this, as that would be equally predictable, but a mixture of large stacks ALONG WITH the non-stacking default that the AI has been doing all along would be very strong. That would force me to keep some cats near the front at all times to hit their stacks, whereas right now I can get by with using cats almost exclusively on offensive. We're so close to getting this balance right, just up the number of units that the AI is using in its stacks WHILE making sure to keep the scattered approach that the AI has been doing all along!
Secondly, I was wondering why I was never seeing AI attacks for so long, then suddenly got a flurry of activity at New Khoisan, followed by a lack of activity again. Then I got a popup stating that I was #1 in military strength, and suddenly I understood it. My cities are all stacked to the gills with military units. The AI knows this, and knows that it doesn't have a good chance to take any of them. As a result, instead of sending units after me in wasteful attacks that go nowhere, the AI simply holds its units back and defends. When the AI sees an opening, as I did at New Khoisan, it was indeed striking at me with a decent amount of force. Once I reinforced New Khoisan, the attacks went away again because the target was now too well defended.
Generally speaking, this is fantastic stuff, light-years ahead of the AI wasting itself in stupid attacks as it did in previous Civ games. But... the AI is taking this too far. The AI needs to be a little bit more aggressive when it comes to fighting wars. From what I saw in this game, if you defend your cities strongly, the AI will simply sit back and not attack at all, and that's not ideal because you can conquer their cities one by one as they do nothing but defend. Probably too late to go messing too much with the AI at this point, but it was quite noticeable how they didn't come after me in this game. Maybe just make them a tad more aggressive at fighting wars?
My civ halfway through the medieval era. I've just prepared my first real offensive group, made up of cats and elephants, designed to take the city of Pasargadae away from the Persians. Why Pasargadae when there are closer targets? Well, I've had a Great Prophet sitting around just for ages with no religion to spend it on, and Pasargadae is the Confucian Holy City. I can therefore capture it, use my GrProphet to build the Confucian shrine, and then basically "steal" the Persian religion as my own! Much easier than trying to research my own religion, heh. Here we go:
Pasargadae is thus captured without much trouble after its defenses are bombed down with the cats. However, once again stretching out to grab Pasargadae has extended my lines a bit and thus caused an opening for the AI. As soon as I presented a target that was NOT stacked full of units, the AI sent out a group to attack it:
This is the first stack in ages that caused me actual worry. The Persians managed to slip a sword and 3 cats past me by using roads in neutral territory (I indicated the path they took to get to their current position with arrows). Let me be clear: their target is NOT Pasargadae. The AI is heading for New Khoisan, which had largely been stripped of defenders in order to make the attack on Pasargadae. This was another clever move from the AI, and if it had sent 8 units instead of 4, I would have been in real trouble.
What happened? Cyrus smoked some serious weed. Rather than continue on to New Khoisan, Cyrus attacked the axe in the forest across the river. He clearly did this to try and capture the worker that was sharing the tile with the axe (of course, the axe was protecting that worker). The Persians killed that axe, of course, but lost two cats doing it and injured a third, which allowed me to take out the two remaining units easily on the next turn. I don't know why the Persians decided to chase after the worker here, but it was a major weed decision. The initial thrust at New Khoisan was a solid strategy. Why the AI changed its mind to chase a worker, I have no idea.
Another Elizabeth mini-stack comes after me. Is their target Pasargadae or a city further back in the lines? I don't know, because I smashed this little group with my cats and took it out with elephants. Good mixing of units, but the AI needs to send more of them to make an effective attack. Larger stacks needed, Soren!
I also still saw some "lone wolf" cats, which I didn't bother to take pictures of. That's something that the AI should be programmed not to use; a lone wolf horse unit makes a nice pillager, but a lone wolf cat? Not a good use of resources.
By roughly 1100AD I had developed my civ well enough to go on the offensive and start really taking some cities from the AI. On this map I needed to have multiple offensive groups on each side, which was fun. Took a couple barb cities, started carving up the nearest English and Spanish cities. I had clearly passed the hump needed to keep advancing and proceeded into the (very long) mopup portion of the Always War game. With v.83 now out, I decided to stop my game and report what I had found thus far. I was just about to enter the Renaissance era on the tech tree.
Already managed to spread Confucianism to most of my cities; also note I'm still running a profit even with this many cities at 80% research. Noble sure is easier on the finances than Emperor.
Now I opened up debug mode after I was ready to stop this game to take a look at what the AI was doing. Two major, MAJOR problems here had me seeing red flags. First of all, an issue that's been reoccuring for ages: the AI settling useless junk cities.
Look at those cities! Oh my, what is Elizabeth smoking?! Hastings is the only one that is even marginally useful. The others are ALL cash drains that will never amount to anything. I can see founding Liverpool if England was desperate for iron, but really NONE of these cities should have been founded. And if you look at the minimap, you can see that the other civs are all doing this too; Mao and Cyrus have cities in the southern tundra, Washington is building them in the north. The AI must stop destroying its finances with these useless junk cities! I've been seeing this for ages, are we going to be able to take care of this before release? It's not only bad because it hurts their research, it's also bad because it makes the AI "look" bad. Nothing says "I am a compilation of lines of code" more than moves like this, and things that suspend disbelief are never a good thing. If at all possible, this should be cleared up (I hope it can be now!)
But there's another problem that's even worse. Soren, the AI civs just do NOT research Feudalism! In an Always War game, that's about the stupidest thing they can possibly do. In 1250AD when I stopped, not a single civ had Feudalism. None of them have it! They're ALL still defending with archers, and thus I can easily run over their cities as soon as I take out the cultural defenses. Washington is researching DIVINE RIGHT (!!!) and he STILL doesn't have Feudalism. It's almost like they're DELIBERATELY ignore longbows!
Whatever the AI priority is for Feudalism, double it. Then, if the AI is at war, double it again. This is a huge problem, because the AI is letting the human rip its throat out while it stubbornly researches Music and Philosophy. This must be fixed before release - it needs to be fixed NOW. I can't recommend it in strong enough words. Feudalism must have a very high priority for the AI.
Overall though, the AI performance was very positive. It's ages ahead of where it was in June when I did the last Always War test. If some of the things I mentioned here can be implemented, we should be in very good shape indeed for release. I just hope there's enough time to take care of them!
Seriously, just compare the AI performance in this game to the AI performance in AGA9 which was also done under the Always War variant. If the AI still did a lot of dumb stuff in this build shortly before release, it remained far better than most other strategy games.
