AI Survivor Season One Alternate Histories

Eauxps I. Fourgott

AI Survivor Nerd
Joined
May 16, 2024
Messages
401
Now that I am indeed starting to post these, the beginning here will serve as the index for each individual writeup, for quicker reference while skipping the chatter.
Game 1
Game 2
Game 3
Game 4
Game 5
Game 6
Game 7
Game 8
Wildcard
Playoff 1
Playoff 2



(Original post)
A newer but (to some, at least) no less interesting feature in Sullla's AI Survivor series is the 'alternate histories': looks at sets of 20 replays of each game to determine the map's 'true' dynamics. Because they're newer, coverage is incomplete for the older seasons, so I've been running them for Season One for some time now.

Sullla declined to post writeups for a season this old on his website, so instead I have been posting them on the Realms Beyond site. Now that I'm involved in the AI Survivor discussions here, it struck me that it might also be good to post those writeups here as well. My question is: Is there an audience for this content that would like for it to be posted here?
 
Last edited:
I would enjoy reading them as a fan of the series, but I don't find the early seasons to hold as much strategic value due to the change in starting techs. I would place more value on complete Alternate Histories for Season Four's Wildcard, Playoffs, and Championship.

Good luck with replays, writeups, and the fantasy contest!
 
Saxo, I hear you on the "strategic value" point and I think you are correct. However, strategic value isn't really my motivation for doing these; I'm more interested in just learning the dynamics for each individual map (although these certainly have influenced my opinions on some leaders and I think offered good insights from time to time). Especially with having put a lot of time into Season 1 now and Myth having worked on Season 4, I currently don't have any plans to go that direction.

Thanks folks for the welcome - with some having expressed interest I'll start posting these here as well! I'll soon have Games 1-3 here since I already posted those previously. I might post Game 4 this week as well since there's no Season 8 game. (I have writeups through Game 6 saved up but am waiting for the right time to put out the last few.)
 
All right, let's hope posting Game 1 goes smoothly here (eta: looks like it did! After a little fiddling with the images, at least.) I will just post this one tonight, since I'm still pretty new to these forums and not sure if there would be an issue with me posting four times in a row tonight or not. Also I need to go to bed as it is.

I do have paragraph summaries of each individual game and endgame screenshots from some of the replays, and can share them on request.

Original game writeup from Sullla's archives: https://sullla.com/Civ4/survivor1.html

---------------------------------

Important note: Unlike the existing alternate histories for Seasons 3 and 4, which were run under modern AI Survivor conditions (by removing the free Deity starting techs and the Apostolic Palace), this alternate history was run under the same conditions as the original game. As a result, we can’t draw as reliable of conclusions about the overall strength of the leaders involved to apply to new seasons, but the histories are a more accurate reflection of how the original game was likely to play out, so we can draw more reliable conclusions about how typical the real result was.

1718078437180.png


1718078456434.png


The first-ever game of AI Survivor turned out to be the tale of three leaders. Peter, Suryavarman, and Brennus were the top three finishers in the real Game One, and the alternate histories showed this to be an expected outcome as they were the clear power trio of this map. Between them they took all but one of the wins, 75% of all top-two placements, and about 80% of the kills. Almost every game would see them as the top three leaders in some order coming out of the landgrab phase, followed by a series of wars to determine which of the three would come out on top this time. Sometimes one would (perhaps with help from the lesser powers) manage to kill both of the other two, allowing a lesser leader to slide into second place, but at no point in any of the games did any of the weaker three leaders assume a stronger position than all of the Big Three – even Napoleon’s single win was the result of a throw by a clearly stronger Brennus.

Which of the three leaders actually came out on top in any given game would depend on the dynamics of that game’s wars. This was a very violent match, with an average of more than 14 wars declared per game and every single game pushing the war counter to the double digits – in this sense the real Game One was highly atypical, as it only featured 9 war declarations. (It’s also worth noting that the Apostolic Palace inflated these totals quite a bit, forcing early peaces and unnecessary dogpiles. Playing these AHs convinced me that removing it starting in Season 5 was the right choice.) However, aside from Ramesses frequently getting dogpiled and Tokugawa frequently attacking Brennus, there weren’t any clear patterns to the wars. This was a highly aggressive field of AIs and most of them were willing to declare war on almost any rival at any given point in time, and that led to high variance in this regard. Most of the time, one of the Big Three would emerge as the leader after the early wars, and then ride on that to victory, but there were also a couple of games where the leader would get dogpiled by multiple rivals later on and get eliminated anyway. The lowest elimination rate on this map was still 30%, which speaks to the fact that nobody was truly safe.

So why did the games end up this way? I’ll go into more detail below, but the short version is that Ramesses and Tokugawa were completely screwed from the start, while Napoleon had a slightly weaker starting position coupled with a worse AI personality that couldn’t make good on it. That left the other three leaders on top in every game more or less by default, and while the lesser three could prove important in deciding the final outcomes of individual games, they almost never had the opportunity to actually jump out in front. Their best outcomes were to end up on the favorable side of dogpiles, and almost inevitably those outcomes would still see a stronger leader also gain territory to maintain his lead over them. As attested by the results, every leader had at least an outside chance of gaining a second place finish, but it became pretty clear that Napoleon could only win as the result of a particularly weird game, and the other two never had any chance to win at all.

Now for a closer look at the individual leaders:

Peter of Russia
Wars declared: 65
Wars declared upon: 32
Survival percentage: 70%
Finishes: 10 firsts, 1 second (52 points)
Kills: 16
Overall score: 68 points

There was no true juggernaut on this map, but Peter was its clear best performer, winning half of the alternate histories, coming close to the lead for kills, and boasting the best survival rate. This was mainly due to his possession of the most favorable starting position on the map: spacious enough to let him always expand to a good size, out of the way enough that every other leader (except maybe the unthreatening Ramesses) had somebody else whom they were more likely to attack, yet well-positioned to get a share of Ramesses’s, Suryavarman’s, or even Brennus’s land when they were dogpiled. As a result, Peter both had the easiest opportunities to get in front and the least likelihood of an inopportune backstab messing up his game, and he was both competent and ruthless enough to take advantage of his this position. Peter’s path to victory was to get out in front of the pack early on by taking some land from one of his neighbors (most often Ramesses, whom he attacked early in almost every game), and then coast to the win from there. If he could get that early lead, he almost always won; there were two exceptions where Brennus leapfrogged in front of him with a mid-game conquest of Tokugawa, and in the dramatic Game 17 he had only a slight lead and was taken down by an attack from Napoleon, but otherwise Peter did a great job of capitalizing on his advantages. Interestingly, he didn’t have any particular preferred victory method; he won no more than 4 times by any single condition and took home at least one win with all four.

On the flip side, if Peter couldn’t get in front early, he almost always got knocked out somewhere along the way. He never really seemed to stay out of conflict, but would always wind up in a fight sooner or later, and if he hadn’t taken the lead, those fights wouldn’t end well for him. As a result, while Brennus and Suryavarman each had solid odds of taking second place if they didn’t win, Peter only got it once – in every other non-winning game, he was either eliminated or at death’s door when the game ended. (If the games had lasted a little bit longer, his ‘true’ survival rate would have been 55% instead of 70%.) He also managed only three kills across his non-winning performances, less than anybody else except for Ramesses. The real Game One was, of course, one of these less successful performances since Peter’s only early conquest was a single city from Ramesses, although that game was a bit unusual in that he never got in a tussle with a stronger foe. Still, stuck in a weaker position, he never accomplished anything there and was forced to hope to backdoor second place. Overall, Peter ended up being the biggest feast-or-famine leader on this map, one who needed a position in front in order to get anything done but was fortunate enough to get that position a lot of the time.

Brennus of Celts
Wars declared: 48
Wars declared upon: 60
Survival percentage: 60%
Finishes: 5 firsts, 5 seconds (35 points)
Kills: 19
Overall score: 54 points

Brennus took home second place in the real Game One, then went on to also place second in the alternate histories, taking home a top two finish in half the games while also racking up the most kills. At first I thought he would be this map’s juggernaut, as he finished in the top two in five of the first six games while surviving all of them, but later on he cooled down significantly, in particular only winning a single game in the back half of the simulations. Like Peter, Brennus had one of the map’s more favorable starting positions, one that had decent enough space, was fairly insulated from aggression, and featured a weak neighbor to conquer. He also proved similarly versatile in his victory methods, winning by all four conditions despite only five total wins. However, Brennus’s position also proved more susceptible to backstabs and dogpiles, as he was attacked nearly twice as often as Peter and three times was First to Die after suffering an early dogpile.

But the biggest reason that Brennus fared worse than Peter was that his weak neighbor was a much more prickly customer. Instead of the weak and unpopular Ramesses who usually was an easy dogpile target, Brennus got the isolated Tokugawa, who never went down without a significant fight and could be quite the thorn in his side. Brennus and Tokugawa’s fates were near-perfectly intertwined on this map: if Tokugawa was killed, Brennus would almost always leverage the extra territory to a top-two position, much as he did in the real Game One. He only failed to do this in two of eleven such cases and only died in one. Tokugawa was First to Die on three different occasions, always by Brennus’s hand, and Brennus was able to win each of those games. On the other hand, if Toku was able to hold out and avoid death, it almost always meant trouble for Brennus. Across Toku’s nine surviving games, Brennus only survived twice, finished in second once, and scored a single kill. Of course, sometimes Toku was surviving specifically BECAUSE Brennus was dying to an early dogpile, but this also showed that Brennus needed the Japanese territory to be truly competitive. Overall, then, Brennus was an average warmongering AI in a fairly favorable position, and would either conquer his weak neighbor en route to a successful game, or fail to conquer him and not be strong enough to compete.

Suryavarman I of Khmer
Wars declared: 48
Wars declared upon: 58
Survival percentage: 60%
Finishes: 4 firsts, 5 seconds (30 points)
Kills: 13
Overall score: 43 points

The last member of the Big Three, Suryavarman won the actual Game One in dominant fashion, but the alternate histories showed that this was a bit of a lucky result. In particular, his extremely strong start in the actual game was an anomaly that only one alternate history came close to repeating; I’m not sure why he was so strong out of the gate in that game, but in most games here he exited the landgrab either only slightly ahead or else behind Peter and/or Brennus. That said, he was still an important player on this map and a viable one, grabbing a top two spot almost as often as those two leaders. Sury’s path to victory in these games was to snowball, getting a lead via conquest early on and then continuing to fight various leaders until he won by Domination. All four of his wins here were from that victory condition, and he was on his way there in the real Game One as well before UN shenanigans ended the game prematurely. However, actually getting the snowball rolling proved a difficult task.

Sury would fail to win his games in one of two ways: sometimes his central position would prove his undoing, as he’d get stuck in multi-front wars and thus be knocked too far back to compete for the win, if not completely killed. The majority of his eliminations started not with him being beaten by a single superior foe, but with somebody else jumping into an existing war and breaking his back. However, in other games, he simply failed to get out in front, not able to keep up with Peter or Brennus, and if he didn’t take the lead early he wasn’t able to win. He wasn’t able to reliably profit off another leader’s demise the same way the other two could, and so he needed a fairly favorable situation in order to come out on top. Still, even when he couldn’t outright win, Sury did a good job of hanging in there given his risky central position; he survived to the finish in half of his non-victorious games, managed to take second place in a majority of those, and never suffered the fate of being First to Die. Overall, these games painted a picture of Sury as one of the better warmongers while still not a top-tier leader. He was a frightening foe when given the advantage and a scrappy survivor when not, but also didn’t have the skills needed to take first place without an early snowball.

Napoleon of France
Wars declared: 58
Wars declared upon: 33
Survival percentage: 50%
Finishes: 1 first, 6 seconds (17 points)
Kills: 9
Overall score: 26 points

Napoleon was perhaps the biggest disappointment on this map. He had a decent starting position, a bit cramped but still viable and with good access to profit off dogpiles on Ramesses or Suryavarman, but while he was often relevant in these games and had a good shot at finishing in second place, he was never the dominant force. Nappy routinely was weaker than the Big 3 exiting the landgrab, and from that point had no route to get out in front. His best path to victory probably would have been to conquer Egypt and snowball from there, but he wasn’t strong enough to take out Ramesses by himself, and a dogpile would result in someone ahead of him getting even stronger. The one Napoleon victory on this map was a fluke result, where he was clearly the second-place leader to Brennus, only to see the Celtic leader turn on the culture slider when he was TWO TECHS from the end of the tree and then turn it off much later – this resulted in Napoleon winning via spaceship THREE TURNS before Brennus would’ve won by culture. There was one other game that Napoleon came close to winning, where he could’ve taken the lead after crippling a winning Brennus with a late 2v1 war, but then dropped out of the war and refused to re-enter it in time to stop the victory. This one may have been my fault, as I forgot to delete the second observer civ (sitting around to prevent an AP Diplo victory during the autoplayed turns) after the construction of the UN, and so Nappy may have just been plotting war against a foe he could never reach. But the overall point still stands: Napoleon needed some extremely lucky breaks in order to have a chance of winning on this map, and otherwise was stuck tagging along behind the big dogs.

My personal theory is that Napoleon performed this poorly as a result of having a weak AI personality for this competition. He’s too militaristic in focus, and that resulted in him not performing as well in the early landgrab, falling a bit behind, and then having no good targets to snowball off of. Nappy as an AI leader needs an advantageous position to come out in front, one from which he can easily conquer a neighbor on his own early on; he didn’t have one in this game, and wasn’t flexible enough to win any other way. He certainly tried his best, declaring the second-most wars on the map, but ultimately it wasn’t enough. He lacked the landgrab or economic skills needed to be successful in this game.

Tokugawa of Japan
Wars declared: 55
Wars declared upon: 20
Survival percentage: 45%
Finishes: 0 firsts, 2 seconds (4 points)
Kills: 4
Overall score: 8 points

Tokugawa was utterly screwed on this map, stuck with a horrible starting position that would never be kept in AI Survivor today. Not only was he crammed in the very corner of the map, but the majority of the land near his starting position was useless ice or tundra, leaving precious little land nearby that was actually good for settling. Brennus would inevitably take a good share of the land between the two leaders, with an end result of Toku having only four good city sites. His settling pattern ended up being very predictable as he settled the same four sites in almost every game, and once he got to those four, he was out of decent land to expand into.

Obviously, just four cities is not enough to be competitive, and Toku had absolutely no chance of winning and only a slim chance of moving forward in these games. His best outcomes came when he successfully grabbed some cities from a neighbor weakened by other wars, but even then he could only get up close to the level of an average AI after the landgrab phase – and the leaders who had started at that level would by this time be doing rather better themselves. Toku did manage to get two second place finishes in these games, but in both of those he was tagging along well behind a victorious Peter. There was a single occasion where he had an outside chance of victory – an unusual Game 18 offered a lategame opportunity for him and Ramesses to attack a leading Peter together and MAYBE overwhelm him with a 2v1 – but even that was no guarantee, and ultimately Toku instead worked with Peter to kill Ramesses.

But even though Toku was stuck in a horrible position and barely had a prayer of finishing in the top two, that didn’t stop him from trying his hardest anyway, and he became a great underdog leader to root for as I played through these games. Despite his weakness, Toku had the highest ratio of aggressive versus defensive wars, going into battle early and often to try and better his position. He and Brennus fought in almost every single game, but it was usually Toku initiating those hostilities. As mentioned before, those two leaders almost never survived together; sometimes there was nothing Toku could do and he was eventually rolled over en route to a successful Brennus performance, but through his sheer persistence, Tokugawa was able to wreck Brennus’s game and outlast him far more often than it seemed like he should have. While he was never in a position to solo kill another leader, his dogged determination was rewarded with a small smattering of kills across these games. Many leaders would have done much worse when starting in this position, and it’s a testament to Toku’s toughness that he managed to perform as well as he did – he may not be one of the best leaders, but this game offers powerful evidence that you should never count on him to be an easy elimination. All things considered, it was a performance to be proud of.

Ramesses of Egypt
Wars declared: 10
Wars declared upon: 81
Survival percentage: 10%
Finishes: 0 firsts, 1 second (2 points)
Kills: 0
Overall score: 2 points

It may not have been immediately apparent going into the original Game One, but in hindsight it was extremely obvious that Ramesses was dead meat on this map, as the lone high peaceweight builder on a map full of low peaceweight warmongers. It was almost inevitable that he would get dogpiled, and that indeed happened early and often in the alternate histories as well as in the real game. Ramesses was always attacked at least twice, and averaged more than four incoming attacks per game, which is especially impressive considering that he frequently was dead by turn 150. The other leaders were drawn to him like a shark to blood in the water, and as a result he was the first to die on thirteen different occassions. In a typical game, the first noteworthy event was for two or more leaders to pile on Ramesses and partition his territory, after which the rest of the game could commence. There was usually nothing that Ramesses could do when faced with so many attacks; he fought the best that he could, sometimes holding out against multiple foes for an impressively long time, but the weight of numbers would eventually prove too much, and so he was stuck with early exit after early exit.

Now, with that being said, this wasn’t a case like we’ve seen in other alternate histories, where a leader was regularly in a strong position before getting dogpiled and was solely done in by a bad diplomatic situation. Ramesses was almost always in the bottom half of the scoreboard exiting the landgrab phase, even before he got attacked, and never in a leading position. Even in the rare games where he managed to hold out and/or get ignored to avoid an early death, he never accomplished much of anything; there were always multiple leaders in front of him, and he never got out to any significant tech lead or took more than a couple of cities from a rival. His one second place finish saw him in third place for much of the game, protected by a Suryavarman who had become his best friend due to shared religion and favorite civic, before Sury killed the second-place Brennus in the lategame to allow Ramesses to backdoor second place. He had an outside chance to take another top two finish in Game 18 as mentioned before, but that was it, and his total lack of kills accurately reflected his failure to ever conquer significant amounts of territory. Essentially, if he hadn’t been constantly attacked in these games, it looks like he’d have performed more like Napoleon than like Peter, suggesting that he’s not a particularly strong AI even by the standards of the peaceful builders.

1718078477977.png


Conclusion
The alternate histories showed that while the real Game 1’s result wasn’t the most likely one, it also wasn’t wildly improbable, and other than Sury’s total dominance, the overall game dynamics were typical for this map. Suryavarman won a decent number of alternate games, Brennus was as likely as anybody to take second place, and Ramesses was by far the most likely to die first. Even a Diplomatic victory had decent odds of occurring, coming up three more times in these games. The most unusual thing about the real game was that it was quite quick and peaceful – every single one of the twenty alternate games lasted longer and saw more wars declared than the real thing. The quick finish was likely due mainly to Suryavarman’s unusual strength, while the low war count was partially due to the quick finish and partially due to Peter and Brennus being more docile than normal. The real loser in this game was Peter, who got consigned to the Wildcard despite being the best leader in the alternate histories, but when he voted himself out of a playoff spot in the real thing, you can’t feel too bad for him! Overall it was quite fun to look at an original game from the perspective of nearly ten years later, enough so that I kept going through these games and have now done nearly four more...
 
Got sidetracked on this for a couple of days, but here are my results for Game 2. I'll post Game 3 as well in just a moment.
Original writeup: https://sullla.com/Civ4/survivor2.html



1718313568785.png

1718313670127.png


The original Game 2 had been the breakout performance of one of AI Survivor’s superstars, Mansa Musa, and the alternate histories proved that this was no accident. Mansa dominated these games, winning a majority, securing a top two spot fully 80% of the time, and beating the next closest competitor by 55 points – that’s eleven first place finishes’ worth! More on this below, but Mansa was given a favorable position in this game, and as quite likely the game’s best pure economic AI, he took full advantage. He would usually take the lead at a fairly early date and then never look back. These weren’t backdoor cultural victories with a small empire; Mansa was winning these games from a dominant position. (In fairness, he was also undoubtedly helped by the pre-Season 4 setup that gave the free Deity starting techs to everybody, thus losing one of his biggest handicaps in the modern competition.)

Aside from Mansa, this turned out to be a remarkably wide-open game. Despite there being only seven games not won by him, five of the other six leaders still won at least once, and after watching these alternate histories, I really believe that any of the seven could have come out on top. Certainly some leaders did better than others and some had only a very small chance at victory, but each leader did have SOME path to a win, which I don’t feel is always the case. It made for an interesting set of games to watch for sure. Sometimes who did better or worse would be determined by initial settling; sometimes by where and when the wars broke out. This plus Mansa’s dominance also made for some tight spreads in scoring among the remaining leaders; just seven points between third place and seventh, and just one between the last two. It was certainly a big contrast to Game One where half the field could be safely written off.

The other general note I have about this game is that it had a rather fast tech pace. The game was usually being won somewhere right around turn 300 – the real game ended up being slower than the average – and an overwhelming majority of games were won via the space race, with just four Cultural and two Domination victories. Oftentimes this could be attributed to Mansa’s economic prowess, but even when he wasn’t the winner, someone else would step in and do just as impressive of a job of finishing the tech tree. The big map and fast tech pace meant this wasn’t the game to try to conquer everybody, and while I did see quite a lot of cultural attempts, they usually couldn’t keep up with the tech either. The real game’s Diplomatic ending was a complete fluke, as well; nobody won via the UN in any of the alternate games.

Now for a closer look at the individual leaders:

Mansa Musa of Mali
Wars declared: 37
Wars declared upon: 50
Survival percentage: 85%
Finishes: 13 firsts, 3 seconds (71 points)
Kills: 20
Overall score: 91 points

Mansa really ruled the roost on this map. He had a great starting position: some of the best land around his capital of any leader, a wide-open peninsula to expand into, no really close neighbors, and only three rivals that he was likely to border, with one fairly weak and the other two likely to attack others first. As a result, he often would peacefully expand to a larger size than anybody else, and rarely faced significant danger from anybody else attacking him, seldom facing two-front wars and often getting to largely tech in peace during the early game. Anybody who’s familiar with Mansa’s AI performance knows that this is a recipe for him to quickly pull far ahead in technology, and that’s what happened in many of these games. Mansa would take the score lead somewhere in the area of Turn 100 and then expand on his lead from there until he won the game. If anything, his performance in the real game was a relatively weak one! Every single one of Mansa’s wins came from first place – he didn’t just take the prize, he was clearly the best-performing AI in these games.

Once Mansa got to around the era of rifles, he was usually unstoppable, and he became a dangerous foe indeed after this point. Sometimes somebody else (usually Mehmed or Justinian) would foolishly try to attack him with inferior technology; they’d learn their lesson the hard way as Mansa almost invariably started storming through their territory. Other times, once he’d assumed a dominant position, he’d decide he didn’t care for a rival (again, usually Mehmed or Justinian), blitz them, and wipe them off the map in the blink of an eye. Mansa often fought with a military tech edge – it was pretty common to see him attacking rifles with infantry and tanks, or sometimes even more lopsided contests. Being a rival of Mansa’s in the late game was quite the frightening prospect, and he defied expectations for a peaceful leader by amassing the most kills of anybody on this map, an even average of one per game, as well as declaring nearly two wars per game. He still was attacked more often than he attacked others, but this was no Gandhi or Hatshepsut. As far as how he won, Mansa almost always went for the spaceship victory. Unlike on some other maps, he rarely pursued Culture here (likely because there were so many other religion-focused AIs founding Holy Cities of their own, which as we’ve learned has a major effect on whether the AI pursues Culture), and usually he’d had his fill of conquering (or simply won by space) before he hit the Domination limit as well. He could have easily won more games by these methods, but simply didn’t choose to most of the time, and in particular going for Culture made it more likely that somebody could slip past him at the end of the game.

So how did Mansa not succeed on this map? Sometimes, he just didn’t get out to as strong of a start, and couldn’t pull ahead in tech like he normally did, eventually falling behind. Sometimes he recovered from slow starts to win anyway, but there were a couple of times where he never did, like in Game 9 where he was out-econ’d by Zara, or his elimination in Game 4 where he was closer to the pack than normal, refused to tech Rifling for ages, and thus got killed by a Mehmed who had no such issues. (Mansa usually teched Rifling plenty early in these games – this wasn’t as much of a problem as we’ve seen for other peaceful leaders in the past – but this was an exception.) There were also a few times where he performed well, but was outdone by a rival who had played an exceptionally strong game – Justinian in Game 14 and Roosevelt in Game 18, although Mansa was in a close second place in both games. Mansa’s other two eliminations in Games 13 and 19 came as a result of him being dogpiled early on, for an actual early exit in the former case and a weakening he couldn’t recover from in the latter. Finally, Game 20 had a wacky ending where Mansa was the runaway, only to pursue a Cultural victory and see Hatty beat him to that condition by about 80 turns. This all added up to enough non-wins to keep these replays interesting, but even when he didn’t come out on top, Mansa was always a significant player. Overall this map was an excellent chance for Mansa to showcase his economic skills, and provided one more lesson as to why he’s such a dangerous player in these games.

Justinian I of Byzantium
Wars declared: 58
Wars declared upon: 34
Survival percentage: 65%
Finishes: 2 firsts, 4 seconds (18 points)
Kills: 18
Overall score: 36 points

Justin finished second in the original Game Two, and backed that up by putting up the clear second-best performance in the alternate histories. While he didn’t quite have the second-best rate of securing a top-two spot, doing that in only 30% of games, he did still put up the second-strongest overall performance and narrowly came in second to Mansa in total kills. Justinian had a much less open starting position than Mansa’s, but it was still one from which he could reliably expand to a solid size, and its position in the corner of the map kept him safe from most aggression from the bigger civs out in the west. Justin also was the only civ on the map to start with Mysticism tech, and he leveraged that to found Buddhism in every single game. The Holy City culture from that would frequently be a significant help in claiming land for him, although there were a few places where that second city would go and he would sometimes put it on the coast where that culture went to waste.

The real strength of Justinian’s position came when it was time to fight and try to conquer more land. Justin benefited from bordering two of the weakest AIs on the map: the perpetually small Roosevelt and the chronically inept fighter Hatshepsut. He fought early with at least one of these two leaders in virtually every game, and he won the war almost every time, whether by fully conquering his rival or simply taking some cities and then signing peace. The gains from these wars tended to keep Justinian relevant – in particular, all of his eliminations came after Turn 200 – and conversely Hatty and Roosevelt’s low survival percentages were largely a result of him frequently weakening or killing them. He could almost never coexist peacefully with them, and only three times did he survive alongside one of the two: a single game where both he and Roosevelt survived in good condition, another where Roosevelt mostly conquered him but let him off the hook with just a few cities left, and a third where Justin conquered most of Hatty before being distracted by a backstab. There just wasn’t enough space out there in the east, and somebody had to go. Justin and Hatty in particular were almost always founders of rival faiths and at each other’s throats for that reason, much like we saw in the real game.

Where Justinian really struggled was in keeping up with the rest of the map. He frequently had one of the game’s worse economies, was rarely able to keep peace with Mansa, and also struggled to beat out Zara and, in the rare games where they prospered, Mehmed and Roosevelt. It wasn’t uncommon for him to emerge from a war with Hatty or Roosevelt with significant territorial gains, only to still be stuck in fourth or fifth place because he’d gotten too bogged down in those wars and not fought efficiently enough, and so be out of the running for a win and sometimes even for second place. Justinian also did die seven times and ended two additional games on death’s door; usually this was at least partially a result of fighting a technologically advanced Mansa late in the game, but there were also three matches where his neighbors did manage to get the upper hand on him and knock him down without Mansa’s help. Justin’s path to victory was to get out in front early and ride an early lead to a win; he did this in Game 11 off an early partitioning of Roosevelt, and then in Game 14 where his economic game was unusually strong and he conquered Egypt and (bizarrely) the Netherlands en route to a narrow win over Mansa. Otherwise, Justinian played fairly well, more consistently than most of the crowd, and had a significant mark on this game, but he clearly didn’t have the best starting land, and that resulted in his successes being only sporadic.

Zara Yacob of Ethopia
Wars declared: 34
Wars declared upon: 32
Survival percentage: 65%
Finishes: 0 firsts, 7 seconds (14 points)
Kills: 9
Overall score: 23 points

Zara has the dubious distinction of being the only leader on this map to win zero of the alternate histories. On the other hand, he still was the leader most likely to advance outside of Mansa, thanks to racking up seven second-place finishes across the alternate histories. All but one of these finishes came behind a Mansa win, suggesting that Zara was an excellent junior partner to the Malian leader. The two rarely fought, although there were a few games where they did come into conflict and each was at least partially responsible for the other’s ouster at least once. Zara had an interesting central position on this map but did a good job of filling it out; his empire would usually be fairly large, but very wide, with a long northern front and cities far to the east and west. This usually equated to a fairly strong score and one of the better economies aside from Mansa. Zara was a leader to be reckoned with, not easy pickings; most of his relatively rare eliminations came late in the game, and he was only First to Die once.

On the other hand, he struggled to actually convert this position into strong performances and couldn’t get out in front on his own. Zara’s path to success was to work with another leader to partition a third party; all seven of his second-place finishes saw him do this at some point, whether attacking Mehmed, Roosevelt, or Hatshepsut. (One outcome that rarely occurred was Zara and Willem fighting. His early attack on the Dutch in the real game was a true oddity.) Unfortunately, he was only getting part of the spoils this way, failing to snowball enough or get out to a big enough size to compete with Mansa over in the west – hence all the second-place finishes. Even though Zara was the only leader to win none of these games, I do think he was a bit unlucky in this regard and certainly had the possibility of coming out in front. He suffered two very close losses in these games: one where he was about to narrowly beat Mansa to space, only to see Mansa go on the warpath and tick over the Domination threshold first; and another where his spaceship was beaten out by just 5 turns by a Hatshepsut Cultural victory. If either of these games had gone a little differently, Zara would have won, and I’m confident that he’d eventually pick up a gold medal if I ran enough of these. But it clearly wasn’t likely.

Perhaps the most telling statistic about Zara’s performance is his low number of kills. He had half of Justinian’s total despite surviving the same number of games, and less than Mehmed despite advancing over triple the number of times. It’s true that he also wasn’t fighting as much to begin with, but overall these numbers (and the dynamics of the games themselves) suggest that Zara just isn’t that effective of a fighter. We’ve seen his reputation diminish over the years since his astounding performances in the rest of Season One, and these results suggest that this is no accident. Zara isn’t great at fighting or economy, likely a below-average leader who was once in the right place at the right times. He’s far from being totally impotent but we’re unlikely to ever see him nearly winning a season again.

Mehmed II of the Ottomans
Wars declared: 54
Wars declared upon: 49
Survival percentage: 40%
Finishes: 1 first, 1 second (7 points)
Kills: 13
Overall score: 20 points

Mehmed was the only true warmonger personality on his map, and his stats reflect that reality. He fought in more wars than anybody else, had by far the highest ratio of kills to survived games, but only finished in the top two twice as he usually couldn’t keep up in tech with every single peaceful leader. Mehmed’s central position in this game gave him enough room to expand and access to easily attack almost any leader; he most often fought with Zara and Roosevelt in the early game while clashing with Mansa more often in the lategame. Mehmed benefited from having a weak Roosevelt next door; he’d often be at least partially responsible for the American leader’s demise, and the games that Roosevelt made it through never ended well for Mehmed – he only survived one, and that with only a few cities left to his name. This helped beef out his territory a bit more, and up until the era of muskets he was usually one of the top leaders. He would often tech Gunpowder at an early date and get good use of his Janissaries while their bonus was relevant.

The big problems for Mehmed came later in the game, though. He simply could not manage his economy properly – from a rather early date he’d usually have fallen laughably behind in GNP, and the results would make themselves felt by the Renaissance as his foes would be able to tech rifles, or more advanced units, well before he could. Mehmed only died once before Turn 200, but he was eliminated an additional eleven times after that point, showing how he’d fallen behind and exited as a late casualty. This fate most often came at the hands of Mansa like we saw in the real game, and there were several occasions where Mehmed attacked Mansa just a bit too late and lost only due to his inferior technology. Mehmed’s central position also proved quite hazardous at times; he was just one war behind Mansa for being attacked the most of any leader, and he was first to die five different times thanks to getting dogpiled or backstabbed.

For a long time I thought that Mehmed’s economic weakness doomed him here, and that it wasn’t possible for him to pick up a win. However, he finally proved me wrong on the second-to-last game and showed that he too had a path to victory. This game saw him get off to a stronger start than normal, snowball off conquests of Roosevelt and Hatshepsut while Justinian and Willem kept Mansa down with two-front wars, then slam his army into Mali at a critical moment to kill Mansa and put himself in the driver’s seat to win. This game notably saw none of the high peaceweight leaders survive; it seems this was a necessary condition for Mehmed to not fall behind in tech. His lone second place finish came in a similar scenario, where it was just the three low peaceweights left and Mehmed had a slight lead in a space race, only to lose to a cultural bid by Willem. These results suggest that Mehmed is strongest when paired with other warmongers, where his economic struggles are less of an issue. Put him with a bunch of economic AIs instead, and he’s simply not effective enough of a militaristic leader to usually come out on top.

Roosevelt of America
Wars declared: 23
Wars declared upon: 40
Survival percentage: 30%
Finishes: 1 first, 4 seconds (13 points)
Kills: 3
Overall score: 16 points

Roosevelt had the most binary outcomes on this map; he was first to die eleven different times, including in a staggering 9-game streak, but on the other hand he took a top two position in all but one of the games where he did survive. This can virtually all be chalked up to his starting position; Roosevelt had a very cramped position with neighbors on all sides, which meant that all of the land around him would invariably fill up quite quickly. He didn’t do himself any favors in this regard either, frequently being one of the slowest AIs to expand thanks to prioritizing wonder builds and whatever else instead. The result was that he could rarely have more than five or six cities, which usually was less than anybody else. This issue was then compounded by two of his neighbors, Justinian and Mehmed, both of whom were almost always bigger and stronger and one of whom would usually come after him fairly early in the game (or, in some cases, come slamming back into him after he’d made an ill-advised war declaration of his own). Roosevelt usually couldn’t hold out against this assault and would eventually succumb, leading to his frequent early eliminations. It also wasn’t uncommon to see him in multi-front wars, and while a less common occurrence, Zara did attack him several times as well.

However, on the rare ocassions where he wasn’t roughed up early, Roosevelt could be surprisingly competitive thanks to the quality of his land. Roosevelt’s starting position lay in a fertile floodplains region, and he’d usually have at least two cities with lots of the wetland to work, plus a third site with double gold resources. As a result, his early economy was generally one of the best, and he would often be keeping up with the leaders in tech before getting knocked down in his wars. When he wasn’t killed early, whether due to his neighbors being more merciful or him getting help from an ally, he was able to leverage this economic strength to eventually become the premiere leader in the east of the map. Note that all four of his second-place finishes came behind Mansa; Roosevelt was growing stronger at the expense of the other leaders, but his small early size and need to split his conquests with partners prevented him from being able to catch up to the Mansanator.

There was one exception, though, the unusual Game 18 where Roosevelt led from an early date. The opening to this game saw Justinian send his Holy City into the northern tundra, Roosevelt plant his second city aggressively in Justin’s direction, and America thus control the entirety of the river valley between the two, when normally Justin’s culture covered at least some of this region. A rougher game than normal from Mansa allowed Roosevelt to maintain a lead until he teched Rifles and conquered portions of both Mehmed and Justin’s land, and he leveraged that to narrowly win a spaceship victory. Thus things needed to line up just right for Roosevelt to come out on top, but it COULD happen! Overall this was an impressive performance from a normally underwhelming leader. Roosevelt had a tough position, but when he was able to get off the ground he proved up to the task of leveraging its benefits and finishing well. Someone who dies first in over half the games scoring this well is quite the feat, but he pulled it off here.

Hatshepsut of Egypt
Wars declared: 18
Wars declared upon: 43
Survival percentage: 30%
Finishes: 2 firsts, 0 seconds (10 points)
Kills: 1
Overall score: 11 points

This game was, to me, an excellent showcase of just why Hatty is the last leader to score any power points in AI Survivor’s history. She had a solid starting position in this game and only one rival whom she commonly clashed with, but still couldn’t accomplish anything of note in 18 out of 20 games. Hatty did a solid job of developing her empire early and was usually one of the better researchers in the early game, but as far as converting this advantage to a strong finish? It almost never happened. Hatty’s path to success in these games was to go unmolested for the whole game, then flip on the culture slider and squeeze out a victory this way. To be fair, this strategy did work in two of the four games where she tried it, and she was just five turns away from beating out Roosevelt for a third win. If nothing else, she was at least clearly ahead of Willem in the culture game. On the other hand… she didn’t even get to try this in 80% of games, and if she couldn’t do this, she couldn’t do anything of note.

The big problem was that Hatty is simply terrible at fighting. She and Justinian were virtually guaranteed to come to blows in this match, as they’d almost always found rival faiths and both took this VERY seriously. There were only two or three games in the entire set where the two could share a religion and become friends instead. And when the two did come into conflict, Justinian won almost every time. Regardless of their relative strength, Hatty simply could not gain ground against him; most of the time her best-case scenario was stalling him out long enough for peace. I even watched one game where Hatty was fielding infantry and tanks against Justin’s rifles, and STILL lost! She just does not know how to fight! To be fair, she usually did a good job of holding out for a while in these conflicts, and was only first to die twice, but she was unable to get the upper hand, and that ultimately would prove fatal. There were only two games where Hatty came out on top; one was Roosevelt’s win, which saw Justinian hamstrung and Hatty beat him to rifles by a significant margin, only to take ages and ages to conquer just two border cities before Roosevelt and Zara plowed in and did the rest. She also once successfully 2v1’d Justinian with Mehmed’s help, and this led to one of Hatty’s two cultural wins. Her other cultural win came in a rare game where she never fought Justin. There were other games where she avoided getting knocked down by Justin – only to get attacked and killed by Mehmed or Zara instead. This wasn’t a super hostile map for Hatty (in particular, I don’t believe she fought her neighbor Roosevelt once across the entire 20 games!), but she did have to do some fighting in order to get through… and she could almost never handle the fighting.

This match paints a clear picture of Hatty as an AI Survivor contender that I think matches what we’ve seen across the rest of the competition. Hatty is good at culturing in situations where she is left entirely alone. However, as we know, this is an extremely rare occurrence; in most cases a leader is going to have to face a hot war at some point or another. And when Hatty has to do so, she’s one of the worst leaders in the game. Hatty lives a very dangerous life in these games, and as of post-Season 7 it’s never worked out for her – but at least she CAN cash in on the occasions in which it does.

Willem van Oranje of Netherlands
Wars declared: 39
Wars declared upon: 15
Survival percentage: 50%
Finishes: 1 first, 1 second (7 points)
Kills: 3
Overall score: 10 points

Willem was by far the biggest disappointment on this map. I’d thought going into these games that his early First to Die result in the real game was a fluke, and in this I was proved correct: Willem only died before Turn 200 once across the alternate histories, and never was the first to die. His position was quite sheltered, he went completely unattacked in many games, and Zara’s early attack in particular was an extremely unlikely result. He had the freedom to build and tech in peace for long periods at a time, and for a Financial leader who’s commonly regarded as one of the best techers in the game, this surely would translate to some excellent performances, right?

Wrong. Willem could never get in front of the economic game on this map; at his peak he’d usually be keeping pace with the weaker leaders in tech, while being soundly outpaced by Mansa or Hatty or even Roosevelt. After some analysis, I’ve concluded that this must be mainly a result of his starting position, which was indeed a quite weak one. There were a lot of low-food tiles, almost no food bonuses of any type, and almost no rivers either – thus leaving his cities stuck at relatively low sizes without any really commerce-rich tiles to work. But even in the couple of games where he was able to break out of this area, he didn’t fare significantly better. Once he took most of Mansa’s core in a 2v1 war, but still couldn’t keep pace and finished second to a strong Justinian in the east. Another time he expanded out to a far larger size than normal and seemed poised to finally have his big game… only to still fall behind in research to a Mansa who had less cities, before getting solo conquered by Mehmed like any other game. (There was also a game where Willem backstabbed Justin while up infantry, trapping him in a two-front war, and still could barely manage to take three cities. His fighting was not good in this game.) Willem survived to the finish in half of these games, and scored points in exactly three of those. Even Hatty had a better conversion rate; for a Financial/Alive leader, these results were simply embarrassing.

Willem did still have a path to victory, though. His one win saw all of the high peaceweight leaders eliminated, paving the way for him to get a Cultural win ahead of Mehmed and Justinian. He came close to doing the same thing in the only other game where those three were the final three, but couldn’t pull it off in time. Willem tried for cultural wins in a number of other games as well, but was never close to success in any of those; in this map against a lot of religious or culturally inclined leaders, he wasn’t able to assemble an effective culture machine, and in particular Hatty would be well ahead of him in any games in which both of them were attempting this win condition. Overall, this was a performance to make me wonder if Willem just got lucky for two seasons, made everybody think he was the bee’s knees, but is truly a mediocre leader who’s simply outclassed much of the time. I need to see alternate histories for more of his less successful games – this may just have been a rotten setup for him – but it may not be entirely coincidence that he was a nobody for the first three seasons of AI Survivor. [Edit from the actual posting of this writeup: I've looked over Willem's other alternate histories and honestly still am not sure what to think of him! This game was clearly at the lower end of his performance due to a bad setup, but I still don't have a good read on his true threat level.]

1718313634731.png


Conclusion

In some ways, and especially in the broad strokes, we saw a quite typical outcome for the real Game 2. Flip Hatty and Roosevelt and the leaders even finished the alternate histories in the same order that they finished in the real game! Mansa was the normal winner for this game, Zara and Mehmed and Justinian would all often be major players with Mehmed dying late, and the other three were usually also-rans. The game length and number of wars were also fairly typical. On the other hand, other specific aspects were fairly unusual: Willem as the first to die and the Diplomatic victory were both total fluke results, and Zara was significantly more likely to come out in second place than Justinian. So while this was far from a crazy game, it certainly did have its oddities.

This game was also the one that firmly hooked me and got me addicted to the alternate histories. They're a significant project, but I have a lot of fun watching the different scenarios unfold and tracking the bigger competition as teams rack up points from game to game. I'm now in mid-Game 8 in these, and don't plan to permanently stop any time soon!
 
And now Game 3, the last one that I've publicly posted so far. I'm planning to release more whenever there's not as much going on - maybe this week, if Sullla doesn't have anything new posted tomorrow. We'll see!

Original writeup: https://sullla.com/Civ4/survivor3.html

1718313765899.png

1718313773340.png

The alternate histories for Game 3 showed a remarkably open game, one where five of the six leaders on the map had very similar chances to succeed or fail. Isabella was the clear strongest leader on the map… but she still fell more than half of the time, and when she fell, it was a tossup as to which other leaders would take her place at the top, with only Joao getting left behind. If you look at the stats for this series of alternate histories, you’ll see that ALL SIX leaders had a survival rate in the range of 40 to 55 percent – everybody was dying half the time, and nobody was a consistent performer here. Similarly, it was a 25 to 45 percent chance of advancing for five of the six leaders, all having a decent but by no means guaranteed shot at the playoffs, and all six leaders suffered the First to Die fate at least once. This was a fairly warlike group of AIs, making it unlikely for anybody to get left off the hook all game, and indeed having just two AIs survive was more than three times as likely as having four make it out! The real question was where the wars would break out and who would get on the wrong side of them in any given game.

The conflict-laden nature of this map also resulted in a lot of Domination victories. In his original writeup, Sullla had commented that this could well be the first such win in the tournament, and while that didn’t happen in the real game, here it was confirmed as a good prediction as it happened over half the time. Many of those Domination finishes came rather early as well, leading to fast games. On the other hand, when nobody got over that hump and the game went on to a more peaceful conclusion, it tended to drag on a bit since nobody had great economic traits or focus on this map. Boudica and Alexander were often decently competitive in research when they had competitive-sized empires, which should speak to the fairly low ceiling for economies in this match.

Now for a closer look at each individual leader:

Isabella of Spain
Wars declared: 16
Wars declared upon: 82
Survival percentage: 40%
Finishes: 8 firsts, 0 seconds (40 points)
Kills: 16
Overall score: 56 points

Isabella was the clear best performer on this map and the “default” winner – she would come out on top unless the other leaders did something to stop her. This can be attributed to her excellent starting terrain – there was some choice land and lots of resources around her capital, allowing her to grow her cities to large sizes relatively early on and power a really strong research machine. Founding one of the opening religions in every game helped further establish control over this desirable region, and she usually could also grab a good share of land around her start, although even in the games where she didn’t she would still be ahead. Isabella was the score leader by Turn 75 in nearly every game, usually by a large margin, once well-established she could hold off any single foe with ease and usually conquer them, and if she wasn’t hamstrung early enough she’d become the tech leader and ride to victory from there. Izzy got into fights early and often, but once she was strong enough that resulted in her streamrolling her opposition. She never went to a peaceful victory, winning all eight of her games by Domination and narrowly racking up the most kills on this map. Her strength was the biggest reason that many of the Domination finishes came so fast, as she would win at early dates including Turns 267, 248, 260 (twice), and an amazingly fast Turn 242 victory in Game 11! By way of comparison, the earliest recorded victory in AI Survivor history was in Turn 252, although in fairness that result WAS posted after the AIs were reverted to their default starting techs – the free Deity techs did help Izzy out a bit in this regard.

But it wasn’t all glory for Izzy on this map. Lush as it was, her start was also centrally located in an aggressive field of leaders, and that resulted in her being invaded a TON – over four times per game on average! Sometimes she could make it through these attacks and dominate anyway (two of her wins came despite being attacked six times in each game), but other times the wars plunged her into a strategic pit that she couldn’t climb out of. There were two ways for Izzy to be brought down on this map. The main way was by her getting caught in ill-timed dogpiles or two-front wars, too many enemies attacking her at once and overwhelming her that way. This could happen early in a game, resulting in her being the first to die, but it could also come late if she hadn’t gotten the snowball running. As one example of how this result could play out, Game 13 saw her get backstabbed by Alex, start killing him, get backstabbed by Boudica, start killing her too, only to then get backstabbed by Qin as well and only then fall. Isabella was eliminated after suffering five invasions in one game, after six invasions in three more, and after SEVEN invasions in two particularly brutal scenarios. I would need to go back and look at the indiviual games a bit more to determine how these scenarios came to pass, but I suspect the single greatest factor was how well she spread her religion. I also noticed in the late games that what happened with Boudica and Louis seemed to be a good indicator of how Izzy would fare – they often fought which was good for her, but on the occassions where they didn’t, Izzy was their only other nearby target, and so would be dogpiled sooner rather than later.

The other, more rare way for Izzy to fail was to be hamstrung early on by Alexander. Alex attacked Isabella early on in the vast majority of games, something like 80-85%, despite almost always sharing her religion. Many times this attack was fruitless and Izzy could fend him off easily, but there were a few occassions where he broke through early on and took a core city to permanently set Izzy back and weaken her for future conquest – or, in one game, went on to conquer her entire empire on his own! This wasn’t a common result but it did happen enough times to not be a total fluke; Isabella needed a little bit of time to properly fortify her cities, and Alex had the ability to come calling before she did so. On the other hand, there were a small number of on this map where Izzy got in the first strike against Alex, and those resulted in her conquering him and going on to win the game.

The result of all this was an extraordinarily binary set of outcomes for Isabella. In EVERY SINGLE GAME, including the real Game 3, she either was completely eliminated or won a Domination victory. There was no in between. She also managed the impressive feat of being both the most likely leader to win on this map AND the most likely leader to die first! I went back and checked and this has never happened in any other alternate history; it’s truly a unique feat for her. I don’t think that we can draw any real conclusions about Izzy’s strengths or weaknesses from this performance – the dynamics seemed to very much be driven by the map, and her starting position. Still, on the whole, despite the mediocre score for a map’s winner, I’d call this an impressive outing. Other than the uncommon cases where Alex succeeded with a very early strike, the only way for Isabella to fall was for multiple other leaders to work together to take her down – she just happened to be in a position where it was very easy for this to happen.

Qin Shi Huang of China
Wars declared: 40
Wars declared upon: 43
Survival percentage: 55%
Finishes: 4 firsts, 4 seconds (28 points)
Kills: 11
Overall score: 39 points

The middle four leaders all finished very close together, and in particular I think the non-Alexander ones can effectively be put at the same level. Qin narrowly claimed the second place position though, thanks to randomly winning a Diplo victory from a trailing position in Game 16 to steal a win from Louis. He did barely get the highest survival rate on the map as well, though the fact that even he died in almost half of all games shows how chaotic this map was. Anyway, Qin put forth a decent performance here thanks to having the best starting position in the north of the map, not boxed in and with a nearby floodplains region (including a gold resource) that he always settled with his first two cities. These factors combined to naturally make him the strongest of the three northern leaders in game after game, and thanks to that strength and his Protective trait, he was virtually guaranteed to make it through the early game – his only two early exits across this set of games came from being dogpiled via the AP.

That was the good side of Qin’s performance. The bad side is that he didn’t accomplish a whole lot with that favorable position. Qin fought a fair amount, but few of those conflicts were decisive, and he usually needed help from another leader to actually take down a foe. Given that he had the strongest northern position and a fertile start, you might expect him to be quite strong in the games where Isabella was taken down, but that wasn’t particularly common; he only assumed a dominant lead twice in twelve such games, and jockeyed with another leader for dominance an additional three times. He couldn’t even get a lot done against Joao; the two were natural rivals, fighting in most games, Qin was almost always stronger, and you would expect him to frequently roll over Portugal to strengthen his own position. Instead, the two almost always stalemated, Qin unable to break through Joao’s defenses without help. He was only able to quickly conquer Joao on his own three times across the twenty games – and he placed in all three of those games, so he clearly would have had a better record if he had fought better against J-man. Instead he mostly just stalled.

Qin’s path to a normal victory in these games seemed to be to watch Boudica or Louis get weakened or killed by early wars then work with the survivor and Alex to bring Isabella down, after which he’d be in a leading position. This wasn’t a common occurrence, but it did happen several times and he wasn’t that much less likely to win than any of the other non-Isabella leaders. Overall, Qin reinforced himself here as a middle-of-the-road AI, competent enough to handle a solid position without messing it up too badly, but not accomplishing nearly as much with it as some leaders might. He was viable on this map but not particularly better than anybody else except Joao.

Louis XIV of France
Wars declared: 37
Wars declared upon: 37
Survival percentage: 40%
Finishes: 4 firsts, 3 seconds (26 points)
Kills: 10
Overall score: 36 points

Louis has established himself as a decent if not amazing leader in later seasons of AI Survivor and had a fairly spacious starting position in this game, so I was expecting much better results for him in the alternate histories than in the real game. He ultimately proved something of a disappointment, though, with only four wins to his name and a track record not significantly better than the other middling leaders on this map. He also became the leader I was most annoyed with in these games, as the victories he did get tended to come rather late and make those games rather a slog to finish, especially compared to the fast finishes Izzy was recording.

The most common dynamic characterizing Louis’s game was his consistent warring with Boudica, as in game after game the two fought for sole possession of the eastern wing of the continent. The two leaders were a rather even match, however, and so in most games this conflict dragged on for a long time, with one leader usually emerging victorious eventually, but not without a lot of stagnation. While Louis would win a good amount of the time, it was usually too late by the time he had, and Izzy or Qin or Alex had taken control of the game in the meantime. Even the one and only time that Louis was able to kill Boudy quickly on his own saw him eventually run over by Izzy anyway. This strategy only worked out for him three times in total, resulting in two of his wins and one of his runner-up finishes: each of those times, Alex and/or Qin had slowly taken out Isabella while Louis was busy with Boudica, and while Louis was able to execute effective backstabs to take the win twice, the third time he was hopelessly behind Qin in second place. Fighting the hardy Boudica simply took too much time and effort to be a reliable strategy.

Those games accounted for about 75% of the total; Louis’s much rarer but much more effective strategy was to instead ally with Boudy, upon which the two would work together against their only other neighbor: Isabella. This only happened in four games, but each of those games saw Izzy as the first to die, and Louis’s strong culture ensured that he got the best share of the spoils and was dominant over Boudica in particular. Louis placed in all four of those games, winning two outright and losing to a random Diplomatic finish in a third; I think the fact that Boudica was the weakest economic leader here made it much more profitable for Louis to work with her than against her. This, then, is where the Sun King really saw success, but it was a rare occurrence.

Louis’s only game that didn’t fall into either of these categories was an ignominious one where Boudica was distracted by a random war with Alex, leaving Isabella free to attack and solo kill Louis for a First to Die finish. Notably, Louis was the only leader who never finished second to Izzy; indeed, he never survived in the same game as her, as they’d inevitably come into conflict somehow and he couldn’t prosper as long as she was there in the center of the map. I’m not sure what exactly made these two such bitter enemies when Izzy could still sometimes coexist with, say, Alex, but there you have it. Ultimately Louis was First to Die four times across the alternate histories, and as one of the more likely leaders to suffer that fate he proved that doing the same in the actual game was no fluke. While the specific series of events seen in that game – Izzy and Boudy working together to partition Louis early on – was fairly rare, only happening twice more in the alternate histories, it still feels like a justified result in light of these games. Louis could sometimes get stuff done, but tended to get eliminated outright when he didn’t (only one Wildcard performance in 20 games, and he barely escaped dying in that game) and, on the whole, underperformed in this scenario.

Boudica of the Celts
Wars declared: 51
Wars declared upon: 31
Survival percentage: 50%
Finishes: 1 first, 8 seconds (21 points)
Kills: 15
Overall score: 36 points

Boudica was the winner in the real Game 3, but while the alternate histories showed her to be the most likely leader to advance, they also proved that she was extremely unlikely to actually win. She only took one more victory across the set, and that was a nail-biter against Qin; her total dominance in the real game was a true outlier result. Instead, Boudica was able to utilize her sheltered yet fairly spacious position to amass a large number of second place finishes. She was consistently a formidable fighter in these games, but had a major weakness: economy! Even Alexander was able to out-scale her economically in games where both prospered, and regardless of who else was around later in the games, Boudy was never able to properly get the snowball rolling. Somebody else would always have just as strong of a territory base, and be able to leverage it more effectively to take the actual win. But between her fighting and the fact that, at the edge of the map, she was less likely to be dogpiled, she would often still be doing better than the rest of the field and move on nevertheless. (Of course, half the time, she still either was killed early or would be blown away late by a superior foe. As said before, nobody was truly safe on this map.)

As was discussed before, Boudica usually clashed with Louis early in the game. While she tended to do better when she won these conflicts and worse when they stalemated, as one would expect, these weren’t hard-and-fast rules and there were exceptions on both sides. Ultimately these conflicts could run the whole gauntlet, from Boudica getting very slowly ground down and eliminated to her quickly winning the 1v1 and coasting to second place from there. She did achieve three quick wins against Louis compared to just one that he got against her, suggesting that she was a bit stronger in this conflict, but less likely to capitalize for a game win, and certainly far from unbeatable. The scenarios where the two didn’t fight usually saw Boudica as a decent but clearly trailing leader; Louis was clearly the stronger partner when the two conquered Spain together, thanks in no small part to Boudica’s much weaker culture, and while she did get two of her second-place finishes from these scenarios, this wasn’t a particularly great route for her.

So what was different in the real game? The most interesting alternate history in that regard was her only win, in Game 19, and it followed a similar pattern: both games saw her work with Isabella to partition Louis, then with Qin and Alex to conquer a religiously isolated Izzy. However, in Game 19, Qin had quickly conquered Portugal and then was an even partner with Boudica in the conquest of Spain, leading to a close finish, while in the real game, Joao had gone undisturbed other than Alex pursuing a futile cross-China conflict, and Izzy had been in the process of conquering Qin when the other two intervened – thus allowing Boudica to take almost all the spoils and exit the conflict as a runaway. Thus it was possible for her to win, but it required everything to line up just right – she needed help to take down her two most dangerous rivals, and for the other two to stagnate and not enter winning positions themselves! Everything lined up more or less perfectly in the real game but it clearly was an outlier result.

Boudica thus exited Season 1 with a good reputation that wasn’t quite backed up by her real abilities. The alternate histories here show her weakness as an economic leader – even here in a pretty warlike field, with the bonus of starting techs, she almost always fell behind in tech – but at the same time they show her tenacity as a fighter. She was able to eke out a solid position for herself in almost half of the games and make something of them, even if that wasn’t a win, and she could do it in a large variety of situations – she finished second to each of the four winning leaders twice! She’s more viable than some of the more insane warmongers just because she won’t self-destruct in the same way. Still, as shown here, she’s a good few steps below the game’s really strong leaders due to her economic troubles. She’s unlikely to often put up as impressive of displays as she did in her original outing.

Alexander of Greece
Wars declared: 68
Wars declared upon: 26
Survival percentage: 45%
Finishes: 3 firsts, 2 seconds (19 points)
Kills: 12
Overall score: 31 points

Alex put up possibly the most impressive display on this map, given the weakness of his starting position. Alex was crammed up in a northern corner of the map without a ton of room to expand in to begin with, then in every single game saw Izzy plant her second city in the most fertile area between their two capitals, found Buddhism, and culturally dominate most of this area. That left him always fairly weak, with only room for five or so cities, not much to work with, and before running these I wrote him off as a likely failure on this map. But I was wrong – Alex didn’t exactly dominate, but he did advance a quarter of the time and win three separate games. Not bad for that starting position!

Alex’s location was one that needed to be fought out of, and fight he did. His most common target was the neighboring Izzy, oftentimes while they shared a religion, but he could also fight early wars against Qin, Joao, and even Boudica in one wacky game. Sometimes these wars would go nowhere, and he would exist for a long time as an also-ran before either finishing outside the top two or getting eliminated late in the game. Sometimes they’d work out, though, and he’d get to a strong enough position to really compete. There was no true road to strength for Alex, his five placements coming from a variety of sources, but unsurprisingly the most common route to success was for him to participate in a dogpile of Izzy which gave him enough spoils for a position of strength. He also survived one game in second place behind Isabella as she or the AP killed all the other strong leaders, and in one particularly impressive performance managed to conquer her at the beginning of the game with a particularly successful early attack, and ride to victory from there. As mentioned in Isabella’s section, while Alex’s early attacks were mostly fruitless, there were a few times when his early strikes were decisive in stopping Isabella and letting somebody else win (though only once did he himself win from this situation).

Perhaps the most impressive fact about Alex’s performance was the fact that, despite his weakness and aggressiveness, he survived to the end in nearly half of the games, and only died before turn 200 three times! Notably, all three of those results saw Izzy go on to win the game without the threat of Greece on her northern frontier. Alex actually won more game than he died first in; both of his first to die results resulted from the rare case of Isabella recording the first strike and crippling him. Overall, Alex’s performance on this map showed how dangerous he can be. While starting techs were undoubtedly a help and he still failed more often than he succeeded, he definitively outperformed his weak position and was still a legitimate threat to win the game. While Alex is always in danger of flaming out early without accomplishing anything, he is at the same time not a leader to be underestimated, a fact which I think this result corroborated.

Joao II of Portugal
Wars declared: 35
Wars declared upon: 28
Survival percentage: 45%
Finishes: 0 firsts, 3 seconds (6 points)
Kills: 1
Overall score: 7

There was one leader on this map who was well and truly hopeless, and that was Joao. The Portuguese leader was dealt a truly terrible starting position, with little land available, rather low quality on the land that he could get, and his only neighbor a Protective leader who could not easily be fought through. He was also rather unwilling to expand directly towards Qin thanks to a pair of peaks right between them, instead sending his settlers to the coast in game after game. This lead to Joao trailing in game after game, and unlike the others, there was never any sort of opening for him to get out in front. He could help dogpile Qin or Alex a couple of times, leaving him still definitely weaker than his partner, and that was about it. Even on the rare occassions where his strength was at a similar level to Qin’s or even higher, he never managed to take more than one or two cities on his own. He did a good job at holding out against Qin’s repeated assaults and wasn’t often killed quickly, but that’s about the best that can be said for him. Joao is not the most dyanmic leader to begin with, and this starting position did him no favors, leaving him to just spar ineffectively in game after game, with Qin, Alex and even Boudica.

It is interesting to note that Joao’s fate was very closely tied to Isabella’s. They had fairly close peaceweight and were far away from each other, resulting in the Spanish queen holding little interest in conquering him. As a result, Joao survived to the end in all eight games that Izzy won, and only once when she didn’t. His three second place finishes came from games where Izzy managed to knock down everybody else before winning Domination, allowing Joao to backdoor the position while doing little himself to earn it. There simply wasn’t much to be done on this map; Season 1 threw out some true dud starting positions, and Joao was one of the unfortunate victims, forced to play an irrelevant series of games while hoping that his best ally could come through.

1718313792967.png


Conclusion
While the alternate histories showed that this was a wide-open game with a lot of feasible outcomes, they nevertheless also showed that the particular outcome that we got was a bit more out there. Boudica’s dominant romp was never repeated, and her win turned out to be extremely unlikely, with four much more likely candidates for that spot. Alex was similarly one of the least likely leaders to take second place, and actually more likely to win! The spaceship victory was also rather uncommon in this warlike field. About the most normal result in the real game was Louis dying first; of course, this also meant we didn’t see his strength on this map (nor did we that of Isabella or Qin), but with such a wide-open setup, it was impossible to see everything in the real game. This turned out to be a very interesting series to run thanks to the wide-open nature, with Isabella’s boom-or-bust performance as the most exciting part. I think the next game might be a bit more straightforward!
 
Add me to the list of interested people.


I would enjoy reading them as a fan of the series, but I don't find the early seasons to hold as much strategic value due to the change in starting techs. I would place more value on complete Alternate Histories for Season Four's Wildcard, Playoffs, and Championship.

I did AH for playoffs 1 (interesting), playoffs 2 (boring), and partially for playoffs 3 (boring too).
Here are the files.
 

Attachments

An update on this, finally, as the current season of AI Survivor draws to a close.

I currently have seven additional writeups complete and ready to post. I think that once Sullla has posted his conclusions page to finish off Season 8, I shall start posting them again at the pace of one writeup per week. More may be completed during that time as well, we'll see. I'm hoping to be able to also run a couple more for Season 8, which would come first.
 
I did AH for playoffs 1 (interesting), playoffs 2 (boring), and partially for playoffs 3 (boring too).
Here are the files.
I don't know how I missed this, but sorry for the delayed reply! Thank you for sharing. While it may seem boring on your end and may very well prove to be of little interest to others, just those results would reshuffle the top tier of the Alternate Histories rankings. For instance, it would take the shine off Willem a bit, push Huayna Capac to the top at least in placing first, and also push Gilgamesh and Darius upwards even with the latter's appalling opener this season. Just a few conclusions that show why I value the missing playoff round from S4 despite my tardy reply :)
 
those results would reshuffle the top tier of the Alternate Histories rankings.
Out of curiosity, how do you compute those AH rankings?
The most obvious way of doing it would be to simply add up scores. Two issues with that, though:
  1. - It greatly inflates the scores of leaders who lucked out with a super favourable start.
  2. - It allows leaders to score in the play-offs when the AH show that they shouldn't have been there in the first place (and conversely robs the AIs who should have played the playoffs of that scoring opportunity)
For the second issue, I remember that my original plan had been to also run "alternate playoffs (and Championship)" AH with the leaders who *should* have been there (a bit like what I did for the Jumbled Rumble, without the Jumbled part ;) ).
But there was the problem of determining who should be in the wildcard game...

As for the first issue... instead of adding up the individual scores, just go for the standard score (5/2/1*) applied to the overall result?
But then how to attribute that score?
Winner: 5 points to the best overall performer or 5 points to the AI who won the most games (and what if it's a draw)?
Runner-up: 2 points to the second best overall performer? Or to the AI who won the second most games? Or to the AI who got the most second places?
Kill points: Even trickier... Just one point to the AI with the most kills? But then, the average kills per game must be close to 3 so that would be underscoring that category compared to the "live" Power Ratings...
 
It's pretty basic. I average the percentages for the different categories and separate them into one table with all of the games and another with just those from the Opening Rounds. This allows ranking for all the percentage categories and also lets me continue to ignore the kills as a score component :mischief:

For the first issue, using percentages creates the illusion of a standard game that sidesteps the exact score accumulated over however many actual games. The gap then is between leaders like Stalin who only have Opening Round games on record and those like Huayna Capac or Gilgamesh who have at least that many games in the playoffs. In terms of favorable starts, I get the sense that some AI are more capable of exploiting that factor and others have yet to show they can be favored to advance on a map.

As for the second point you raise, you can see how averaging the percentages partially mitigates the issue. If an AI does badly in the playoffs, it just hurts their overall performance. That said, the asymmetry in chances to score remains. Running a whole bunch of games would yield more insights, which is generally how you run your leagues anyways.

Overall, it's a way to make sense of the Alternate Histories data. I would be curious your thoughts, especially when I can update with all of this season's games.
 
It's a larger version of the following:
S4S5S6S7%1st1st Rank
Charlemagne5%5%5%0%3.8%44
For each season, I take Charlemagne's first place percentage from the Alternate Histories, and then I take the average and rank it against the same calculation for all the other leaders. Charlemagne has a terrible track record in the Alternate Histories, which is partly why I always have an eye on him in your leagues.

I do the same for second place, first-to-die, and survival as well (the last two not traditionally scoring categories). The discrepancy arises when including the playoffs, as leaders vary in the number of games they have played, but I just take the average for whatever games they were in. I do not have an overall ranking system, but the components are telling.

I also like a category I call relevance, which is advancing (placing first or second) divided by survival. Charlemagne is middle of the pack for the Opening Rounds and ranked 46 in the composite. Just a reminder, these are the Alternate Histories for a current Pool One leader...

Hopefully that makes more sense!
 
Ah ok, I see now. :)
Since the % are simply "nb wins x 5", you're essentially averaging the number of wins.

Using a sum for the current set of AH would amplify the outlier effect, so using the average is indeed the only thing that would make sense here: we can consider live game playoffs composition to be more or less random (outlier effect), and thus those games can be considered as another set of "opening round games".
On the other hand, using the average for "alternate AH" (playoffs replayed with the "correct" leaders") would run into the same kind of problem I faced with my scoring for the Rumbled Jumble: since competition is fiercer there, for some leaders, participation in the playoffs would end up degrading their stats, potentially pulling them below leaders they'd beaten in the opening round!

However, it does favour "lucky" leaders. The win% for a season is the total win score, without runner-up points or kills points. So Caesar gets an 85 from his utterly unbalanced start in S4... Charlie's gonna need to win quite a few to make up for that! :lol:
That said, it is true that it usually takes a stronger leader to fully exploit a good start's potential. Those OP contexts where almost any leader would thrive were rare. I think. :)

Anyway, this discussion has led me to start another project, in parallel of the League: I don't want to use aiplay for the League games, but I could use it for other stuff when not running League games.

I'll share when it's a bit more advanced, but it'll be interesting to compare the results. :)
 
It's funny but I ignore the live results in this format :lol: Theoretically I could tack them onto the Alternate Histories as one more game. The only oddball round with more than 20 games is Season Seven Game Five where we ended up with two sets thanks to Slashin'.

Some leaders do better, others worse, in the playoffs, it's true. I don't understand the variables there, but I suspect it generally has to do with the success of low peace weight leaders. I'm not even sure the playoffs are consistently more competitive. If Huayna Capac has five games on top of the Opening Rounds and stays near the top, while Alexander and Mehmed deteriorate, it might be more of a clarification of competitive matchups rather than just a degradation of stats.

There are a lot of interesting angles to explore. I like the way you group together the yet-to-win AI in your League, and it could apply to this ranking to see distinctions at the top or bottom. It just represents a fair chunk of time and energy to run Alternate Histories to get started!

Looking forward to your next project! (And thanks Eauxps for your patience while we sort out this side business)
 
What I meant was the Playoffs games that AH are run for stem from the live games results.
If Charlie (for instance) won a live opening game, we'd get him in the playoffs, even though AH for that game later would show that HC should have made it.

So that's why I say that AH for the existing playoffs could be put on the same "level" as the opening round games AH: in many instances, the leaders who made it there weren't the "best" leaders in their game.
Now, if you were to replay the AH with those "best" leaders instead of those who made it there in the live game, then competition would be tougher (hopefully: because if it isn't, it means the whole exercise is pointless ;) ).
And an above-average AI, but not top tier, would tend to get crushed there and drop to below-average.

So with the current set of AH, an average is fine (and a sum wouldn't be).
It's with that theoretical set of AH featuring the "best" leaders that an average would be an issue (and a sum would work better).
Not sure I'm clear. :lol:

Btw, in the files I've posted here, you also have another set of AH for S4 games 5-8 and the wilcard, if you want more data. :)
 
Almost time to start these up again! But not quite. I think I'll give it another week to let the hubbub from our exciting season finale simmer down. Expect to see the first teaser next Saturday, with full results the Monday after and the cycle then repeating until I run out of reports.

Now does seem like a good time, though, to mention a couple of caveats to these that came to my attention after I started running them.

First: espionage. I know Sullla used a different means of gaining espionage points against the AIs in this season than he did later on (from some writeups it seems that he spawned in Great Spies on the fly as needed to boost his total). I, on the other hand, did not think about adding espionage points at all until several games into this. Once that aspect was pointed out to me, I made the conscious decision to remain consistent for this season and continue to not gain any espionage points (there's no need, anyway, as debug mode lets me view all relevant information at will). I don't know how much this has affected the replays; for my part, I haven't seen any discernable difference in the game or AI dynamics.

Second, barbarians. I completely missed this for ages, but the barbs in my replays have not been starting with Archery tech, leading to a lot more warriors and easily captured cities (they do pick the tech up on their own eventually, but not right away). How much has this affected the overall dynamics? It's hard to say. But given that the barbs are a more random factor, and the biggest influence I've noticed is when a leader is stronger than normal thanks to an opportunebarb city capture, I suspect that the biggest impact has been to add more random strong results from otherwise weak leaders. It thus seems likely that the main dynamics of the games haven't been that strongly affected. Anyway, I will fix this in all future replays moving forward, but for most of Season 1 it will be in effect.

Third, starting rolls. Most alternate histories (including almost if not all of the official ones on Sullla's website) are played all from the same save file - which *probably* means that all start-game rolls such as peaceweight are the same for each replay - and different from those of the real game, since some work with the WorldBuilder is required before running the AHs. This will also apply to most of my Season 1 AHs. However, for my latest set, I instead started from the WorldBuilder file for each replay, which worked well and I *think* avoids this issue.

Finally, I should note that I have been using an AI observer civ (ol' SITTING_OUT_BULL) in all of these replays to avoid Apostolic Palace victories. Other than one case I noted in the writeup for Game 1, I don't believe this has affected the competition.

See you all in another week - I'm looking forward to finally sharing these!
 
Finally, I should note that I have been using an AI observer civ (ol' SITTING_OUT_BULL) in all of these replays to avoid Apostolic Palace victories. Other than one case I noted in the writeup for Game 1, I don't believe this has affected the competition.
The one thing to watch out for is that each additional civ lowers the Domination land threshold by 2% I believe (it's 64% for the standard 6-player games, 62% for the 7-player games).
Most of the time it shouldn't matter, but in my AH I've had Cathy win by spaceship not once, but twice when she was being conquered and her opponent was at 63.61% land area (just unbelievable). With an extra civ, she would have lost (and honestly, deservedly so ;) ).
 
Back
Top Bottom