The empirical game-balancing thread

[Viatos]

Don't have the Turn 1 for this, sorry.

This is a Jotnar game on WorldofErebus with Blessing of Amathaon, Living World, and the option that scales difficulty with score. Also, flavourstart et cetera. Also, I turn off Victory Conditions except for Conquest.

You will note right away that it has not been a well-played Jotnar game; I have only three cities, a handful of military units, and my 250+ XP Wild Troll has never been upgraded because I never spent the five turns it would take me to learn Tracking. And yet there is pretty much no hope for the AI to do anything; I've scouted to the Malakim and Ljosalfar. In fact my tech tree is relatively random based on what I felt like doing next rather then what would really be most helpful.

If I wanted to, I could pick up Tracking, move my singular Brass Drake, take them both out simultaneously with Egrass on the Malakim, stealthing through their deserts and blasting their defenses to nothing, and the Brass Drake acting as softener for the King of Rock and Troll (should rename the unit) who can stealth through Ljos forest.

The Austrin were quite close north of me, but luckily I had found Orthus on the western tip of the continent and monitored his movements - when I saw him heading into Austrin land, I quickly signed Open Borders and sent the King to tag along, hoping to steal his axe.

Little did I know they had a fort right between their two cities, which Orthus ate and claimed before beginning the most meandering back-and-forth ineffectual assault you can imagine. Anyway, I stole the fort (Orthus ignored this), promoted the Herredcarl to Stormkin, proceeded to Troll Orthus and devour their Archerless top city. Their capital had archers, but with the help of lightning bombardment I was attacking with 80% odds, so that didn't last. A stack of five Psions with like 7 strength each interrupted, but I gave them the lightning and Trolltanked them, and after that the end was swift.

I then settled in for a long period of growth and development, unmolested except by random Air Elementals (so OP, thankfully Jotnar are made of defense) and Gnoslings, feeding my Troll a steady stream of barbarians and animals (many of which, particularly Boar, are so wonderful for the Jotnar in their cities). Egrass explored wherever the Troll wasn't, picking up a respectable amount of experience as he did so. At some point I got a Troll Goblincatcher from an event; I gave him Seakin and then he died like six turns later.

The key victory conditions here were extremely strong invisible units and being careful with my Troll in the endless deserts WorldofErebus insists on shoving everywhere. Popping lairs didn't actually do that much for me, but killing their guardian minotaurs with forests to avoid retaliation was great. I am also thankful Orcus didn't attack me, which would have been distressing and likely cost me as many as three units.

 

[Koravis]

Just sayin', guys.

As for "incomplete games" not being used to balance... that really isn't the point. As several have needlessly pointed out, there is a clear tipping point in games beyond which, continuing to play is essentially beating a dead horse. The devs are trying to balance the play experience *up to that point* -- I mean, it's not as if they're trying to somehow make the AI capable of coming back after the player has decimated the world population with military might. That is simply unrealistic.

I thought the idea was pretty simple -- post two data points and draw the line connecting them. That is, briefly describe important play decisions that can't be gleaned from world builder or the end-game animations and graphs. There's absolutely no reason why hundreds of saves would be needed before the devs could get any use out of this... and if that were truly the case, by the way, then why have feedback/balance threads at all? Answer: the devs aren't sociologists, they aren't looking for a statistically viable sampling of feedback comments.

I guess people think boiling a 400-turn game down to three paragraphs of complaints based on a limited (sometimes totally absent) understanding of the mechanics they're complaining about is more useful than this. So be it.

heh, I'm sorry I probably didn't get what I wanted to say out really well, I didn't mean to bash the idea or the thread at all, just discuss the proposed method and the additional information needed besides the savegames.(And my view that the actual save games are either superflous, or too little) I think a thread to post a summary and details of one's game for balance purposes and the like is a great idea, and I know it can be very useful, and you've got my kudos for making it heh

 

[Tsathoggua]

heh, I'm sorry I probably didn't get what I wanted to say out really well, I didn't mean to bash the idea or the thread at all, just discuss the proposed method and the additional information needed besides the savegames.(And my view that the actual save games are either superflous, or too little) I think a thread to post a summary and details of one's game for balance purposes and the like is a great idea, and I know it can be very useful, and you've got my kudos for making it heh

I do agree that, as the game gets closer to what the devs have in mind, posting saves will be less useful.

But, right now, I don't think anything better demonstrates the impact of, say, imbalanced lair results or the AI's inability to deal with the unhealth redesign, than to open up an end-game save, resign, and watch the little vid that shows the expansion (or lack thereof) of all of the civs in one view pane. The graphs drive it home even further. I don't expect players to be downloading and looking at these saves, but if you did download mine you'd see what prompted me to begin this thread in the first place. There's actually a lot of good information presented in those end-game screens if you're looking to assess the overall balance of the game, and it isn't really that difficult to parse.

 

[Castar]

Soooo. Played with the Elohim some. Mind, the final save is not where I thought I won, it's where I gave up. It was kind of a slow start, but I managed to keep up with the AI and surpass them all, as usual on Noble. I basically went for Monks + Altar of Luonnotar (the victory condition was turned off) + (Great) Prophets. Everything all fine and dandy, not much happens, but I notice the Jotnar staying well ahead of me in score.

I found out why (but not how) when I finally went to war against D'Tesh. I had to cross Jotnar territory (with whom I was friendly), and I saw this. Some of them are in stacks of three or more.

It was a fun, albeit slow, builder game up until then, and I would have confidently crushed D'Tesh (who was struggling) with my army of Undead Slaying monks, but I just gave up there because I realised I would eventually have to face the Jotnar and then I would lose. It's too bad, it took me quite a while to get to that point (boy do I love/hate Monks, love them because of the Flavor, hate them because they're bloody useless until you can pump them out at level 4 or so... I'm going to change the Pacifist promotion for my next Elohim game).

 

[Viatos]

It was a fun, albeit slow, builder game up until then, and I would have confidently crushed D'Tesh (who was struggling) with my army of Undead Slaying monks, but I just gave up there because I realised I would eventually have to face the Jotnar and then I would lose. It's too bad, it took me quite a while to get to that point (boy do I love/hate Monks, love them because of the Flavor, hate them because they're bloody useless until you can pump them out at level 4 or so... I'm going to change the Pacifist promotion for my next Elohim game).

Yeah, I could have done this in my game no problem by tanking my research rate for like...10 turns. Huscarls are not only damn near impossible to attack, they can bombard, and you can spam Thrall Militia for them to eat.

 

[hix]

In the last two games I have popped an angel in the first 50 turns or so, and as much as I love it, it doesn't seem fair to the already pitiful AI when I can wipe out my first civ in the first 100 turns with just an angel and 5 melee units.

 

[dunedainjedi]

In the last two games I have popped an angel in the first 50 turns or so, and as much as I love it, it doesn't seem fair to the already pitiful AI when I can wipe out my first civ in the first 100 turns with just an angel and 5 melee units.

Has anyone noticed the AI with a mid game unit early game? I always seem to end up with a mage and champion or two(ancient cities makes this worse), but I seem to be the only one getting them. Seems like the AI should have gotten one, or do they not explore lairs?

 

[tufloss]

Originally Posted by hix View Post
In the last two games I have popped an angel in the first 50 turns or so, and as much as I love it, it doesn't seem fair to the already pitiful AI when I can wipe out my first civ in the first 100 turns with just an angel and 5 melee units.
Has anyone noticed the AI with a mid game unit early game? I always seem to end up with a mage and champion or two(ancient cities makes this worse), but I seem to be the only one getting them. Seems like the AI should have gotten one, or do they not explore lairs?

I have seen the AI explore lairs (Usually near my borders, popping high strength barbarians which proceed to kill me They probably do get those high level units but are not as good as a human player at keeping them alive.

 

[El Yahpo]

I can't post saves. When i try to upload them I go to my save file and there is nothing

 

[Evalis]

The issue with 'empirical' measurement is that it's just as subjective as any other method of determining game balance. Each player has different methods of play, and different ways to exploit the mechanics.. it's possible that one player (playing on any difficulty setting) would perform vastly different than another, based on reloads, world edits, or just general skill level.

The only way to measure in this manner would be to have one player play multiple games until the 'end' and then measure them against each other... not to measure each random person's games against another random person's game. Even if this were accomplished however.. it doesn't say 'what the problem' is, and it would still be up to the devs to find that out.. which I suspect would be much easier to do looking at spreadsheets.

 

[tufloss]

I can see the value of this thread even if you were to discount the effect of the player in the game. Look at the AI civs in all these games, is one civilization always on the top or bottom of the scoreboard? By looking at that information you could quickly see which Civilization needs a nerf or a buff (or if the AI needs to be taught how to play that civilization.)

Where did the Civilizations get put, does the map flavour start need to be adjusted? (And probably many more observations that could be made that are independent of each individual player)

 

[Calavente]

The issue with 'empirical' measurement is that it's just as subjective as any other method of determining game balance. Each player has different methods of play, and different ways to exploit the mechanics.. it's possible that one player (playing on any difficulty setting) would perform vastly different than another, based on reloads, world edits, or just general skill level.

The only way to measure in this manner would be to have one player play multiple games until the 'end' and then measure them against each other... not to measure each random person's games against another random person's game. Even if this were accomplished however.. it doesn't say 'what the problem' is, and it would still be up to the devs to find that out.. which I suspect would be much easier to do looking at spreadsheets.

I disagree. We are not comparing our game with the Ai's game but as tufloss said :

I can see the value of this thread even if you were to discount the effect of the player in the game. Look at the AI civs in all these games, is one civilization always on the top or bottom of the scoreboard? By looking at that information you could quickly see which Civilization needs a nerf or a buff (or if the AI needs to be taught how to play that civilization.)

Where did the Civilizations get put, does the map flavour start need to be adjusted? (And probably many more observations that could be made that are independent of each individual player)

With such an analysis you get 3 info :
-how did the AI play, against barbs/animal/AI.
-is your civ too easy/too hard compared to the AI (early game + mid game power curve : effect of the health system, no-unhealth for some civs, much health for others...)
-What did you get from goody huts/dungeon, invisible exploration...Etc
it shows how such moves influenced the game.
A turn 100 vicar doesn't change your game if you discovered him the other side of the world as you'll be eaten alive before coming back home.; But a turn 100 champion/mage kill the civs near you. or a turn 20 Thane of Kilmorphe near your border allow you to take RoK out of the dwarves mouth....Etc

 

[Pohlmann]

i just lost a game to 1 unit. on turn 160 a hidden nationality strength 5 aerons chosen assassin with decent mutation killed one of my units until i ran out of units.

12 strength 5 first strikes 20% postcombat heal are unbeatable on turn 160

 

[Castar]

Probably an enemy who popped an early Assassin out of a lair (since lairs currently don't scale with game progress), combined with the ridiculously broken Aeron's Chosen promotion.

Both are known and will be fixed/balanced afaik.

 

[hbar]

The issue with 'empirical' measurement is that it's just as subjective as any other method of determining game balance. Each player has different methods of play, and different ways to exploit the mechanics.. it's possible that one player (playing on any difficulty setting) would perform vastly different than another, based on reloads, world edits, or just general skill level.

The only way to measure in this manner would be to have one player play multiple games until the 'end' and then measure them against each other... not to measure each random person's games against another random person's game. Even if this were accomplished however.. it doesn't say 'what the problem' is, and it would still be up to the devs to find that out.. which I suspect would be much easier to do looking at spreadsheets.

There is a well known solution to this: autoplay. I believe you can set up the game to have the AI play itself. Run a minimum of 30 test games for each leader, each map size, speed, difficulty, etc. It sounds extreme, but it would provide the best data.

Of course, by the time you're done, Civ 7 will be going gold. Perhaps forgoing a little rigor in the name of results isn't such a bad compromise.

I finally have time to play this week, so I'll post a save here sometime. BTW, great idea for a thread (IMHO).

 

[Evalis]

hbar.. what? So you're saying that a human being actually playing 30 games for each leader until it's conclusion would be faster?! I somehow doubt this, even if you had 30 people doing it.. of course the answer for the AI's performance (or lack there of) is already known. The barbarians rape their sons and daughters. Add in the archos or the clan, or any nation that gets lucky enough to appear behind another and you will see. Auto terraforming is another issue (not for the malakim) because it automatically gives high food, production, and possiby commerce. Thus saving them from the barbarian rapes.

Posting save games does not provide the answer, as far as I'm aware this is an igcognito attempt to grow your epeen. All it becomes is information overload. If you recognize something out of sync with the rest of the game, that belongs in the balance thread. Don't post saves as this creates 'more' work for the already overworked dev team! Actually spend the time to detail the problem, and you'll find the changes implemented to combat them much more in tune with what was causing the imbalance.

 

[Venir]

Seems like a good idea, honestly.

Well, uh... seems the people actually making the mod disagree with you..?

 

[hbar]

hbar.. what? So you're saying that a human being actually playing 30 games for each leader until it's conclusion would be faster?!

Um, no. I am saying that the computer playing 30 games against itself would be faster then a human playing one game.

Also, I forgot to use the sarcasm tags. My bad.

 

[Tsathoggua]

blah blah blah

You're free to not swing your 'epeen' to your heart's content.

On a related note, I had no idea winning at single-player games with horrendously unbalanced factions against ill-equipped AI contributed to epeen.

Let me start up a couple more D'Tesh games.

/sarcasm

 

[Valkrionn]

There is a well known solution to this: autoplay. I believe you can set up the game to have the AI play itself. Run a minimum of 30 test games for each leader, each map size, speed, difficulty, etc. It sounds extreme, but it would provide the best data.

Of course, by the time you're done, Civ 7 will be going gold. Perhaps forgoing a little rigor in the name of results isn't such a bad compromise.

I finally have time to play this week, so I'll post a save here sometime. BTW, great idea for a thread (IMHO).

Turn on Chipotle (cheat code in the Ini), hit ctrl-shift-z, put in the number of turns you want, and sit back and wait.

The game will turn off the sound, and will make a loud alert noise (and turn the sound back on) when the autoplay is up.

Honestly, that's a big part of how we balance. We play a game or two with the mechanic to see if it's fun, and working well for the human... And then run multiple AI games.