Social Effects Comparative Analysis

[Alpha Centauri Bear]

Hello all mod designers and mod lovers.
I've seen many discussions and flames about social models comparison as well as countless attempts to mod them to improve "balance". Most of these mods and opinions are completely subjective and lacking rationalization. They maybe correct and right on the point but lack of objectivity denies any further shared discussion and mutual contribution in term of gradual modifications.
So I thought whether I can compare social effects using game mechanics plus some math. Here is the first draft. I will appreciate any comments or improvement suggestions. Keep in mind that I did it mostly to help out SE mod designers in weighing and balancing their creations.
http://alphacentauri2.info/wiki/Social_Effects_Comparative_Analysis

My questions to the community.

• Do you think the idea of evaluating and comparing social effects is helpful for SE mod designers at all or it is a complete waste of time?
• Do you think my idea of converting the one social effect impact to another using game mechanics is legit? In other words, is the correct way of comparing social effects in general?
• If you answered yes to both above then what do you think about my assumptions about the game progress and averaging? Is this on target? Any suggestions or statistics to add?

[bvanevery]

Most of these mods and opinions are completely subjective and lacking rationalization. They maybe correct and right on the point but lack of objectivity

My own extensive SE choice mods have been driven by lots and lots of empirical testing.  I have played many test games and made many releases of my mod.  Currently working on version 1.21, with each 0.01 representing a release, so 21 previous releases.  I say that empirical testing is a rational basis for analysis.  Indeed, to me it is the only basis that ultimately matters.  You have to look at the end results you actually get.  Various approaches, beliefs, or tools can provide help in designing a SE table, but ultimately, the results are the results.

denies any further shared discussion

Actually it doesn't deny shared discussion.  People just don't have a basis for agreement and extract whatever they like out of such discussions. 

and mutual contribution in term of gradual modifications.

Fairly quickly, I moved quite delibreately towards a gradual system of SE choice effects.  I wanted the player to usually be considering a +1 or -1 change to their SE choices, not a +2.  I didn't want the player to be able to easily blow the lid off of the top range of effects, as that's generally just a waste of resources.  In many cases I did achieve this, if not all.  My SE table is designed for much more of a "mix and match" style of gameplay than the original game.

Details social engineering choices

social engineering choices

1. Do you think the idea of evaluating and comparing social effects is helpful for SE mod designers at all or it is a complete waste of time?

The basic question is whether the comparisons somehow agree with someone's sense of the empirical reality of what happens as a consequence of different SE choices.  And whether various observers agree, and whether various observers are testing under similar conditions.  For instance, you mentioned that you test on standard sized maps.  I always test on Huge or Enormous (80x160) maps, because it's too easy to conquer things at closer ranges.  My testing gives factions more time to build up before they come into collision with each other.  There are also some distorting effects of research speed due to map size.

Also, I have modded all my factions, so their SE bonuses and penalties would need to be taken into consideration as well.  Not only that, I've changed their AI parameters and restrictions for how they play.

2. Do you think my idea of converting the one social effect impact to another using game mechanics is legit? In other words, is the correct way of comparing social effects in general?

It's not dumb, but without rigorous testing corroboration, it becomes merely an idea, heuristic, or rule of thumb rather than a verifiable statement about the realities of the game.  I think you would need some kind of standardized testbed to quantify your claims. 

3. If you answered yes to both above then what do you think about my assumptions about the game progress and averaging? Is this on target? Any suggestions or statistics to add?

I think it is hard to rigorously verify so many game variables at once.  Even my extensive playtesting over the past 5.5 months is probably undersampling the problem space.  Nevertheless it seems like I've done enough to make improvement in how things play.  I can't claim that large numbers of players have played my mod though.  My latest mod release has gotten 22 downloads in 12 days, which is typical of my releases.  If I had a larger sample of playtesters, then their own abilities would have to be taken into account as well.  I can run AI vs. AI test games, but it has to be acknowledged that the AI plays poorly in various respects.

Looking at your various proposed weights, I think your rating of SUPPORT is far too low.  My opinion is it's the single biggest game breaking variable of them all.  Ergo in my mod I "democratized" it.  No faction gets a SUPPORT bonus or penalty.  Everyone has the same basic access to achieving up to +3 SUPPORT, minus whatever political choices they can't or won't pursue in practice.  I removed most AI "secondary compulsions" so that they're not overconstrained and can more freely mix and match to achieve objectives such as "I want more SUPPORT".  And luckily, removing such constraints actually worked: the AI does indeed exhibit a greater range of SE choices according to extant game conditions.

The value of ECONOMY is skewed by the AI's unwillingness or possible inability to rush production.  You seem to have proposed a human-centric value for this weight.  That can only be valid if you test a lot of multiplayer games, which is a hard thing to do in the real world.  Even if you propose to test single-player human games vs. 6 AIs, it will take you far longer to run such tests than AI vs. AI.  And you'll probably be doing 95% of those tests yourself, as far as getting rigorous reportable results, so your own biases of playing style should be properly considered as well.

I'm not really sure if GROWTH is as important as you're making it out to be.  In my mod, following the trend you observed with other modders, I made it more difficult to pop boom.  Nevertheless the AI seems to grow big cities just fine when food is available.  I have not rigorously analyzed this growth.  I'm not sure if Transcend difficulty provides bonuses here either.  I do recall that RESEARCH and INDUSTRY bonuses are provided.

[Alpha Centauri Bear]

Most of these mods and opinions are completely subjective and lacking rationalization. They maybe correct and right on the point but lack of objectivity

Hi bvanevery.
Thank you for replying. I'm glad I caught your attention.
Now about this saying above. I knew that whatever wording I chose it'll sound bad anyway.
First of all, I didn't want to offend anybody. This is from the pure observational point of view. Of course, my observation may be incorrect, my conclusions may be incorrect, etc. That's why I am calling to your (and other modders) expert opinion. So feel free to correct me or contribute.
Second, I didn't even mean other people observation, experiments, experience are wrong. Never ever. On the contrary, I saw many people like you actually invest a lot of time in research. I am eternally grateful for all the empiric and statistical input I can find here and in other places. Thank you and other researchers for that!
What I meant is that besides statistical data those can be verified, any attempt to validate and compare society effects and models are subjective and empiric in the way that you cannot actually explain this knowledge to others so they can follow the chain of your thoughts. So you cannot use help of other contributors to criticize of improve your findings. That's all. I didn't mean to say that your findings are wrong. In fact, I believe they are pretty close to the truth. So I devised this math framework to involve people like you with your ideas and feelings.

I hope it clears this little misunderstanding.

[Alpha Centauri Bear]

Fairly quickly, I moved quite delibreately towards a gradual system of SE choice effects.  I wanted the player to usually be considering a +1 or -1 change to their SE choices, not a +2.  I didn't want the player to be able to easily blow the lid off of the top range of effects, as that's generally just a waste of resources.  In many cases I did achieve this, if not all.  My SE table is designed for much more of a "mix and match" style of gameplay than the original game.

I've read your mod. It's pretty great! I like most of the ideas there including the one above. That is definitely the way to go. It adds to variations allowing more different mixes yet keeps player from reaching the top levels quickly let them work on it. Cool stuff.

Anyway, I think I should remove these stupid words from there. They are going to be misinterpreted in worst way by many people.

[Alpha Centauri Bear]

Looking at your various proposed weights, I think your rating of SUPPORT is far too low.  My opinion is it's the single biggest game breaking variable of them all.

The SUPPORT to INDUSTRY is actually the easiest conversion of them all. They both result in minerals. You don't even need extensive empirical testing for that. Just pure math.
One more unit to support frees one mineral. So if your base puts 11 minerals into production each turn then +1 INDUSTRY will do the same effect as additional unit to support. Hence at this point they are equal. With less minerals per base SUPPORT will weight more than INDUSTRY and vice versa. That is what it actually says on my worksheet. SUPPORT weigh 1.33 in early game and then falls as you build up your mineral output. Seems pretty logical and straightforward for me.

The value of ECONOMY is skewed by the AI's unwillingness or possible inability to rush production.  You seem to have proposed a human-centric value for this weight.

Well, of course! The use of this weight is to detect SMs those would rarely be used by human. That's why this approach targets SE attractiveness for human. Other than I don't think anyone cares much whether AI is using some models more often then others or how much it benefits from them.

I'm not really sure if GROWTH is as important as you're making it out to be.  In my mod, following the trend you observed with other modders, I made it more difficult to pop boom.  Nevertheless the AI seems to grow big cities just fine when food is available.

GROWTH is your future yield. As my emulation shows, +1 GROWTH makes your empire grow 12% faster. That translates to minerals yield increase and energy yield increase, the latter in turn goes to labs and energy reserves (those can be used for rush production). So it should equate to about 1.2 * (1.25 INDUSTRY + 1 RESEARCH). Which is pretty hefty bonus and that is what you see in my worksheet. Of course, this is based on base growth emulation which may be far from reality. That is why I like to verify my result with others.

[bvanevery]

Overall I think you're doing fine... there's just so much detail, that anyone trying to do what you are, will have a long way to go.

I believe you are making an important mistake on SUPPORT vs. INDUSTRY.  If you achieve +3 SUPPORT, you get support equal to 4 or the size of the base.  The relationship is nonlinear.  It's complex according to base size.  If your model were to compare base sizes, you'd also need to compare a faction's ability to grow and hold onto bases.  Which invokes map size, start position, and quality of defense.  This makes other SE factors non-trivially important as otherwise you aren't going to have beautifully large bases.

A case in point is the Pirates.  At some point in my modding I recognized that they had this giant moat around them, and huge minerals and energy advantages over other factions.  Rather than having them be the annoying cusses they are in the unmodded game, I made them Passive entrepreneurs who pursue Wealth.  As a result, they are one of the most dominant AIs in the game now, because they are leveraging the huge geographic advantage they have over other factions.  This advantage has been noticed by other modders, who have had different responses to it.  A common one is to make more than 1 aquatic faction so that the Pirates can't so easily rule the roost.

The use of this weight is to detect SMs those would rarely be used by human.

But you are assigning the weight, not detecting something.

Also if your end goal is to opine what SE choices are "rarely used", your exercise is inherently fruitless.  All of the models get used at various times in a game according to what faction one is playing and what else is happening.  There's no such thing as a "never used" SE choice in unmodded SMAC.  My weight of evidence in this regard is over a year's worth of AARs available in the AAR section of this site.

Other than I don't think anyone cares much whether AI is using some models more often then others or how much it benefits from them.

Um, I think it's fundamental to improving AI performance without actually patching the .exe.  Also, the AI vs. AI behavior is the only thing you can test relatively rapidly.  Relatively rapidly.  I sure spent a lot of time today waiting for results, and it's delaying the release of my next mod iteration.

That translates to minerals yield increase and energy yield increase

You're assuming a lot about the quality of the land at your disposal, the speed at which you can improve it, and its defensibility.  Particularly from an onslaught of Needlejets against your Formers.  GROWTH unfortunately does not offer a straightforward benefit, because there are several other limiting factors.  One thing is certain though: any faction starting on the Monsoon Jungle dominates the early game.  Until possibly someone shows up midgame to take it away from them.

[Alpha Centauri Bear]

I believe you are making an important mistake on SUPPORT vs. INDUSTRY.  If you achieve +3 SUPPORT, you get support equal to 4 or the size of the base.  The relationship is nonlinear.  It's complex according to base size.  If your model were to compare base sizes, you'd also need to compare a faction's ability to grow and hold onto bases.  Which invokes map size, start position, and quality of defense.  This makes other SE factors non-trivially important as otherwise you aren't going to have beautifully large bases.

You are absolutely correct. Many effects are non-linear. That is why it is pretty difficult to determine effect weight as a value of a level change on a scale as this change value varies depending on the current position on a scale along with other variables, of course. Imagine that you are to compare two models and one of them has SUPPORT while other doesn't. How would you value a contribution of SUPPORT in the first model? If we are somewhere around 0 on a SUPPORT scale then the weight is average (as in my calculations). Whereas if we are at +2 and the average base size is much larger than 4 and we support a lot of units per base then the improvement will be significant. However, we need to average it somehow to say that "the contribution of +1 SUPPORT in the model worth this amount with all other conditions equal". So yes, this averaging approach is quite subjective.
I thought about new averaging technique which is in line with non-linearity. I am going to estimate the impact of each transition on a SE scale and take weighed average of them assuming you are more often in the middle of the scale and rare on extremities. That is all, of course, with regard to some normal game circumstances. Like GROWTH is usually get +2 due to Children Creche pretty early in the game and later on you are more likely to end up with positive rather than negative values. So on average you'll be somewhere in +1 to +4 range with values beyond this range having declining probabilities to occur. Same story with SUPPORT and other non-linear SEs. This way I can account for possible impacts of all values on a SE scale yet give them different probability to occur.

The use of this weight is to detect SMs those would rarely be used by human.

But you are assigning the weight, not detecting something.

Also if your end goal is to opine what SE choices are "rarely used", your exercise is inherently fruitless.  All of the models get used at various times in a game according to what faction one is playing and what else is happening.  There's no such thing as a "never used" SE choice in unmodded SMAC.  My weight of evidence in this regard is over a year's worth of AARs available in the AAR section of this site.

It's just a matter of wording. I am assigning value to SMs for comparison purposes. This comparison is mostly for SM balance. In other words, you calculate SM values then you use these values to adjust them so that none of them is significantly lower than others. So the detection part is to find discrepancies and correction part is to balance them.
Again, this is the matter of perception. Surely you can use any model sometimes. However, if you use one of the model pretty rarely like not even in each game, then you perceive it as inferior and worthless. You probably can eliminate it from the game and don't feel a difference. My point is to make all of them more or less equally usable.

Other than I don't think anyone cares much whether AI is using some models more often then others or how much it benefits from them.

Um, I think it's fundamental to improving AI performance without actually patching the .exe.  Also, the AI vs. AI behavior is the only thing you can test relatively rapidly.  Relatively rapidly.  I sure spent a lot of time today waiting for results, and it's delaying the release of my next mod iteration.

Imagine we are about to start a game with completely random factions both you and AI (or all AI). Can we say that redesigning SMs would make AI play better? I don't think so. Social engineering is there for everybody with same exact effect. On large average scale no changes in social engineering makes AI play better. Of course, some faction would do better with some choices amplifying their innate abilities but this is exactly what I abstract from in my calculations by globally averaging everything. In this study I am trying to give better game variability perception to human player and don't care about AI performance.

That translates to minerals yield increase and energy yield increase

You're assuming a lot about the quality of the land at your disposal, the speed at which you can improve it, and its defensibility.  Particularly from an onslaught of Needlejets against your Formers.  GROWTH unfortunately does not offer a straightforward benefit, because there are several other limiting factors.  One thing is certain though: any faction starting on the Monsoon Jungle dominates the early game.  Until possibly someone shows up midgame to take it away from them.

Yep, I assume a lot. One cannot get far in research without assuming first. Later on these assumptions can be adjusted to better values if they were too off initially. That is why I present my assumptions and subsequent computations to the public to get a feedback on them. By feedback I mean some constructive criticism of particular parameter, assumptions, and technique rather than "You're assuming a lot".

Besides, random resource bonuses like Monsoon Jungle have nothing to do with relative SM values. They are additive, you know. Regardless if you are lucky to be in Jungle or not you still have option to chose some +X GROWTH model if you so desire. The benefits of resources and SE will stack. In other words if you feel like SM A is much better than SM B on average then you most likely will feel the same under any game circumstances be it Monsoon Jungle or something else.

By the way, I thank you for all the input. It shows me that the concept of SE weight is not easy to grasp. I am going to add more explanatory words in my article.

[bvanevery]

Imagine we are about to start a game with completely random factions both you and AI (or all AI). Can we say that redesigning SMs would make AI play better? I don't think so.

You are wrong.  This is proven on my box using my mod all the time.  Of course one has to set an expectation for how much better.  There's a point at which I can't make a silk purse out of a sow's ear, it takes .exe patching to do that.

Social engineering is there for everybody with same exact effect.

Not true.  Most factions have restrictions of things they can't use, and leanings towards things that make it more optimal for them to use.  This is inherent to the notion of "having factions".  If we didn't have factions, we'd just have generic nationalities ala Civ II.

In this study I am trying to give better game variability perception to human player and don't care about AI performance.

But this doesn't have to be inherently meaningful, because the dynamic range of human strategic choice is much larger than the AI's.  What do your equations have to say, for instance, about a Recon Rover rush?

[Alpha Centauri Bear]

Imagine we are about to start a game with completely random factions both you and AI (or all AI). Can we say that redesigning SMs would make AI play better? I don't think so.

You are wrong.  This is proven on my box using my mod all the time.  Of course one has to set an expectation for how much better.  There's a point at which I can't make a silk purse out of a sow's ear, it takes .exe patching to do that.

We are talking about different things here. I do not argue with your findings because I didn't do similar research. I definitely may be wrong. However, I still cannot grasp the idea how generic and gradual adjustments in social engineering make random AI faction generally tougher? Let me repeat myself. You are playing random faction against random AI factions with certain social engineering table. Now you tweak one of the SM giving it more INDUSTRY, for example. Are you able to tell now based on your empiric data whether it makes your next random game tougher against AI and by how much? I feel like even if such dependency exists it would be impossible to detect.

Social engineering is there for everybody with same exact effect.

Not true.  Most factions have restrictions of things they can't use, and leanings towards things that make it more optimal for them to use.  This is inherent to the notion of "having factions".  If we didn't have factions, we'd just have generic nationalities ala Civ II.

You are right again and again we are talking about different things here. That is true that faction abilities combined with certain SM produce slightly different effects. Add to that faction choice abilities and preferences. So yes - certain faction may be strengthen by certain social engineering tweaks. However, if you randomize them and play many games then sometimes AI will be that powerful faction and sometimes you. So on average it would give human player as much change in benefits as AI. So again don't see how social engineering help human over AI or vice versa.

In this study I am trying to give better game variability perception to human player and don't care about AI performance.

But this doesn't have to be inherently meaningful, because the dynamic range of human strategic choice is much larger than the AI's.  What do your equations have to say, for instance, about a Recon Rover rush?

Nothing, not surprisingly, as this is not a scope of my research.


We are studying and working on different things. You are fine tuning social engineering to achieve better faction power balance. I.e. to make each faction win about same number of times. That accounts for all faction specific parameters and possibilities.
I, on the other hand, focus on comparing relative SM usability. I.e. perceived value of some SM choices over others. Such things like "Democracy is overpowered". This is not an exact study but some practical help tool that allows quickly and dirty evaluate some random SE combination to decide whether it makes sense at all or should be discarded as not viable (or adjusted).
Yet we probably can share some thoughts on intersected areas even though our goals are different.

[bvanevery]

My thoughts are, we are having a classic argument about how to benchmark stuff and what is even a valid benchmark.  You don't think you can detect things about AIs; I think you can detect far less about human play.  You do not have enough humans with enough variation in expertise to run enough games to provide enough samples.

You can try to funge the mathematical relations, but they are complex, with many dimensions, and not easily tractable.  Even if you try to compare SUPPORT and INDUSTRY directly, I think you are going to find they're not equivalent.  You would need to analyze a multi-dimensional graph of the following inputs: cost of units to produce, number of units to produce, size of city in which they are produced, starting minerals of city before production begins, ongoing growth of minerals if the production time is long, SUPPORT number, INDUSTRY number.

If the goal of contemplation is to produce new SE tables, you can do that by simply producing a new SE table.  Why would you choose to produce any one SE table over any other?  In my own work, I eventually adopted a consistent regimen of bonuses and penalties.  3 bonuses, 2 penalties for Political, Economic, and Values choices.  4 bonuses, 2 penalties for Future Society choices, except for Eudaimonic.  That's because 2 Secret Projects get rid of the penalties for the other 2 Future Society choices.  There's no way to get rid of penalties for Eudaimonic, so I gave them only 1 penalty.

I deliberately adopted a regimen of graduality.  Bonuses and penalties are +1 at at time in many instances, far more instances than in the original game.  Especially, rather than give a -2 in 1 category, I give a -1 in 2 categories.  This allows the player to "take a hit" with some choice and often not be greatly affected, if they only take one hit in a category.  It's easier to spread penalties across 2 categories than to take it all in 1 category.

There is certainly no -5 POLICE, -3 PLANET in my regimen.  I think that's ridiculous.  I think Free Market being some total peacenik category is ridiculous.  I spread those penalties across many other SE choices so that the effect would be gradual.

I believe SUPPORT is the most influential bonus in the table, so I made obtaining it rather gradual, and didn't allow factions to have SUPPORT bonuses.  The latter actually took a lot of design iterations and releases to decide that was the right thing to do.  The uniform number of bonuses and penalties wasn't arrived at immediately either.

I have a bias against PROBE.  I wouldn't mind it being out of the game entirely.  It's overpowered, in that you can buy an entire city full of military units for only the cost of the city, not the units.  I took most PROBE bonuses out of the game.  This did have a desirable effect: the AI stopped spamming Probe Teams quite as much, because it didn't perceive them to be quite as valuable.

I could go on, but basically I arrived at a new SE table by lots and lots of design iterations and playtesting.  You can try to get a theoretical handle on how SE relationships work all you want, but ultimately, all that matters is you actually make a new SE table.  Given that I just put 5.5 calendar months into iterating on this, it'll be interesting if you can come up with even a theoretical basis for making a better one, or a substantially different one.  While still retaining something like the flavor of the original game, not just going off into "hey let's give everyone Talents!" territory.  Or "once you get Power you're AWE-SOMME!!" and can totally stomp everyone because you've gained the golden choice that puts all others to shame.  Making sure the table does produce the distinct bonus and penalty ranges of the original game, for the most part, is a serious constraint on the number of sane tables possible.

Oh, and did I mention that I changed the entire tech tree, so that Future Society choices would come mid-game?  And not just give you overpowered uber endgame bonuses, that you don't even need when you've got Quantum and Singularity engines anyways?  Quite deliberately, players have far more flexibility of SE choice from midgame onwards, than in the standard game.  Being able to gradually tweak SE choices up or down +1 or -1, means you're a lot less constrained.  Ergo, no such thing as a rare SE choice in my mod.  Doesn't happen.  It's deliberately a game of inches, 1 bonus or penalty at a time, just as someone would do in a tabletop board game.

The way to make a new SE table, is to decide what you don't like about the game, and then endeavor to fix it, as expressed in the SE table.  Which will inevitably draw faction bonuses and penalties into the design.

[Alpha Centauri Bear]

Checked out and verified some formulas.
Updated some calculations with more detailed analysis adding more variables and tried not to cut corners wherever I can.

Some notable changes in results:
Economy goes quite high showing highest weight among other effects, which is more or less expected.
Support raised a little thanks to bvanevery for drawing my attention to it.
Same story with the Police.
Morale and Planet - I've reviewed my approach to them a little. They still stay as most speculative estimates, though.

[bvanevery]

You could learn something differentially about SE values, if you ran AI vs. AI tests, on the same faction for all 7 players, changing only 1 variable by +-1.  So any given SE choice, you could test the -3, -2, -1, 0, +1, +2, +3 versions with AIs, by making custom factions.  That may only measure starting / early game advantage, as when the game progresses, more SE choices may blow the lid off of the value range.  Although, it may also result in cost shifting, if the AI is given flexibility to choose whatever they like.

[Alpha Centauri Bear]

You could learn something differentially about SE values, if you ran AI vs. AI tests, on the same faction for all 7 players, changing only 1 variable by +-1.  So any given SE choice, you could test the -3, -2, -1, 0, +1, +2, +3 versions with AIs, by making custom factions.  That may only measure starting / early game advantage, as when the game progresses, more SE choices may blow the lid off of the value range.  Although, it may also result in cost shifting, if the AI is given flexibility to choose whatever they like.

That's a great suggestion. Thank you, bvanevery. I probably should try it to get alternative point of view on this.

Few question regarding that if you don't mind.

How do you set up AI only game? I never did this before. I probably can figure it myself too but you can push me there faster.

What are criteria you use to measure effect of your changes? Like you give some bonus to one faction and then what? Do you count how much more often it wins? Do you measure how bigger faction power it gets at certain point in time? If so, how do you measure faction power? Game doesn't give you number just a chart of faction relative powers. Or you just take their current score into account? Or something else? In other words, what kind of statistics you gather to answer your own questions?

How do you stop and restart AI only game at certain points? Like if you want to stop at 100th turn, write down some statistics and then continue?

How do you make it run fast? Do you blacken screen so you don't see AI moves?

[bvanevery]

How do you set up AI only game? I never did this before.

See this recent post of mine.

I probably can figure it myself too but you can push me there faster.

It's actually quite a bit more painful to figure out than you would expect.  I posted, got feedback from people... finally figured it out by trial and error.

What are criteria you use to measure effect of your changes? Like you give some bonus to one faction and then what? Do you count how much more often it wins?

I'd stick to gross effects at first.  Like if the factions with the most bonuses consistently and overwhelmingly wipe the floor with the other factions, that tells you something about the relative value of the bonus differences.  If one faction has +3 and another has +2, does it take a long time to wipe the other faction out?  Or is it over quickly?  Do they stalemate?  You'd need to watch a fair number of games to "get a feel" for how much advantage or disadvantage a particular SE choice gives.  But if you watch enough games, you'll start to see patterns.

Do you measure how bigger faction power it gets at certain point in time?

Probably not, because you'd be privileging arbitrary cutoffs.  Maybe it's harder for everyone to fight on a particular map, for instance.  Maybe one side of the map has it easier than another.   Maybe a specific faction has trouble with something, that another faction doesn't.  So 7 Spartans on the same map may not get the same outcome as 7 Morganites.

How do you stop and restart AI only game at certain points?

Oh, that's easy.  Remove the weight you've made out of pennies and duct tape from the Enter key!  Running AI vs. AI is seriously low rent, they didn't plan this capability much.  Sometimes the game is going to stop of its own accord.  Then you'll have to switch your view to a different faction to get it going again.  Also you can't multitask with your machine while you're doing this, you have to let the thing run monopolizing your computer.  I suggest having something else to do, like reading a book!

Like if you want to stop at 100th turn, write down some statistics and then continue?

You could just watch it like a ham cooking in the oven.  If you overshoot it a little bit, that would be ok, because you do have autosaves.

How do you make it run fast? Do you blacken screen so you don't see AI moves?

It is faster to not use the Omniscient View.  OV shows you what's happening with all the factions.  The default is to only show the faction you are watching.  Once you've activated the Scenario Editor, I think OV is switched on and off by hitting the 'Y' key.

You have to watch all the AI moves.  So it is best to enable options like Fast Battle Resolution, etc.

[Alpha Centauri Bear]

Thanks, bvanevery. I'll try everything out.