AI Wonder Priority?

[Quibblesome]

While reading the amazing game "slow and steady wins the race" by Seraiel he performs this amazing move where he gifts music to force an AI to kill a great engineer on Sistine instead of MoM (which he is aiming for).

This suggests that AI's have a wonder priority. Is there a list of the priorities anywhere? I'm wondering if its a viable play (for example) to gift an AI early stone to try to get them to build mids for you while you just make the troops required to take the city with mids. But it kinda depends on how valuable mids is to the AI.

 

[Manco Capac]

Indeed, that is a question that tickled me too, but without the code reading keen ability, I let down the questioning.

But wonder building is a pretty random thing that varies a lot in different attempts of the same maps. Sure, each leader has its own probability of wonder building. Louis XIV and Ramesses have the highest odds (yes, HC has lower odd than those two, I was surprised too) while Genghis, Shaka, Monty (e.g. ) have the lowest.

Why Freddy prioritized Sistine Chapel over MoM? I can't say for sure, but perhaps he was starting the wonder and a hard-coded script forces the AI to rush wonders that haven't been invested at all over those that are already half done.

 

[iggymnrr]

While reading the amazing game "slow and steady wins the race" by Seraiel he performs this amazing move where he gifts music to force an AI to kill a great engineer on Sistine instead of MoM (which he is aiming for).

Perhaps the AI is just getting more hammers from the engineer for the Sistine instead of MoM.

 

[Quibblesome]

Where is the code? I can have a butchers if I know where I am looking.

 

[Seraiel]

While reading the amazing game "slow and steady wins the race" by Seraiel he performs this amazing move where he gifts music to force an AI to kill a great engineer on Sistine instead of MoM (which he is aiming for).

This suggests that AI's have a wonder priority. Is there a list of the priorities anywhere? I'm wondering if its a viable play (for example) to gift an AI early stone to try to get them to build mids for you while you just make the troops required to take the city with mids. But it kinda depends on how valuable mids is to the AI.

Thx for the compliment first

I know from AI having different Wonder-priorities, as I've played a lot of games lately, and some Wonders always get built real fast, some get built real slow.

Wonders that the AI prioritizes heavily are:

SH (almost impossible to get without sacrificing a lot, anyhow, this wonder get's built so early mainly because many Civs start with Mysticism)
Oracle (get's prioritized highly, I've seen AI's chop out the Oracle in 10 turns (Marathon speed where 1 chop = 9T) anyhow, the tech Priesthood doesn't have high Prioirity for the AI)
Mids and the GLH also always go early, if one keeps in mind that the Mids are extremely expensive and that the GLH is almost as expensive because it doesn't get fastened up by a Ressource. Latest Mids I've seen were 1500 BC, latest GLH must have been somewhere around 2000 BC.

Then come the 3 Aesthetic Wonders, from those I know, that Shwedagon Paya has a higher Priority than the Parthenon, or, more Civs start with Gold than with Marble, both possible, but Shwedagon always goes shortly after SoZ, which costs a lot less Hammers.

I know sixtine has one of the highest priorities for the AI, as does the Colossus, those get built lightning-fast. In contrast to that, the HGs, the MoM, the Hagia Sophia and the Chicken Pizza get built slowly, but those are again crappy Wonders (excluding the MoM, that one is the most awesome early to mid game Wonder after Oracle, HGs are good on Huge maps) .

AP also get's built slow, but that might again be because it's twice as expensive as the Mids, because it doesn't have a specific ressource.

I know from AI, that having a specific ressource hightens the chance that they build a Wonder sped up by it. I know that, because I've seen the Taj Mahal get built in 10 turns by Civs having Marble, and I've seen Civs not having Marble not even starting on building it for 20-30 turns. Might again be coincidence.

About the late game Wonders I don't know, most times my games don't go that far or I'm so far ahead that I get the Wonders.

Maybe you can find out for us? Woldbuilder an AI, give it all techs, gift it some GE's one after another, look in which order they build the wonders?

Indeed, that is a question that tickled me too, but without the code reading keen ability, I let down the questioning.

But wonder building is a pretty random thing that varies a lot in different attempts of the same maps. Sure, each leader has its own probability of wonder building. Louis XIV and Ramesses have the highest odds (yes, HC has lower odd than those two, I was surprised too) while Genghis, Shaka, Monty (e.g. ) have the lowest.

Why Freddy prioritized Sistine Chapel over MoM? I can't say for sure, but perhaps he was starting the wonder and a hard-coded script forces the AI to rush wonders that haven't been invested at all over those that are already half done.

You don't believe me

I've just loaded the savegame where I did that move again, and didn't gift Freddy Music. Look what happened:



No joking, I know that AI has Wonder priorities. I. e. the GL always get's built with medium speed, while the HGs and the MoM get built slowly and the Shwedagon Paya gets built lightning fast. With 17 AIs in my games, the variance concerning wonders is very small, as someone always decides to build it. Interestingly enough, I do believe though, that certain AIs have even specific Wonder priorities, otherwise I cannot explain why Frederick always gets the Mids in my games. Yes, he has Stone, and Roosevelt and QSH are also hot candidates, but I've seldomly seen Louis or Ramesses build them, those two may be hindered by SH.

And concerning your point that AI doesn't prioritize Wonders where players invested hammers in: cseanny also assumed that, I don't know, as I have no proof of that, but again, I didn't test it.

This would be a perfect task for our friend DanF!

Perhaps the AI is just getting more hammers from the engineer for the Sistine instead of MoM.

Good point. Maybe they coded something like "cheap wonders, worth less, lower priority, expensive wonders, high value, high priority" . Shwedagon is more expensive than Parthenon, still, Shwedagon gets built earlier. HGs are cheap, they get built late. Anyhow, MoM = same Hammers as Shwedagon, but get's built late. TGL would fit again, medium hammers, medium build-speed.

Sera

 

[vranasm]

Where is the code? I can have a butchers if I know where I am looking.

you could check install folder for *.cpp files I think they are in folder CvGameCoreDll or something like that

 

[coanda]

And concerning your point that AI doesn't prioritize Wonders where players invested hammers in: cseanny also assumed that, I don't know, as I have no proof of that, but again, I didn't test it.

There was a thread about a year ago which tested the impact of player actions on AI wonder likelihood rather comprehensively.

http://forums.civfanatics.com/showthread.php?t=430214

Trying to summarize...
Anecdotal experience had suggested to some players that when a human put hammers into a wonder, it discouraged the AI from starting/continuing to build that wonder. Because it seemed a bit odd and controversial, people went debugging and code-diving.
My takeaway conclusions from that thread...
-The AI does not consider other players when deciding whether to build a wonder. It doesn't matter if you have 0 hammers, 1 hammer, or 499 hammers into Pyramids; 1 turn or 100 turns remaining. The AI is just as likely to start it (or stop building it and swap to something else) in any of those cases. It doesn't matter if they have a monopoly on Masonry, or every player in the game has Masonry, IND, and Stone to boot; it's equally likely to try for Pyramids.
-The AI does not care how many people already know a tech when deciding whether to research it... it doesn't care about "trade-bait" value when making it's own research decisions. So simply researching Priesthood will not discourage the AI from researching Priesthood itself, then possibly building the Oracle.
-The AI is less likely to value a wonder technology if the wonder has already been built. So building Oracle will make AIs slightly less likely to research Priesthood (you're free to try to direct the AI's tech path this way... good luck. It's a very weak effect.).
-The AI also values a technology less if a first-to-discover bonus for it has already been taken. E.g., discovering Code of Laws makes the AI less likely to research CoL (because it no longer founds Confucianism), which could provide a small edge in winning those critical races for Chichen Itza. As a side note, this is also why Philosophy is so nice to get first. Once Taoism is founded, the AI gives Philosophy almost no further value, so it tends to stay a monopoly tech for a while. Since you need Philosophy for Liberalism, a human getting Philosophy first can buy quite a bit of time for the Lib. race.

-Anecdotal examples to the contrary are tricky things... if a very small action changes the number or order of calls to the RNG, it can have a large impact on the game. The relevant example was that swapping to work one more mine caused one more RNG call, which changed the random number generated for an AI decision on whether to continue Oracle or not. Because of that change, working that one extra mine happened to let the human get Oracle instead of the AI. As this is based on random chance, there is no way for a human to make use of this without save-scumming (effectively, it lets you partially emulate the "new random seed on reload" option without actually selecting that). But changes on average would be just as likely to work the other way - by swapping to work the mine, you might easily make the AI keep building Oracle when they would have stopped if you left your tiles worked alone.

 

[Kid R]

I don't have the SDK installed anymore to check but I'm pretty sure there is no hardcoded preference for specific wonders - the code is in CvCityAI.cpp or something like that.

The wonders do have different categories though - e.g. MoM is listed in the XML as an "economy advisor" building and the Sistine is listed as a "culture advisor" building. The sistine also has a culture "flavour". Mom has no flavour. It is possible that the same AIs who like wonders also tend to be ones who like culture, so that would add a bias towards the sistine.

Following that line of thinking for some of the other ones mentioned above:

• SH = culture advisor, culture flavour
• Mids = economy advisor, no flavour
• GLH = growth advisor, growth flavour
• Paya = religion advisor, religion flavour
• Parthenon = culture advisor, no flavour
• SoZ = military advisor, military flavour

Even if it was random of course Seraiel's gambit would be a fantastic idea, since by offering two wonder choices instead of one it would reduce Freddy's chance of rushing the MoM from 100% to 50%.

 

[TiglathPileser]

Another valid question is does the AI always use Great Engineers on Wonders if it can? Or will it settle or golden age them even if wonders are available...

 

[Seraiel]

I have seen Willem not use a GE on the SoL in my current game. He did not start a GA either. I don't know if AI can save GPs (he was short before Railroad -> Mining Inc) but I don't think so, therefor, I think he settled him.

 

[TiglathPileser]

Are you sure? You can't insta-build the SoL, iirc you get just over half the hammers required from rushing with the engineer.

 

[Habitus]

-The AI does not care how many people already know a tech when deciding whether to research it... it doesn't care about "trade-bait" value when making it's own research decisions. So simply researching Priesthood will not discourage the AI from researching Priesthood itself, then possibly building the Oracle.

Didn't realise this, I was pretty sure that code diving had shown that the AI prefers tech others don't have, thou its affects aren't enough to always make it a low probability to research it.

*edit*

Hmm thats even the thread i remember discussing it, so looking thru now to see when the tech priority was discussed

*edit 2*

Just read your bit about the conditions on tech value and thats enough for me

 

[Quibblesome]

Well I had a fair old dig through the sources of BetterAI which I seemed to have lying about in my dev folder and from what I can tell there is NO PRIORITY LIST.
However there are some factors that make specific wonder building more likely which would explain a lot of the patterns that we're familiar with.

Essentially if its deciding to make a wonder the code more or less cycles through every possible wonder and assigns each one a value. It then takes the "top" valued wonder as the one that is built. This value is adjusted by lots of different aspects but some of the major modifications include:

• How many turns it takes to build (this is precisely why SoZ/Great Wall/Oracle/SH are so popular).
• What "wonder value" the leader in question has (iWonderConstructRand). It says rand because this is a value fed into the "RNG" that directly effects the size of that value (which more or less makes it not very random tbh).

These are the biggest factors. For those literate in the ways of code you can appreciate that these modifications are large:

value = (value * 1000) / (turnsToBuild + 3);

value = (value * randomNumber(seededByWonderConstructRand)) / 100

Its worth noting that there are prior factors that effect the value before the above tamperings occur. It's quite interesting because there is specific code listed for so many wonders that increment this value, such as GPP generation (Parth), increased Trade Routes (GLH), Production modifiers (Kremlin), Free Specialists (SoL/GL), Free Buildings (SH, Three Gorges Dam) but NONE (that I saw at least) that covers wonders such as Mids or MoM. Therefore I think its actually LESS likely those two wonders are built.

So here is the contrast of the smaller mods to the "value" based on the features of a building/wonder. Note, that from a code perspective it treat wonders and buildings relatively similarily.
(It tries to re-use the code for both building types but then adds in so much building/wonder specific code that it makes me think it does more harm than good trying to re-use that code.)

// kremlin
value = value + 66;
// pentagon
value = value + numberOfCities * 6;
//SoL
value = value + numberOfCitiesOnContinent * 12;

I think one of the reasons that Wonders like SH and Mids _feel_ popular is because there is a general paucity of wonders available at those times (for Mids its the case after Walls and Oracle are nabbed) therefore its likely that AIs that decide to build wonders build them. Not because the AI beelines them but because it just wants to build wonders and those are the only wonders available!

Therefore I think (back to the original points) that Seraiel was relatively lucky that Freddy built Sistine instead of MoM (I actually tested and the AI doesn't always use GEs for the most expensive wonder) but most importantly that gifting an AI stone and masonry makes the AI slightly more likely to build Mids. However it isn't deterministic and if they're already building a wonder then I don't think they'll switch. .

Oh also I looked through the research code, it's pretty similar and the AI doesn't seem to know what the player has so I don't think "tech blocking" is a valid strat. Looking at the code techs that aim for wonders have a largish random modifier on them whereas techs that allow improvements (AH, Pottery, etc) have a much higher deterministic effect on the "value". Any research that unlocks a religion has a pretty high value for religious AIs which means that AIs are probably going either Med/Poly and then either Mas => Mono or Agri/Hunt/Pottery/BW (to catch up on worker techs). Again its somewhat randomised so its really hard to rely on. However I figure any slow Oracles are due to AIs catching up on worker techs as opposed to "blocking".

------

As an aside though, that code (specifically the code that dictates which building is made) is UGLY as funk. I mean you can see above that it arbitrarily assigns hard-coded values willy-nilly (I'm personally a big fan of TurnsToBuild+3 ) to various parts of buildings/wonders which must make it a complete pain in the ass to edit/test or comprehend as a whole. I mean the randomness somewhat ruins one's ability to test anyway it in any deterministic fashion. I can only sympathise with the former devs of BetterAI and current ones of K-mod. I'm amazed that anyone could edit that with some semblance of intent, it hurt my head a fair bit just trying to read it for a few hours .

 

[Zx Zero Zx]

Well I have found there are techs and tech lines that different AI favor over others, so this wouldn't surprise me.

 

[Seraiel]

Are you sure? You can't insta-build the SoL, iirc you get just over half the hammers required from rushing with the engineer.

You can see the hammers AI has invested into something (formula by T-Hawk in Replay #4) . Willem got the GE and didn't build any Wonder with it (no hammers greater than 200 invested into anything) , but he didn't start a GA either, therefor, must either have saved him or settled him.

And I wouldn't be too sure that one cannot insta-build the SoL Value of a GE is defined by the production value of a city, all one needs is a city that's production is great enough so that the GE has a value of the SoL, though I know, the SoL is really expensive so this might be difficult, but maybe an extreme form of IW-City could still handle it.

Well I had a fair old dig through the sources of BetterAI which I seemed to have lying about in my dev folder and from what I can tell there is NO PRIORITY LIST.
However there are some factors that make specific wonder building more likely which would explain a lot of the patterns that we're familiar with.

Essentially if its deciding to make a wonder the code more or less cycles through every possible wonder and assigns each one a value. It then takes the "top" valued wonder as the one that is built. This value is adjusted by lots of different aspects but some of the major modifications include:

• How many turns it takes to build (this is precisely why SoZ/Great Wall/Oracle/SH are so popular).
• What "wonder value" the leader in question has (iWonderConstructRand). It says rand because this is a value fed into the "RNG" that directly effects the size of that value (which more or less makes it not very random tbh).

These are the biggest factors. For those literate in the ways of code you can appreciate that these modifications are large:

value = (value * 1000) / (turnsToBuild + 3);

value = (value * randomNumber(seededByWonderConstructRand)) / 100

Its worth noting that there are prior factors that effect the value before the above tamperings occur. It's quite interesting because there is specific code listed for so many wonders that increment this value, such as GPP generation (Parth), increased Trade Routes (GLH), Production modifiers (Kremlin), Free Specialists (SoL/GL), Free Buildings (SH, Three Gorges Dam) but NONE (that I saw at least) that covers wonders such as Mids or MoM. Therefore I think its actually LESS likely those two wonders are built.

So here is the contrast of the smaller mods to the "value" based on the features of a building/wonder. Note, that from a code perspective it treat wonders and buildings relatively similarily.
(It tries to re-use the code for both building types but then adds in so much building/wonder specific code that it makes me think it does more harm than good trying to re-use that code.)

// kremlin
value = value + 66;
// pentagon
value = value + numberOfCities * 6;
//SoL
value = value + numberOfCitiesOnContinent * 12;

I think one of the reasons that Wonders like SH and Mids _feel_ popular is because there is a general paucity of wonders available at those times (for Mids its the case after Walls and Oracle are nabbed) therefore its likely that AIs that decide to build wonders build them. Not because the AI beelines them but because it just wants to build wonders and those are the only wonders available!

Therefore I think (back to the original points) that Seraiel was relatively lucky that Freddy built Sistine instead of MoM (I actually tested and the AI doesn't always use GEs for the most expensive wonder) but most importantly that gifting an AI stone and masonry makes the AI slightly more likely to build Mids. However it isn't deterministic and if they're already building a wonder then I don't think they'll switch. .

I appreciate the work you do very much

Anyhow, Frederick constructing the sixtine had nothing to do with luck. I just tested this with the savegame from Replay #4:

At first, I only gifted Music to Frederick. Then I made sure that the random number was a different one (by i. e. gifiting all techs to him, gifting only music, mooving units, etc.) , which I could proove it was, because totally different AIs came asking to me (or none at all) the next round. Frederick built the sixtine in 10 of 10 cases!

I then went 1 step further and gifted Music aswell as Theology to Frederick, made sure that the RNG was different each time again, and Frederick built the AP in 10 of 10 cases!

So this has nothing to do with luck. Even if there is not "Wonder priority list" that has the name "Wonder priority list" , the factors you mention make the AI choose specific wonder over others. MoM has lowest priority, therefor, that move was no luck

What we need is someone who does the test I've proposed:

Give an AI all techs, gift it a GE. See what wonder that AI builds. The settings have to be as close to a real game as possible as we see from your work, that number of cities, maybe even population etc. matters. Then we can write "our own wonder-priority list" which of cause has no 100% accuracy, but should show very well, which wonders are valued higher at realistic settings than others .

Regards, Sera

 

[iggymnrr]

At first, I only gifted Music to Frederick. Then I made sure that the random number was a different one (by i. e. gifiting all techs to him, gifting only music, mooving units, etc.) , which I could proove it was, because totally different AIs came asking to me (or none at all) the next round. Frederick built the sixtine in 10 of 10 cases!

I then went 1 step further and gifted Music aswell as Theology to Frederick, made sure that the RNG was different each time again, and Frederick built the AP in 10 of 10 cases!

So this has nothing to do with luck. Even if there is not "Wonder priority list" that has the name "Wonder priority list" , the factors you mention make the AI choose specific wonder over others. MoM has lowest priority, therefor, that move was no luck

It may be that the AI is prone to building the first available wonder it checks for. That is the order in which the wonders are encountered. Does the AP appeared on the parsed list before Sistine?

 

[Manco Capac]

This would be a perfect task for our friend DanF!

I wish too, but he returned to his long slumber of 2 years.
Maybe during the next cycle.

Wait I'm seeing strip of codes...

 

[Quibblesome]

What we need is someone who does the test I've proposed:

I did a brief test at the same time I was checking the code. It was a Deity pangea with a bunch of civs and new seeds on reload. I gave them all three cities, masonry, aesthetics and music and watched to see what they'd build.

The civs ranged from Hannibal and Qin (40 on wonder building) to Ragnar and Jules (10 or 20 I think). Most popular choices were Great Wall, SoZ, Oracle (they spent the first 10-20 turns building archers so they had time for Priesthood) and Stonehenge. No-one ever built Sistine in like 5 or so runs of the test. However the result varied. Once Jules went straight for the ToA!?!?!
I messed Hannibal up as I accidentally gave him an extra city which resulted in him never making a single wonder (even with a high wonder making value) I'm guessing he thought he was "in financial trouble" which was a clause I saw in the code that avoids building wonders.

I tried giving them all quarried stone, that didn't stop them from making the Oracle though.

I then tried giving them GEs and it was more or less the same pattern. Sistine was like 4th or 5th choice at best.

I'm not saying you're wrong in what you've witnessed nor do I think that this test ^^ is by any means conclusive nor is my reading of the code (I am pretty new to it so my understanding may well be off). However I still presently believe from my examinations that Wonder building appears to rely on a huge number of factors including a bit of RNG so it might be hard to reliably state:

"Give Freddy music, he'll build Sistine with the GE instead of MoM"

He might do it every single time from a particular point. Perhaps his AI has looked at his environment and decided that Sistine is the _only_ thing he should build the values somehow were too large for the RNG to derail it. Perhaps he saw cultural victory as his only method of victory but its also possible that you just rolled a one of six ten times in a row (if you regenned the seed each time)!

I personally don't know if making different movements gifting certain techs or changing your worked tiles changes "the seed" as such. I've looked again at the random code and the seed is the only random part of it and it gets altered by NON random numbers each time its called. This means that as long as the RNG is called _precisely_ the same number of times then it WILL return EXACTLY the same value for the same case. As far as I'm aware the only 100% reliable way of ensuring the seed is going to be different is by setting the game to "regenerate seeds on reload".
I don't even know if any of the actions you performed would have called the RNG a different number of times. The only way you can know for sure if by enabling random logging (not sure how you do this myself) and looking for the order of the messages:

• Rand = x on x AI Best Building ASYNC
  or
• Rand = x on x AI Best Building

If they always appear in the same order from the load then you haven't triggered the RNG any more than you originally did which means that it makes sense that the results are deterministic.

 

[Seraiel]

I did a brief test at the same time I was checking the code. It was a Deity pangea with a bunch of civs and new seeds on reload. I gave them all three cities, masonry, aesthetics and music and watched to see what they'd build.

The civs ranged from Hannibal and Qin (40 on wonder building) to Ragnar and Jules (10 or 20 I think). Most popular choices were Great Wall, SoZ, Oracle (they spent the first 10-20 turns building archers so they had time for Priesthood) and Stonehenge. No-one ever built Sistine in like 5 or so runs of the test. However the result varied. Once Jules went straight for the ToA!?!?!
I messed Hannibal up as I accidentally gave him an extra city which resulted in him never making a single wonder (even with a high wonder making value) I'm guessing he thought he was "in financial trouble" which was a clause I saw in the code that avoids building wonders.

I tried giving them all quarried stone, that didn't stop them from making the Oracle though.

I then tried giving them GEs and it was more or less the same pattern. Sistine was like 4th or 5th choice at best.

I'm not saying you're wrong in what you've witnessed nor do I think that this test ^^ is by any means conclusive nor is my reading of the code (I am pretty new to it so my understanding may well be off). However I still presently believe from my examinations that Wonder building appears to rely on a huge number of factors including a bit of RNG so it might be hard to reliably state:

"Give Freddy music, he'll build Sistine with the GE instead of MoM"

He might do it every single time from a particular point. Perhaps his AI has looked at his environment and decided that Sistine is the _only_ thing he should build the values somehow were too large for the RNG to derail it. Perhaps he saw cultural victory as his only method of victory but its also possible that you just rolled a one of six ten times in a row (if you regenned the seed each time)!

I personally don't know if making different movements gifting certain techs or changing your worked tiles changes "the seed" as such. I've looked again at the random code and the seed is the only random part of it and it gets altered by NON random numbers each time its called. This means that as long as the RNG is called _precisely_ the same number of times then it WILL return EXACTLY the same value for the same case. As far as I'm aware the only 100% reliable way of ensuring the seed is going to be different is by setting the game to "regenerate seeds on reload".
I don't even know if any of the actions you performed would have called the RNG a different number of times. The only way you can know for sure if by enabling random logging (not sure how you do this myself) and looking for the order of the messages:

• Rand = x on x AI Best Building ASYNC
  or
• Rand = x on x AI Best Building

If they always appear in the same order from the load then you haven't triggered the RNG any more than you originally did which means that it makes sense that the results are deterministic.

Interesting post

First of, I can asure you that the RNG was triggert, even without using new random seeds on reload. I know that, because totally different events happened in the turn after I gifted Freddy something (or not) , i. e. sometimes someone made peace, sometimes someone declared war, sometimes an AI came to me asking me for something, sometimes, nothing, etc.

Constant was, that Freddy prefered the AP over the sixtine over the MoM.

Concerning your test, you've proven, that the number of turns an AI needs to build a Wonder, has the greatest impact on its decision, the cheap wonders were built first, ToA was a rare random-choicce.

Therefor, you must gift the AI GEs so they have 1T of buildtime, and then note the results. If sixtine was 4th or 5th choice (as you said) , note it. If there are 15 wonders to build, 5th place is high priority, if there are only 5, it's low, therefor, you must gift more techs to the AI than just Masonry and Asthetics, give them size 10 cities, gift them all techs until Nationalism. Re-run the test, and you'll get a list with the relative value of each Wonder. You can use advanced starts, to get more realistic scenarios.

Sera

 

[UnforcedError]

I personally don't know if making different movements gifting certain techs or changing your worked tiles changes "the seed" as such. I've looked again at the random code and the seed is the only random part of it and it gets altered by NON random numbers each time its called. This means that as long as the RNG is called _precisely_ the same number of times then it WILL return EXACTLY the same value for the same case. As far as I'm aware the only 100% reliable way of ensuring the seed is going to be different is by setting the game to "regenerate seeds on reload".

Setting any city on auto-build or automating workers (or setting units on auto-explore as long as there are more than one fogged tiles), moving units into the fog etc etc will affect the random number sequence Freddy gets to "decide" what to do, since apparently the game uses the same seed when having to determine actions for the human player. However, gifting more or less techs or moving units differently within your own cultural border won't necessarily make more or less calls to the RNG, you're absolutely right.
Btw I'm seeing the code for the first time in my life as well, and I totally agree with you. It makes me scream