(5-CP) Counterproposal: Japan Nerf

[azum4roll]

Counterproposal of:

[Vote] - (5-19) Japan Adjustment Proposals

Current bonuses on Japan: Shogunate: When a GGeneral or GAdmiral is born, gain 50% completion towards the next Great Artist/Writer/Musician in your capital +1 :c5culture: and :c5faith: to all military training and defensive buildings 10 buildings are boosted by the UA: Military training...

forums.civfanatics.com

(5-CP) Counterproposal: Japan Tweaks

Counterproposal to (5-24) Japan Nerf by pineappledan. Proposal Remove the Dojo's "yields on leveling" mechanic. Remove the yields on Defense and Military Training buildings from Japan's UA. Japan's UA gain the following effects: :c5greatperson: Great General and Admiral birth also...

forums.civfanatics.com

Current Japan UA:
+1 Culture and Faith from Defense and Military Training Buildings. When a Great Admiral or Great General is born, gain 50% progress toward a Great Artist, Writer, and Musician in your Capital.

Current Dojo (uniques in underlined italics):
2 +3 Science
3 Culture
1 +1 Supply
20 +5 Experience
Melee, Gun and Naval Melee Units trained in this city gain the Bushido promotion
Gain Science and Culture in this City when Units trained here picks a promotion on level up

As shown in the latest AI only statistics, Japan performs too well in Emperor but under par in Warlord, which points to the XP handicap being likely the problem.

Proposed Japan UA:
+1 Culture and Faith from Defense Buildings. When a Great Admiral or Great General is born, gain 50% progress toward a Great Artist, Writer, and Musician in your Capital.

Defense buildings include Walls, Castle, Bastion Fort, Arsenal and Military Base only. Current boosts to Military buildings provide 2 in Ancient and make early culture/religion game too easy.

Proposed Dojo:
Remove Bushido on Naval Melee
Remove yields on level up
Gain 1 Science and Culture per XP in this City when Units trained here gain XP (note: does NOT include training XP)

Comparison (assuming only Barracks and Dojo are built, and no handicap XP):

Level/XP	Current Dojo Total Yields	Proposed Dojo Total Yields = XP Gained
3/30	5	N/A
4/60	21	20
5/100	57	60
6/150	121	110
7/210	221	170

Comparison (assuming only Barracks and Dojo are built, and Emperor AI handicap XP):

Level/XP	Current Dojo Total Yields	Proposed Dojo Total Yields = XP Gained
3/30	N/A	N/A
4/60	21	0
5/100	57	40
6/150	121	90
7/210	221	150

As you can see, the proposed version is difficulty neutral, and is actually a slight buff for low difficulty AI and humans at low unit level assuming no Teocalli/Brandenburg Gate/Military Academy are built.
It also no longer has yield jumps in specific XP thresholds and gives a much more consistent stream of yields.

As for the removal of Bushido on Naval Melee, see pdan's thread.

 

[azum4roll]

I sponsor this.

 

[Zakuzelo]

Alright, sorry in advance, and especially sorry to do it in your thread, azum4roll, but I gotta vent here. This situation is ridiculous.

The problem with AI Japan is how the leveling mechanic interacts with the bonuses to experience the AI gets at higher levels. We know this for sure, because without those difficulty bonuses the problem doesn't exist. On Emperor they overperform, on Warlord they underperform.
That's it. That's the problem. That's not an opinion, it's a statistical fact, this is what the data everyone's basing these proposals on shows.

So why is it that every one of them is also set to remove or nerf the +1 from military and defensive buildings, the bushido on naval melee, or both? Those mechanics are not overpowered. A Japan that does get +1 from military and defensive buildings and bushido on it's boats, and doesn't break it's leveling mechanic with AI handicaps is, according to the data we're basing this all on, the eighth worst civilization in the game. When they're not abusing AI handicaps, Japan fudging sucks.
So why are you nerfing these mechanics? It was just supposed to be an early bonus that "extends its usefulness to much later eras?" It "make early culture/religion game too easy?" Nope, sorry, that's not true. A Japan that gets an easier region and early culture gains, and then gets gets a bunch more free culture and faith latter in the game as those bonuses scale, is a Japan that STILL LOSES. They're boring abilities and/or not historically accurate? Debatable, but alright, that's better, what buffs will they get to compensate? None? Well how the hell is that acceptable at all!?

Two of these propsals try to address the actual problem, and then move on to throw some extraneous other nerfs on underpowered mechanics. One of them ignores the problem entirely and just includes those nerfs. Which means, ridiculously enough, that none of these proposals are acceptable.
Why? How? How did this happen?

 

[Rekk]

Alright, sorry in advance, and especially sorry to do it in your thread, azum4roll, but I gotta vent here. This situation is ridiculous.

The problem with AI Japan is how the leveling mechanic interacts with the bonuses to experience the AI gets at higher levels. We know this for sure, because without those difficulty bonuses the problem doesn't exist. On Emperor they overperform, on Warlord they underperform.
That's it. That's the problem. That's not an opinion, it's a statistical fact, this is what the data everyone's basing these proposals on shows.

So why is it that every one of them is also set to remove or nerf the +1 from military and defensive buildings, the bushido on naval melee, or both? Those mechanics are not overpowered. A Japan that does get +1 from military and defensive buildings and bushido on it's boats, and doesn't break it's leveling mechanic with AI handicaps is, according to the data we're basing this all on, the eighth worst civilization in the game. When they're not abusing AI handicaps, Japan fudging sucks.
So why are you nerfing these mechanics? It was just supposed to be an early bonus that "extends its usefulness to much later eras?" It "make early culture/religion game too easy?" Nope, sorry, that's not true. A Japan that gets an easier region and early culture gains, and then gets gets a bunch more free culture and faith latter in the game as those bonuses scale, is a Japan that STILL LOSES. They're boring abilities and/or not historically accurate? Debatable, but alright, that's better, what buffs will they get to compensate? None? Well how the hell is that acceptable at all!?

Two of these propsals try to address the actual problem, and then move on to throw some extraneous other nerfs on underpowered mechanics. One of them ignores the problem entirely and just includes those nerfs. Which means, ridiculously enough, that none of these proposals are acceptable.
Why? How? How did this happen?

I guess you should have made a counter-proposal sometime during the 2.5 weeks it was open, then.

You can make a new proposal next session in two months.

 

[Anarcomu]

I guess you should have made a counter-proposal sometime during the 2.5 weeks it was open, then.

That's sometime the problem with the congress (not really the congress, but those using it, including me).
We're not happy with some proposal, but we don't take time to create a proper counter proposal.
Then, we complain about "this would have been better" or "that is npt the real problem". But too late.

 

[Anarcomu]

And sometime, we discover a proposal by the time it is to vote, and that's too late to propose a counter proposal ...

 

[Stalker0]

The problem with AI Japan is how the leveling mechanic interacts with the bonuses to experience the AI gets at higher levels. We know this for sure, because without those difficulty bonuses the problem doesn't exist. On Emperor they overperform, on Warlord they underperform.
That's it. That's the problem. That's not an opinion, it's a statistical fact, this is what the data everyone's basing these proposals on shows.

With respect I will refute those facts.

The Warlord data is not good data, because what it showed was that the AI is not capable of putting together a victory strategy on that difficulty. SV wasn't even a factor on that difficulty level, and time victories were a huge portion of the % because the AI doesn't know how to win at that point, it just runs out the clock.

Most civs that utilized SV as a basis for a number of their victories (japan included) saw a large drop in their win rates. That shouldn't be surprising, if SV is no longer viable...you would expect civs that utilize SV to drop in the rankings. So did Japan fall mainly because of the leveling mechanic, or was it simply that it lost one of the pillars of victory that was giving it a lot of wins. Its hard to say at this point.

There is no point in having a balance discussion on difficulty levels where the AI isn't trying to win....its already "casual" at that point so strict balance isn't necessary. We then go back to the Emperor data where the AI IS competitive, and we do our best to assess based on that.

 

[pineappledan]

1. The AI bonuses are part of the game, and we balance for a specific difficulty (emperor), which means we balance for a specific set of AI handicaps. AI handicaps changes can be done, but we need to be careful (see point 4).

2. To reiterate, the Warlord results do not matter, because we don’t balance for warlord. Even if we did, the warlord test results do not show us anything useful. With 45% of games won by time victory, the main thing the warlord tests show us is that the AI can’t reliably win within 500 turns, unless they are playing as certain civs that are laser-focused on certain VCs, and that Japan is merely not one of those 10% of civs. To say it indicates anything about the AI handicaps specifically is reading too much into it.

3. Japan’s balance is not his reliance on 1 handicap. Japan is the sum total of all his abilities. Any nerf is a nerf to the whole package. Your assertion that no nerf is needed either needs to prove that the emperor win statistics — the difficulty we care about and balance for — are bunk, or that the nerf we suggest will not actually make Japan weaker. You do neither of these things, and in fact you rely on the AI statistics to make your own point.

4. If you change Japan you change Japan. If you change an AI handicap you change every civ to one degree or another. That includes civs that are affected by XP handicaps as much or more than Japan, like the Zulu. You are arguing that the scalpel change is wrong, and that we should use a sledgehammer instead. Actually, you are arguing that the sledgehammer IS a scalpel. In order to make that argument you need to explain away why we shouldn’t care about the effect such a change will have on all civs.

 

[Legen]

The Warlord data is not good data, because what it showed was that the AI is not capable of putting together a victory strategy on that difficulty. SV wasn't even a factor on that difficulty level, and time victories were a huge portion of the % because the AI doesn't know how to win at that point, it just runs out the clock.

2. To reiterate, the Warlord results do not matter, because we don’t balance for warlord. Even if we did, the warlord test results do not show us anything useful. With 45% of games won by time victory, the main thing the warlord tests show us is that the AI can’t reliably win within 500 turns, unless they are playing as certain civs that are laser-focused on certain VCs, and that Japan is merely not one of those 10% of civs. To say it indicates anything about the AI handicaps specifically is reading too much into it.

Warlord data is of interest because it acts as a control group for how the AI is being balanced around handicaps. To say it doesn't matter because the AI struggles to win without handicaps is like arguing that the control group in a medication study is unimportant because the placebo used doesn't actually cure anything.

And the way you use a control group isn't by just checking its results in isolation, you compare with the actual intervention group. The relative differences between the results of both groups is what matters there, and you can then use those differences to raise hypothesis of how the intervention is generating its effects. The Warlord-Emperor comparison gave quite some important things to consider regarding this topic:

• Militaristic civs performed better average at Warlord than at Emperor.
• Militaristic civs that rely on leveling as a core mechanic performed worse instead.
• Japan was the civ with the largest rank variation going from Emperor to Warlord (from 3rd to 36th, 33 ranks worse).
  • Only two other civs experienced a comparable variation, Netherlands (from 32th to 2nd, 30 ranks better) and Carthage (from 38th to 10th, 28 ranks better), and they did on the opposite direction.

These all point to Japan being highly influenced by handicaps, and that the results at Emperor have less to do with the strength of the civ's kit and more with how the handicaps interact with that kit.

Most civs that utilized SV as a basis for a number of their victories (japan included) saw a large drop in their win rates. That shouldn't be surprising, if SV is no longer viable...you would expect civs that utilize SV to drop in the rankings. So did Japan fall mainly because of the leveling mechanic, or was it simply that it lost one of the pillars of victory that was giving it a lot of wins. Its hard to say at this point.

The important info you're neglecting is that Japan's kit and AI don't use SV as a basis, which makes Japan's SV wins at Emperor a topic of major concern. Remember that Japan's science yields are currently not tuned for a scientific wins in human hands, and that that AI Nobunaga has the lowest values for science/spaceship flavor (in other words, that AI isn't even trying to win a SV, it actively focus on something else instead and a SV happens nonetheless); yet, AI Japan can win SV at a comparable rate with scientific civs, who are tuned for SV in human hands and have AIs with the highest science/spaceship flavors (a.k.a. trying the most at winning a SV and actively tailoring their decisions in that direction).

Given that AI Nobunaga isn't even trying to win SV due to having the lowest science/spaceship flavors, I don't think SV should be treated as a pillar of victory for that civ. Neither human balance nor AI flavors go in that direction and, as such, AI Japan not winning SV at Warlord isn't a good argument for disregarding Japan's low placement at Warlord. In fact, given that AI Japan is actively truing to win something other than SV, we would be expecting it to perform better at Warlord than at Emperor instead.

3. Japan’s balance is not his reliance on 1 handicap. Japan is the sum total of all his abilities. Any nerf is a nerf to the whole package. Your assertion that no nerf is needed either needs to prove that the emperor win statistics — the difficulty we care about and balance for — are bunk, or that the nerf we suggest will not actually make Japan weaker. You do neither of these things, and in fact you rely on the AI statistics to make your own point.

He doesn't need to prove Emperor results are bunk, he just needs to point out that AI Japan's balance at Emperor is, in fact, practically all about handicaps. And the conclusions we can form by comparing Warlord and Emperor point out to AI Japan being the civ that most clearly relies on handicaps to perform at Emperor than any other AI, and to such an extreme that those handicaps are producing unintended results (lowest spaceship flavor AI competing in SV with those that have the highest values).

 

[pineappledan]

Warlord data is of interest because it acts as a control group for how the AI is being balanced around handicaps. To say it doesn't matter because the AI struggles to win without handicaps is like arguing that the control group in a medication study is unimportant because the placebo used doesn't actually cure anything

Using your analogy, the both of you are drawing conclusions by treating your control as an experimental group, and reading into the results given by a placebo.

Militaristic civs performed better average at Warlord than at Emperor.

That’s not what it shows. What it shows is that militaristic civs are better at getting Score. Winning a time victory says more about how the game scores actions like cities, tiles, wonders etc. than it does about likelihood of winning by an actual VC.

He doesn't need to prove Emperor results are bunk, he just needs to point out that AI Japan's balance at Emperor is, in fact, practically all about handicaps.

And he fails to prove that. As do you. You just have a correlation and a theory that you believe to be true. Maybe it is; you have made a proposal to change the AI handicaps (way too much), and I am nominally in favor of trying it, but you claim to have evidence that you frankly don’t have.

And I say again, even if you’re right, and Japan is swinging heavily based solely on the XP handicaps, we care that he is balanced on emperor, not warlord. With those handicaps, not without them. Just because an AI benefits from a handicap (big surprise) doesn’t mean you have established that is bad or unwanted. Nor does it justify nerfing Zulu and Sweden and the other dozen or so other militarist civ who also disproportionately benefits from XP bonuses so you can “fix” Japan.

 

[Legen]

Using your analogy, the both of you are drawing conclusions by treating your control as an experimental group, and reading into the results given by a placebo.

That would mean both of us thinking that the Warlord group is the one with handicaps, and the Emperor one is the one without. How is that supposed to make sense?

That’s not what it shows. What it shows is that militaristic civs are better at getting Score. Winning a time victory says more about how the game scores actions like cities, tiles, wonders etc. than it does about likelihood of winning by an actual VC.

We've been discussing about Emperor results that militaristic civs are better at forcing Score victories in the first place, with arguments about Score victories there likely being the result of Domination victories still in progress by turn 500. Which makes sense; militaristic civs are the best ones at crippling other civs, preventing their rivals from completing a victory, and many militaristic civs lack focused tools for other victory types. Score victories were, in fact, largely won by militaristic civs. If we were to buff the warring capabilities of militaristic civs further, we would expect an increase in score victories at Emperor as a result (all other factors being equal).

As such, the increase in Score victories at Warlord isn't just about the AI struggling to fulfill the requirements of the non-Domination victories, but also possibly about non-militaristic civs being slowed down a lot more by militaristic civs around them at that difficulty. The question is if such effect is by a small or large margin.

And he fails to prove that. As do you. You just have a correlation and a theory that you believe to be true. Maybe it is; you have made a proposal to change the AI handicaps (way too much), and I am nominally in favor of trying it, but you claim to have evidence that you frankly don’t have.

And I say again, even if you’re right, and Japan is swinging heavily based solely on the XP handicaps, we care that he is balanced on emperor, not warlord. With those handicaps, not without them. Just because an AI benefits from a handicap (big shocker, that) doesn’t mean you have established that is bad, or unwanted. Nor does it justify nerfing Zulu and every other militarist civ who also benefits from that XP bonus so you can “fix” Japan.

Japan's sensitivity to experience handicaps is well known, since the leveling mechanic was tried back during the Sakoku UA experiment and changed for this reason. We were even arguing on how easily it could get out of control at Deity, and how the only way for a human to replicate that was to get really lucky with the starting scout and ancient ruins.

At the time, we even had the following two feedbacks in the same day, from the same person:

I think you guys are focusing too much on scouts. On almost any map your scout can reach level 3 pretty easily, thats 25 very early on. If you get more, great, but Japan is a viable civ without getting cheesy amounts of yields from scouts. If you do get some really yields from scout XP, its just a bonus

You can get a lot of high leveled units very quickly. I think its essential to attack a CS early on, and after that try to keep moving between different AI enemies. Each turn your units should either attack, heal, or move closer to a target.

Okay I got got a start with 2 map ruins and its just ridiculous. Its far too much culture early on, I agree it shouldn't apply to scouts.

The patch notes also include this issue at its rationale (in bold):

Spoiler :

New Version - September 7th (9-7)

Hey all, New version inbound. General Smaller bugfixes and whatnot from github AI improvements and adjustments, especially settling and naval ops General Balance Bonuses from golden ages now happen whether or not a golden age is already active. 'Chaining' golden ages now equally as...

forums.civfanatics.com

Japan - new UA: +1 Science and Faith (typo: is Culture and Faith) from Defense and Military Training Buildings. When a Great Admiral or Great General is born, receive Great Artist, Writer, and Musician Points in your Capital.

• UB - regains Yields from leveling ability
• Old UA was creating weird strategies, and scaled poorly at higher difficulties. The Trade Route idea, while interesting on paper, never really came into play (and was usually just frustrating for both sides). New UA is a blend of the prior one and a static buff on some buildings, giving Japan a reliable (small) steroid, a wartime steroid, and an incentive to focus on small, highly-trained armies.

If the leveling mechanic wasn't so sensitive to bonus experience, it would likely be part of the UA by now, instead of being on the Dojo.

Also note that we care not just on balance at Emperor, but also how that balance is being achieved. Otherwise, we wouldn't have been complaining so much about culture victories being prevalent before, or how civs that had little to do with CVs were winning it so much. Or how Domination victories were essentially absent in AI tests.

And part of how victories are earned do include a case of an AI/civ that doesn't even try seriously to win SV is getting comparable SV wins as AIs/civs that are heavily focused on it. It's ok for you to want SV to be a part of Japan's gameplay, but you also have to consider that the civ, and its AI, is not currently designed with SV in mind. If Japan is to be winning SV at the current AI Emperor rates, it should be because it has both the right kit, and the AI should be focusing on it; having it as a byproduct of handicaps is not acceptable.

 

[pineappledan]

We've been discussing about Emperor results that militaristic civs are better at forcing Score victories in the first place, with arguments about Score victories there likely being the result of Domination victories still in progress by turn 500. Which makes sense; militaristic civs are the best ones at crippling other civs, preventing their rivals from completing a victory, and many militaristic civs lack focused tools for other victory types. Score victories were, in fact, largely won by militaristic civs.

Look again at the policy choices from the warlord games. It is not just domination civs, it is all civs. nearly 1 in every 2 games was a time victory and all civ VC focuses and policy branches are represented. No, the time victories do not in any way indicate that militaristic civs are doing better at lower difficulties. All they show is that game score favors wide.

Score victories were, in fact, largely won by militaristic civs. If we were to buff the warring capabilities of militaristic civs further, we would expect an increase in score victories at Emperor as a result (all other factors being equal).

That's not how any of this works. If civs got better at domination victories you would see more domination victories, and sooner. the amount of time victories would most likely go down.

At the time, we even had the following two feedbacks in the same day, from the same person:

He's talking about how the old UA giving yields on scouts was broken. He is not talking about AI handicaps. This doesn't support your argument. Quote mining 6 year old posts doesn't help your point either.

Speaking of which, you are almost every 4th post in that thread. I think you need to take a step back here.

 

[Legen]

Look again at the policy choices from the warlord games. It is not just domination civs, it is all civs. nearly 1 in every 2 games was a time victory and all civ VC focuses and policy branches are represented. No, the time victories do not in any way indicate that militaristic civs are doing better at lower difficulties. All they show is that game score favors wide.

That's not how any of this works. If civs got better at domination victories you would see more domination victories, and sooner. the amount of time victories would most likely go down.

You're not understanding the point. Militaristic civs tend to slow down the pace of non-domination victories by crippling their neighbors. This is observed at Emperor, and it is expected to also happen at Warlord. There's a reason why civ feedback sometimes includes a "this civ becomes a runaway if its AI is left unmolested, but usually gets attacked by a warmonger and is kept in check instead".

He's talking about how the old UA giving yields on scouts was broken. He is not talking about AI handicaps. This doesn't support your argument. Quote mining 6 year old posts doesn't help your point either.

The point in common between the AI handicap and the scout discussion is that the leveling mechanic is highly sensitive to experience adjustments/variance. And I think I was very clear when I said:

We were even arguing on how easily it could get out of control at Deity, and how the only way for a human to replicate that was to get really lucky with the starting scout and ancient ruins.

I doubt you're not intelligent enough to understand the relationship between the two points.

 

[Zakuzelo]

With respect I will refute those facts.

The Warlord data is not good data, because what it showed was that the AI is not capable of putting together a victory strategy on that difficulty. SV wasn't even a factor on that difficulty level, and time victories were a huge portion of the % because the AI doesn't know how to win at that point, it just runs out the clock.

Most civs that utilized SV as a basis for a number of their victories (japan included) saw a large drop in their win rates. That shouldn't be surprising, if SV is no longer viable...you would expect civs that utilize SV to drop in the rankings. So did Japan fall mainly because of the leveling mechanic, or was it simply that it lost one of the pillars of victory that was giving it a lot of wins. Its hard to say at this point.

There is no point in having a balance discussion on difficulty levels where the AI isn't trying to win....its already "casual" at that point so strict balance isn't necessary. We then go back to the Emperor data where the AI IS competitive, and we do our best to assess based on that.

Thank you for the respectful refutement.
You're right that I didn't consider the point that the AI generally can't win at all on Warlord, but I don't think it changes my point. Even if it isn't hard proof, it's certainly indicative of such. What else could have caused a gap in performance this large if not the leveling mechanic?
And I don't mean to say we should balance for Warlord, I'm just using it as a reference for how big a difference the leveling mechanic makes.

1. The AI bonuses are part of the game, and we balance for a specific difficulty (emperor), which means we balance for a specific set of AI handicaps. AI handicaps changes can be done, but we need to be careful (see point 4).

2. To reiterate, the Warlord results do not matter, because we don’t balance for warlord. Even if we did, the warlord test results do not show us anything useful. With 45% of games won by time victory, the main thing the warlord tests show us is that the AI can’t reliably win within 500 turns, unless they are playing as certain civs that are laser-focused on certain VCs, and that Japan is merely not one of those 10% of civs. To say it indicates anything about the AI handicaps specifically is reading too much into it.

3. Japan’s balance is not his reliance on 1 handicap. Japan is the sum total of all his abilities. Any nerf is a nerf to the whole package. Your assertion that no nerf is needed either needs to prove that the emperor win statistics — the difficulty we care about and balance for — are bunk, or that the nerf we suggest will not actually make Japan weaker. You do neither of these things, and in fact you rely on the AI statistics to make your own point.

4. If you change Japan you change Japan. If you change an AI handicap you change every civ to one degree or another. That includes civs that are affected by XP handicaps as much or more than Japan, like the Zulu. You are arguing that the scalpel change is wrong, and that we should use a sledgehammer instead. Actually, you are arguing that the sledgehammer IS a scalpel. In order to make that argument you need to explain away why we shouldn’t care about the effect such a change will have on all civs.

1. I'm not saying otherwise.
2. I'm not saying we should. I understand that the Warlord data isn't useful for balancing Warlord, and perhaps not even as a reference to other difficulties in general, but considering the difference in performance and how the leveling mechanic interacts with Emperor handicaps, I think that in this case it's very useful and indicative of where that over-performance is coming from.
3. I understand that, but if we have evidence to indicate that the leveling mechanic is the real issue, then that's what we should fix. You're trying to fix a flat tire by changing the oil, it doesn't make any sense. And I didn't assert that no nerf was needed to Japan as a whole, just to these two mechanics, and I think I was pretty clear on that, so I don't know what the hell you're on about with the rest of this point.
4. I'm not proposing we change AI handicaps. Maybe there's a discussion to be had there, but it's not the one I'm making here. That is that, this is this.

 

[Stalker0]

combined with other counterproposals for master voting thread