Yeah that was unnecessary actually. 450 food, 400 gold for only +1 attack, +5% accuracy and +5 HP is really overpriced imo.
That’s actually pretty good to me. With so many civs, it is better to see more variety of units instead of just knight and archer (and CA). It is really good to see eagle (only Aztecs though) and Camels have matched with them. Now EA is close to that. Need more help for the unit or those two civs and we will see one more unit often.
*80 food. So even harder to get compared to CA.
I was thinking about to move it to Castle Age. Persians gets it for free and Saracens get it in Feudal as new civ bonuses. But all of these mean, as you said, nerfing camels, all in scouts, and even CA and similar units while just buffing Franks. So I gave up the idea.
I’m starting to like this idea as you provided the reason behind it. Also high elo players are less tend to garrison villagers. So it will hit the exact elo range where it should be.
A random thought: stagger Bloodlines into 2 techs 10 Hp each? Not sure of price but feudal castle deal. This might make feudal archers too strong if you need to rely on scout to counter archer instead of Skirms.
Any change to BL hurts scouts in feudal eve more than anything else, but also hurts LC, camel, CA, UU cav, etc.
The problem with cav, is primarily knights, nerf knights or buff their counters
And most importantly any nerf to BL is also a buff to franks of all things. And of all things franks need the nerf more than knights. Yes knights are dominant. but a big portion of that is due to franks and their almost singular gameplay.
Proof of this is Portuguese. Discounted knights, but still bottom of the ladder. So the problem clearly isn’t “just knights” either, but the ecos backing them.
Laughs in Berbers. An exceptionally good civ with very little eco bonus. Or maybe 10% faster villagers are better in practice than we what we get in scenario testing due to saving more villagers in raid.
This why other than increasing Cavalier upgrade cost as I see this very cheap, I don’t want to nerf knight stats in general. All OP civs, regardless of tag, needs to be nerfed individually.
Well the Bereber discount to knights is really strong. If you look at when Berbers win it’s usually their early castle age powerspike where they start to shine. Berbers even can go super cheap camels which I think is kinda bad design: It has a bonus to the strogest unit in the game AND to it’s counter. This shouldn’t be a thing imo.
But yes this is “compensated” by having only average eco so it’s balanced.
Portugues on the other hand have less discoun on their knights, no eco bonus at this stage and no camels at all. Ofc Portuguese could go archers, but as they have no bonus to wood they can have problems with the production. I never saw Portugues opting for 3 archery ranges in feudal. Seemingly this is just not a viable strat even with that big of a discount.
I also don’t think this is needed. I think there are several things that could be improved generally to reduce the knight powerspike - examples:
A) Stronger but more expensive walls: Yes, higher investment but also more reliable protection against a pure raiding party. The knight player then has to scout wether the opponent is fully walled if it makes sense to try raiding. On the other hand a small amount of knights would still be viable to restrict the opponent expansion.
B) Faster pikes: As knights are generally a bit slower than the scout line they would be more effected if their counters are just a bit faster. This would also possibly enable light cav play in castle age more. Cause with the supeior mobility of the light cav you possibly would have an easier task outmanouvring the pikes.
I wouldn’t say “counter”, but it’s kinda obvious that you can make a lot of damage to opponents with only 3-4 knights which is the typical raiding party size.
On the other hand if the opponent makes only 3-4 pikes + the upgrade he invested more but would still lose the fight.
Then we have the thing with the mobility what hera also talked about.
Ofc you could opt in first making just spears and see how many knights are made and then click the pikeman upgrade, but that’s also associate with risks.
Borderline meaningless on archers, worse than the port bonus on kts due to the quantity of wood , and the fact it’s wood (harvested the easiest in the game) total discount is a negligible 7% before even factoring over abundance of wood.
Coupled to the fact he said
Archer/xbows aren’t the strongest unit, so it’s more acceptable if they and their counter get a discount
In the same way that goths much larger total discount to infantry means much less because it’s not a strong line.
So like a point between stone walls and palisades? Haven’t palisades just been nerfed? Would Devs even consider reverting the nerf which then leads to more stagnant gameplay all over again? (Not that I’m against it, it just doesn’t seem to fit with Devs plan for the game)
I think it’s difficult to get the point across while Heras trying to cast, but I’m sure this has been covered by a number of casters in a few different videos. This vid also wasn’t a good example for Hera, because the opponent Never managed to get around the pikes or even bother with a counter to them. (KT + counter beats pikes + counter)
IDK if this is really the right call, but imo the TCs could just get +1 atk in castle and another +1 in imp.
Currently TCs do 3 damage per shot vs cav in castle age and imp if you have bracer. But the archer line does 3 respectively 4. Vs Paladin, TCs only do 2 damage per shot which leads to them having actually negative value damage output (If you garrison vills to shoot down Paladins, you actually lose netto ressources because of the idle time). Interestingly even hussars are close to being “effective” under TC fire currently. It is weird that TCs are basically the only thing that doesn’t scale with ages.
IDK if it’s intentional, but I would like to ask the question if it’s not probably a good idea to overthink the stale 5 atk of TCs at all ages.
I meant knights. Yes I think knights are the strongest unit in the game. Cruzify me for that, but it’s my honest opinion.
I don’t want to revert anything. I’d like to see a change to palisades so they cost more but have higher HP. In my opinion the meta has shown that the current design just leads to an inclusion of walls into the general meta. And I think with that tweak you could achieve to make walling a more strategical decision. Maybe I’m wrong with that assessment, but it’s my opinion on that topic. Make Palisades more costly but more durable could diversify the meta gameplay.
I think it’s a very interesting topic and I respect hera to have the guts to speak about it. Especially as Hera is famous for his own Knight play. I respect even more that he actually criticizes the effectiveness of his own playstyle. Is he right? I think to some extend, yes. In other aspects he is surely overexaggerating things here. But I think that’s also fine.
The poinr I wanna make is, that the Knight powerspike is real. The question for me is more… Is it really too big? Or is it actually good to have that big of a powerspike in the midgame?
I think it’s an interesting topic. And I haven’t really decided yet what path I would go. I tend generally more to just accept the knight powerspike in the midgame but would tune down cavalry and cavalry raids in general a bit. That’s why I proposed to improve the speed of the spear line. But I think there are definetely valid arguments for the other way around.
@casusincorrabil I’m quite curious as to how you arrived at that confidence interval value.
Did you use a binomial proportion confidence interval? If yes, an inherent assumption of the binomial distribution is that the probability of success is the same for all trials - which is obviously untrue for this environment.
Probability of success is variable for each trial, with elo being a far more accurate marker of that probability than civ, not to mention the probability of success is also dependant on the opponent’s civ.
If you’re using a simple confidence interval like that, imo the necessary sample size would be a few orders of magnitude higher than what is normal to do away with those inconsistencies.
I used the Formula of statistic error. If you are familiar with this kind of statistics you know that binomial, confidence and error all tend to go to the same limit at high number of tests and 50 % p value. So as it is very close to 50 % you get very close to one another with just using one of them.
And yes I used the statistical error calc as it is indeed the easiest here.
Never heard such an argument in a honest discussion about statistics… Even as Physicists where we try to eliminate all secondary effects etc. this just doesn’t make sense. It’s the wrong approach and doesn’t lead to any valueable results.
If you want to make a multi-factorial analysis, yeah. But then please use the correct approaches for this.
Are you serious? Do you even know what an order of magnitude is? 11
Well you can make demands and criticize all day, but I can tell you however you want to calc yourself you will never get far away from my 1.7 % confidence interval.
But yes, go ahead and calc yourself if you like. You are free to do so. But pleas stop just criticizing without providing any alternative calculation. That’s hypocrit.
And if I ever made such an assumption so easily without confirmation, I would get kicked out of my job.
Assuming that the average p value of each trial will tend to 50% is a big ask, and necessary if you want to assume that elo has played no factor in this result (another MASSIVE assumption)
What “statistical error calculator” are you using if you don’t mind? Does it assume the distribution is normal?
Big sigh. I know what you meant. Read what I said. I was pointing out how another civ with double discount isn’t oppressive because it applies to less powerful lines
Or maybe you missed the part where someone else pointed this out
Either you are going to make walling more effective or you won’t
I think its better if TCs get the bonus without needing a tech, otherwise it simply makes non archer matchups lopsided (ie you need to take arrow upgrades to defend, you get absolutely nothing if you don’t)
So anybody that falls behind with their own knights, or is making infantry isn’t getting any of the this benefit. When the problem still exists in those same mactches.
So while archer civs likely need it more, I don’t think they should be the only civs to benefit from it.
One of the reasons we don’t see infantry siege in open maps is due to counter raiding, which shouldn’t be that strong. It’s not only Vs archer civs that raiding shouldnt be that strong.
And obviously the infantty siege will eventually need to get the arrow upgrades, but in the interim it’s a bit odd that TCs are only doing 1 dmg to cav, when cav are already so good at everything
Is it?
I mean in theory we actually have to work the other way around and make a preassumption. Then calc the confidence for whatever probability we want to achieve, whether it be 67, 90,95 or 99 %. Then also calc both primary and secondary error… That’s the right approach if you want to test theories like this. Cause if you want to account for secondary effects you need to check if these are all accounted for and this is only possibly by theory and test.
So yeah… I don’t think this is a viable method here anyways, so I stay in a shallow error analyzis.
IDK. Maybe if you are really statistician and they have high emphasis on formal correctness. In physics we learn to assess the influence of certain mistakes and if the mistakes in a range of 2 + order of magnitudes smaller than our destined accuracy we neglect them. It’s also basically the only way how you can make physical experiments cause you will always have this kind of influence.
And as long as the deviation from p = .5 isn’t too big (.45 => .55), the error is always smaller than 2 magnitudes between either of these calculations. So I neglect it.
That’s why I told you: Please make your calc and then propose it here so we can compare it
Cause I know it’s not 100 % accurate. But it’s definetely accurate enough for the accuracy of this application. It doesn’t really matter wether it’s 1.6 % or 1.8 % here.
BTW as real statistician you should know that it’s sometimes better to stay in the shallow raw stat analyzis than trying to correct by theoretical influence factors. Cause you never really know if these are real, they apply in this way you imagine, if there are other influences you neglect or even worse you have a ### hoc ergo propter hoc fallacy.
I was curious about using the basic CI formula vs accounting for Elo in a logistic regression model so did a simple simulation. The findings were that, assuming differences in Elo is randomly and independently distributed, that accounting for Elo in the model results in narrower confidence intervals. How much more narrow is proportional to the effect size of Elo and the number of samples. This is fairly intuitive as the less samples we have the less evenly distributed Elo is.
That being said, at the effect size of Elo we see in the data the difference between the two is pretty much negligible at 10k samples.
In all cases the point estimates are the same.
Main thing to note though is that assumption of randomly independently distributed Elo scores. I haven’t checked this but based on the match making I would expect to see a slight skew as Civ picking is not evenly distributed across the Elos and if you are above average Elo you are slightly more likely to be paired with people of lower Elo than yourself (I think). Therefore I would expect that the naive formula is slightly biased and has a slight overestimated confidence interval, I wouldn’t have expected it to be orders of magnitude out though.
I think a much bigger influence is single civ picking which heavily biases win rates to 50%.
Agree, this is probably the most influential factor we have here. And it’s really complicated to factor this out, so I jsut think we have to accept it’s there.
I won’t answer any questions from you as you failed to provide any own argument here. It looks to me like you try to ridicule me. Like Nili tried with hera.
It’s even the wrong question my friend.
TBH I even think you don’t know what you are really talking about here cause your way of asking me out is very shady. But I think someone who really understands the stuff wouldn’t ask these questions in this manner cause they are basically leading into brutal misconceptions about statistics. Maybe you try to make suggestive questions here but then I even see less reason to even try to answer them.
Go ahead and show off your own calc. Then we can talk again.