Abstract
The aim of conducting this research was to determine whether the first general sent in an attack adds a 3% bonus to offense. In the process, it uncovered many factors about the random factor which existance had also been in question. The research was performed by a variety of methods, including dedicated research provinces, and with 50 attacks closely examined which were then correlated with data provided by other researchers. It concluded that the first general does indeed give a 3% offensive bonus, and that there is a random factor in attacks with a 95% chance of success when sending 2.5% more offense than defence.
Introduction
It is well known that when we send an army off, we need to use at least one general, and that sending more generals gives us a bigger attack bonus. There has, however, been much debate about whether the first general sent also gives the same 3% bonus that additional generals provide. This issue stems mostly from the unclear wording used in the pre-OMAC guide (note that the current advanced guide is player modified and thus unreliable). The guide stated:
- "You have a number of Generals, each of which can lead an army - you will need at least one available to send any given attack. However, you can send additional generals and gain a bonus to your offensive strength of 3% per general sent"
Many players have taken this to mean that any additional generals after the first one gives a 3% bonus each, whereas some have taken it to mean that you can send more than one general, and every general you send gives a 3% bonus. Not being satisfied with heresay and unfounded opinions, I began research to answer this question of whether or not the first general sent in an attack provides a 3% offensive bonus.
Method
Data for this research was obtained by a number of different means, each useful in their own way.
Firstly, two provinces on Genesis were setup specifically for research purposes (originally for thief and BE research), under my control (to eliminate sources of error), and set up to attack each other. Both provinces were Orc/Rogues between 500-550 acres (daily growth), having a base ME of 100% with the only change from that being increasing DME due to GB protection. Some attacks were against Town Watch defense only, although most were against a normal defence. All attacks were made with a combination of soldiers, orc elites and war horses with one general each. Both provinces were monitored closely during attacks to pick up all relevant data.
During the research I was approached by csarmi, who had previously done some research into this area with a similiar idea in mind. csarmi is well known to be a very skilled player and researcher, and getting a hold of his data proved very useful in advancing this research.
For simple checking purposes, I also started using the findings of this data in my own WoL play while playing in Sanctuary and Copy Cats in order to test the reliability of the findings.
* Note. From here in I will always refer to oversends assuming the 1st general does give a 3% bonus.
Results and Discussion
The initial results included 12 attacks with offense sent between 98.5 -103.0% of targets defence per hit, and gave the first definitive results. Two hits with 98.5% and 99.9% offense both bounced, two hits with 103.0% targets defence succeeded and, most importantly, eight hits with between 100.0-102.8% offense all succeeded.
If there was no first general bonus those eight attacks would have been undersends with 97.1-99.8% offense, and thus should have failed. Further attacks have all agreed with this, with no data clearly disagreeing, but as usual, things decided to get complicated.
Another debated issue about attack mechanics is whether there is a random factor involved, and this question came up in the second lot of data. This data, from csarmi and my further research, showed clearly that sending the same offense would not always give the same results with successes with as low as 100.7% target offense, and failed hits with 101.7% target offense in this particular data, and many reports of bounces on 'safer' attacks.
Investigation into the likelyhood of succeeding at a particular percentage oversend suggested that to get a 95% chance of success, you would have to send around a 2.5% oversend. Data taken in this manner provides too small a data set to account for extreme cases, as such a formula is too difficult to determine at this stage.
I personally believe it is in result an asymptotic function*. I have seen a number of attacks with massive oversends which have failed for no justifiable reason (and many more from nub mistakes). Whether these fails are due to a small chance in an asymptotic function, or if they are data transfer issues between a player's browser and Utopia, is an entirely different issue.
* asymptotic function in this case implies that regardless of the amount you send there is always some (albeit small) chance that the attack will fail
Conclusion
This research affirms that there is indeed a 3% bonus given by every general sent in an attack - including the first. It also suggests a random factor of around 2-3%, with the data suggesting that by a 3% oversend the chance of failure is negligible, while a 2.5% oversend gives around a 95% success rate.
The main reasons I believe this wasn't picked up before is because many people believed there was no random factor based on the fact that the attacks they sent were actually at least 3% oversends, not 0.1% oversends as they thought (as they had not included the first General bonus). So in a way this research has helped to clear up two big questions in attacking. As I mentioned, I have been hitting at around a 2-2.5% oversend normally in wol play (both while playing in Sanctuary and Copy Cats) with great success.
Acknowledgements
I would like to extend a big thank you to csarmi for his data, oyzar for his expertise, Catwalk and jdorje for allowing me to test some things in wol and clarey for editing this article and her help with the research.
Future Research
I plan on starting up more research on Genesis to look into a broad range of things including success rate of magic (both self and targetted), finish my work on thievery success rate, military losses in attacks, gains formulae, some further work into random factor and some general op gains formulae like NS and MS.
What would make this work much easier and more effective would be if I could get more people to help me by running a dedicated province on Genesis purely for this research. It will mean that the provinces won't interact with the rest of the server (if you see a prov called Research - please dont op/attack or even cb it). This does also mean whoever helps by running a province will of course get immediate access to the data and findings.
Data collation: Click here to download
If you have any questions regarding this research or would like to contribute to further research, please visit irc.utonet.org #Strategy and contact me on my IRC Nick: Clampy
Comments
39 comments postedExcellent work, looking forward to seeing more!
So basically, there is a random factor of up to 3% , and the first general is now found to give +3%
So not much has changed, apart from people who blame the random factor for bouncing just nubbed it up ? hehe
Nice research and thank you for sharing :)
Snatch, it has little practical implication yes. But the random factor is bigger than 3%, and possibly asymptotic as Clampy speculates. Which means that any attack can fail, but the risk becomes infinitely small as you send more. If that's the case, should look into cases of 1-5% oversend to see what's worthwhile. I'll probably just keep assuming +9% and oversend by +2%, habits are hard to change.
Well done research Clampy!
more more more :D
<3 rosedragon
and btw... if u guys are confused.. download the zip file and check em for more infos
i am realy realy sorry but this is not a proof for the "random factor"
the first gen have a bonus of 3%
i believe that if there is any "random factor" its will be in the range of 1%
so you will start failing around som+som*98.1% and
if there isnt any "random factor" you will start failing around som+som*97%
why 98.1???? well simple if there is a "random factor" and if its realy 1% max then the clac is simple
your off :100%
1st gen bonus: 3%
enemy real def (som+som) : 100%
max random factor : 1%
(100%+1%)/(100%+3%) = 101%/103% = ~98.1%
so if you wanto hit with a calc of a random factor of 1% you need to send above (som+som)*98.1%
if there is no random factor you need to send above (som+som)*97%
just so you guys know i didnt test it myself below 99% (just yet!)
i test it only on 99% and did 16 hits - all worked!!!
i am soon going to test it on 98% and then on 97%
thats all i have to say i think :)
lol @ all these "leet" ppl didnt even know first general counted.
i thought everyone knew the guide was wrong >.<
^^ there really isnt a random factor , inless being incompetent counts for "random" lolz
Truly leet ppl do not bother with such useless shit. They just dice ^^
That post would have made vines proud Bijo.
wow, so i read the first sentence of this crock of shit, and was forced make an account to comment.
First question that needs to be asked is: Do you smoke rocks?
Now on to the article itself. You clearly have no idea how to setup a proper test to understand what you are researching. Your test has not shown shit, except expose yourself as a complete 2 bit dumbfuck.
Lastly what Drug says is indeed correct. Now if only he can dice as leet as me.
realest so friendly
/me hands Realest a mirror
i think my post summarize it pretty well
with "random factor" its (som+som)*98.1% or (som+som)*1.01/1.03
without "random factor" its (som+som)*97% or (som+som)/1.03
(you should also add abit more off units so it would be the above value)
the champ is here :D
rofl
It's nice to see that what we suspected might be true indeed. Had a great time testing stuff like that back then (with oyzar, for example).
:-)
Your post summarizes your idea pretty well Bijo. You do not back it up in any way nor do you address anything written in the article.
Hi csarmi, are you still around? Still lots of stuff in need of testing.
Stupid multi
For simple checking purposes, I also started using the findings of this data in my own WoL play while playing in Sanctuary and Copy Cats in order to test the reliability of the findings.
While Clampy multi'd on genesis for his test province purposes, he did not multi in Copy Cats and Sanctuary as far as I know - he played briefly in Copy Cats, before moving on to Sanctuary after which he took a break from the game and is back now.
Stupid Snow ^^
Clarey is always there to clean the sheet of any cheater in a kd she luvs!
Just clearing up allegations thrown at one of our Guest Writers. Although we are pretty lax with our moderation standards, personal attacks on our staff is disliked as they are providing their time and effort to contribute to the site. If you have any issues, take it up with them or an executive member in PM.
If you disagree with his research then feel free to respond as such, or prove him wrong.
This is exactly what I love about utopia.
It has always been a numbers game - hard to argue it not being! Therefore it's very much appreciated when people team up on such projects - like it was the case several years ago when researching on the freak personality.
I'll happily offer my assistance on the Genesis, and I can only recommend others join this project too. Especially, if/when they data is being announced on uTimes.
----
Clarey, I would much appreciate if you would to setup future research "reports" like one normally do when writing BA/BsC reports. Such as e.g. the conclusion being placed on the top!
see catwalk, you presenting me a mirror has no bearing, as you are amongst the most visible clueless idiots in this game. The best example of your cluelessness can easily be seen by looking at your past attempts to figure out NS gains, Land gains, and so forth. Zero idea how to isolate variables and completely wrong experimental setups to test your null hypothesis. With the right setup, 1 tank of stealth, or 1 day of attacks can yield you conclusion you want. Next time you try to hate, make sure your house is not made of glass.
As for this experiment writeup, it clearly has not demonstrated anything, so clarey, thats why there are flames. There is nothing you can conclude from this data, except well.. that the writer is an idiot like i said. Most obviously, if you assert there is a random factor, then did the first general really mod data or was it just a bigger random factor (which doesnt exist to begin with). This article can be scrapped on the premise of "learn to calc your hits".
PS: Drug still sucks.
-------------------
moderator note: a sentence was removed here due to it being a personal attack. please speak to an Executive (&) at utimes if you have any issues Realest
"Drug still sucks" > Realest anyday.
I did briefly run two provs in gen, but only ever one in wol.
Only other thing to mention is that the newest of the data used was taken over a year ago, so I guess it is of course possible that things might have changed since, though I haven't seen any evidence to suggest as such.
To those who are asking why I'm not addressing flames, I am a scientist by profession in a multi-billion dollar company and I have learnt to ignore opinion, heresay and conjecture and instead to focus on the evidence at hand, so until someone does put forward any evidence to the contrary there is no point saying anything. If anyone has any evidence to suggest otherwise, please contact me, I'd love to look at it and talk it over.
Hard to believe a professional scientist is unable to design a proper experiment to accumulate useful data to illustrate a point. I'm not an accomplished scientist by any means, though I do have a science degree, but even I can spot the painfully obvious flaws in your design.
Inquisitive minds are great, as discussion and discovery it can bring can be fun. However, it fails to happen here.
Well as I mentioned it was a retro-active experiment, it was setup to test other things and this came out of the results, then after I mentioned it csarmi told me about his previous research into it. So all of the data in the spreadsheet was collected before I even planned to do any experiment on random factor. There are of course many ways which would give better results, but that takes time to get new data. To be an efficient researcher it is important to be able to utilise data already available to you - for my main project at work presently (which is worth many millions of dollars), most of my data has been from previous studies by others. As long as your careful about what you have and what you do with it, there's no reason a good conclusion can't be reached.
The most important part is that there are a large number of attacks which disprove a lack of random factor, therefore there must be some randominity involved. As for the chances at different rates, they're approximations and based purely from trending 50 data points without a formula being discerned - so only useful as a guide at best.
You can stop referring to your real life work, as I honestly do not care about it. What you are doing there is akin to me prefacing every point I make by saying "as you all I know, I have a 10 inch long penis". There is no way to prove whether this is true or not, and it adds nothing to the discussion.
Next, if you are saying 50 data points only make for a guide, why are you able to conclude as you presented in your abstract? You also claim not to go by hearsay, yet you are using 2nd hand data, hypocrite much?
Your data here is 2nd rate, and your method is faulty. Therefore, your conclusion is garbage. Try again next time.
Well, it is clear you are intent on living up to your reputation, and you haven't dissapointed here. I understand that you will not accept anything which isn't delivered by your own methods and are clearly more interested in personal status and attacks than any data, information or abstractions. As such I see no reason in continuing this, believe what you will, it doesn't harm me. Have a good age.
lol another failed 'research' thread?
<3 Realest
(some say his posts are worth millions of dollars)
RE:Realest: Likewise without any contribution to the actual research from yourself, you stand out to be a most spectacular asshole and a hypocrite. If you're just gonna flame and spew out non-constructive criticism and flowery words of your own intellect - then you subject yourself to that very same criticism you subject Clarey to. It isn't relevant, so zip it.
So either bring an example of how this research should be done - in your opinion, and get off your proud ass and do a bit of work yourself, before you start bitching at other people doing it all wrong. It's like going up to a ballerina doing the "Nutcracker and Mouse King" and correcting a flaw in her lines, like you had a loaf up your ass. You either show how to improve those lines. Or you stand out like a laughable piece of shit.
There: I put it all civil and nicely. And it was even constructive, in contradiction to Realest.
In response to your queries:
Only generals after the first add a bonus - that is, generals 2, 3, and 4.
There is randomness to attacks, both in offensive and defensive
strength. This randomness does, however, have a cap.
Thanks,
Roisin
roisin > clampy
Roisin verified that there was a random factor, which is what Clampy's research showed. So Clampy is, as always, very ftw.
His research also showed that the first general gives a bonus, which is wrong. So he's clearly not ftw :P
That would mean that random factor is rly rly crappy..
comming up with something compleatly incorrect seems a far way away from being correct
as for the random factor, it must be tiny as i send about .5% over generally and never bounce unless i make an error like no war horses etc.
He made an error in presentation by stating that the first general gives 3%. It is logical to conclude that it does because of results concerning the random factor (which was both proven above and later confirmed by Roisin), as it would be strange if the random factor mainly applies downwards. The distribution makes much more sense if the first general counts. Still, it's possible that a strange distribution was chosen. That minor error in presentation doesn't change the fact that Clampy is very much ftw and that there is a random factor.