Rise of the Witch King

Intro:
gamereplays.org/community/index.php?showtopic=755440

Not very conclusive. This time around, things are getting more prostrat as I have created a macro to do the data collection part and as such I can execute many tests in a relatively short amount of time with me doing something else. (+/- 35 sec per run or 822 in 8 hours). Huzzah.


First test (control)

NOTE: The macro uses the keyboard to get to FoI2 (10 down presses). This
is different from what most players do.
Results:


Actually better than expected, still men is obviously broken. The probability of any one faction not being picked at least once is terribly small and horrible to calculate. To give you a feel, this is probably a good approximation:
Chance of motw not being picked in 144 trials: ((6/7)^144) = 0.000000000289 (2.289e-10) or 0.0000000289%
Since there are 7 factions that could have not been picked, multiply by 7 gives: 1.6024e-9 (I know I'm ignoring overlap).

Fun comparison: the probability of somebody randomly guessing a 4-digit PIN within 3 attempts is about 0.0003. (3e-4). That's more than 180000x more likely to have happened to me than this outcome (if rotwk RNG was fair).

So what I wanted to ask was, which factors would you like to test? Skirmish profiles? Video/sound settings? Game specific settings (colour/team/map)? Good/evil CaH settings (is is 1/4 probability to get mordor with an evil CaH or is that screwed up too)?


Would be cool to have RandomEvil and RandomGood crappy CaHs in 2.02 v5.0 that aren't worth their money.


Second Test (map)

So... I ran a new test with 3 simultaneously executed, alternating scenarios (to minimize the effect of time-based factors). Unfortunately, I played some games before doing the run and I forgot to reset the colour so now all runs have player 1's colour set to green.
Scenarios:
1. Control scenario
2. Argonath (first map in the list, so no action to select map)
3. Argonath (scrolling 5 maps up and down again to match the 10 key presses in the control scenario)

Results:


Conclusion: The differences between the 3 scenarios is relatively small. Probably the map setting and scrolling through the map list does not effect the outcome of the RNG.

However, test 1 and test 2's scenario 1 are effectively the same (except player 1 colour), these can be compared. What's striking is the difference mainly in motw (0% -> 16.9%), mordor (6.3% -> 18.11%) and isengard (19.4% -> 7.1%). Remember the only difference is player 1's colour!


Third test (colour)

Since the previous test seemed to indicate player 1's colour affects the random generator I decided to dive more into that. Again, the scenarios were run alternatingly. Colours used: ?, pink, turqoise and black/grey.

Results:


The variations are not as big as they were in the green colour scenario. Elves 30% (pink) -> 19% (turqoise) is pretty substantial though. Especially since the mean is supposed to be 14% (or 17% if you subtract motw).
Anyway, after I run Damul's test I will test ? vs green again with >800 samples check whether colour really affects the RNG. This test does not show a whole lot, sadly.


Fourth test (green revisited)

The following test distinguishes colour (picked) an colour (preset). A picked colour was altered using in-game controls, a preset colour was set in the 'network.ini' configuration file located in 'appdata' prior to starting the game. In the first two scenarios, whenever the colour green is picked, it was set to '?' and vice versa (since the game saves the previous colour setting).

Results:


Yeah so this isn't making much sense. It seems up to now all scenarios that are run simultaneously produce more or less the same results. Next up:
- How do CaHs influence the RNG (e.g. if you select an evil CaH is the probability 1/4 you get angmar?)
- How does bfme2's RNG compare to that of ROTWK?
- Does removing/adding a faction help?


Fifth test (CaH)

OK so, when an evil CaH is picked in combination with random, the RNG won't select good factions. For this test, I selected Thrugg (the default Uruk CaH).

Results:


The distribution of Mordor, Angmar and Goblins seems fair. Isengard messes it up by having about 2x the probability. It could be RotWK developers wanted to create a system where the race (uruk -> Isengard) of your CaH influences the RNG and in the process ended up messing with the core workings of it.

This post has been edited by brabox: Jul 18 2015, 10:12 AM

Good read, keep this coming.

Not a good read, I don't like math. Give me sci-fi, fantasy, guns, and hamburgers...please.

I'd like to help with this if you're open to it. If you give me the set of variables and method you're after I can do some additional data gathering, it would be manual however. I can do this in the background whilst studying.

Try on other 2 maps to see if percentage are similar. As motoma is, so i'm interested about "how do you did it?". Do you have same percentage if you start it from network? Have you tried some other tests with the same condition?

I could do some graphic statistical but i need more data and anyway " how do you did it".

Anyway where is the code that choose random team?

Sorry bushit, I should've known this sort of stuff scares you




I'm using a AutoHotkey (FOSS program) script I could send you, then you don't have to do it manually. You'll have to edit the script though to make the mouse click coordinates match your resolution I guess but that's maybe 5-10 min work.
What the script does is click my rotwk shortcut, go to multiplayer->network, create game, select FoI2, create brutal opponent, play game, F12 screenshot, kill game.dat, repeat. (CaH is disabled in my network profile, so I don't do that every time)


That's a good suggestion, I will probably test that first. It's good to start simple and eliminate factors.

Just fix the random somehow for 5.0. Add the chance of getting men/mordor/goblins by x10 and reduce dw/ang x20

"somehow" is the problem... which are parameters that influence random? is quite simple make percentage (1/7) but where are codes?? BOOOH .

The code that actually does the RNG is compiled into game.dat and as such can not be studied, sadly.


If it were this simple it would've been fixed long ago. As can be seen from the data, the chance of getting motw in one specific scenario is (very close to) 0%. By what number would you multiply it? Before this can be solved more samples must be collected.

gamedat can be read, its whether or not your willing to put in half your life getting through safedisc. Some folks have done some limited work with it

Awesome, I enjoyed your last topic on this and this one too. It's worth testing everything you can so we can get a good idea of exactly what affects the RNG algorithm.


Can you clarify what you mean by graphic statistical? Assuming you had a data set showing the percentage changes of getting each faction in different situations, what would you do with it?

All the game's core source code is compiled in game.dat, but it's encrypted using Safedisc 4 technology and the source code cannot be accessed without some hacking skills. No one has figured out how to do it yet though as far as I know, but it's something that is being looked into. The changes you see in 2.02 are really just changes to configuration files (and some others) that aren't encrypted and can easily be extracted, read, and modified with FinalBig.


That's not likely to be fixed because we can't access the RNG algorithm code. If we did fix it though you would have a 1/7 chance of getting any faction.


Well in theory it can be studied, but it would require unpacking it and figuring out how to get past Safedisc 4, which requires some pretty impressive hacking skills, or a lot of time, or both. Multiple people have tried to figure out how to do it over the years. As far as I know none have yet succeeded.

My brother like to program (he is graduated in informatics in security system) i could ask him something about (hacking skills). about graphic now i thinking that isn't useful...

(i now that this hacking is really hard... but i try)

This post has been edited by Mako: May 20 2015, 14:52 PM

You could ask your brother to take a look if he is interested, but it's going to be time consuming either way.

After safedisc there might be even more encryption if you're unlucky. Besides, if you rename game.dat to game.exe you can see it's actually an executable itself. Reverting 12mB of object code to something like C is to my knowledge already quite impossible.

However, since we all know 1v1 games where motw was obtained from random, there must be some way to influence rotwk's RNG. It's a matter of finding out what methods work best!

Soo... who here wants random factions revealed on the loadscreen? I can't fix random, but I can give you that!

You have been trying to do that for a long time.

I'm OK with it, it does remove the early-game element of surprise, but since that's pure luck and chance thus is crappy IMO.

Imagine having Elves vs Isen. Neither of them scout eachother. The elven player takes the risk and goes for a Lancer start. The Isengard player takes the risk and goes for a Wildmen start. The Elven player actually won, not out of skill but out of pure luck.

But by having factions being displayed, you make sure you don't make silly mistakes and go for the right start/BO.

I'd like to see what others think.

revealing all armies at start could be kinda cool, you could still spy to see starts anyways

No way, random is meant to be surprise or forcing enemy to spy.

An option to turn this on would be handy

This is a topic I never bothered to explore so I don't know how far it can be exploited. I do know that certain people managed to pull out certain relevant pieces of information by reading the RAM in real-time while the game was running though. Reading encrypted files saved on a hard disk may be improbable, but reading raw memory is certainly doable and since the machine works with the raw data in order to perform its work you can gain a picture of what was originally in the files. Even passwords have been determined by hackers by accessing the memory of the machine processing the password. From what I recall this is how the OpenSSL issues arose - people accessing random pieces of memory potentially containing confidential information.


This will never gain enough community support to go through I think. I hate Random too and I especially hate it when players who don't understand that the RNG is broken try to tell me not to pick factions. But it's not enough to know the faction. If you're Men vs. Elves for example, you're still clueless whether your opponent is going for a Green Pasture or a Barracks start, and getting it wrong and making the wrong choice will cost you dearly.