Random Generator Bug?

[Snoil The Mighty]

I've done some of my own research on this. Not because I suspected a bug, just like to fiddle sometimes. I didn't find anything out in the 3 standard deviation range, in fact nothing anomalous at all. So the query is, what's the bug? I looked for a thread that described bug in detail and a lot of threads came up but didn't find anything descriptive. Anyone out there to point me in the right direction or describe what the issue seems to be?

[Skott]

What exactly is it suppose to generate? Or is that what you are trying to determine?

[Snoil The Mighty]

What exactly is it suppose to generate? Or is that what you are trying to determine?

Well there's a few threads where it has been cited that the random number generator is bugged somehow. However, that's all I ever get to read, I haven't seen a description of exactly how it's bugged. Basically, it's just is supposed to generate random numbers that solve probability problems that we see as action. IE-Your spy has 44% chance to get into a castle. You send him to the target there is then a random number generated to solve the probability issue. Depending on what the number is-success or failure is the result.

[Neoncat]

Best example about this one is those 'I have only girls' or 'I get only boys'. Don't know what the percentage is for getting girl or boy, but when you get 10 boys in a row and percentage is 50% you have just beaten up statistics.(~0.097% chance to get 10 boys in a row.)

Example -> http://cs.joensuu.fi/~ppakar/seka/mtw2_006.jpg
(~0.3% chance to get 8 girls with 50%. If recall correctly it got even worse, but I haven't taken screenshot about it. This happens way too easily somehow.)

When testing this bug you shouldn't do every action in a row, but you should do a action then do something else like you do in real game and then again try that action and so on.

Skott:
Random generator affects whole game. Nearly everything you do is calculated with random generator somehow.

[Durallan]

people are complaining that they fail missions with 95%, but its only 95% not 100% theres still a chance that the mission can fail.

[Alcorr]

Yes I have noticed this. In fact I KNOW its bugged. Either that or im astronomically unlucky. Just 10 mins ago I saw a heretic surrounded by 11 priests. Every priest had a 44% chance of executing him. I used all 11. The heretic lived.

I have seen many cases of this to varying degrees.

Do I need to say any more?

[Laconic]

You'd probably be hard-pressed to put a finger on any bug there might be unless you could run a series of pure numerical tests using the generator itself, not the game. I don't know how you'd go about doing that. I just take it in stride, my "luck" with the odds runs in streaks; I once assassinated three popes in a row, all with a 40% chance to succeed, and I've also missed two different 95% chance targets on the same turn (and one assassin was even killed in the process, getting me in hot water with an "ally").

[[GG]vonDöbeln]

I tend to agree the random generator is bugged somehow. No matter what my chance is of success (wether it's aquiring foreign merchant's assets, assasination or removing heretics) I fail my first attempt 90% of the time, even with 80%+ chance. After 5-10 attempts with 50% chance you'd think at least one attempt would succeed, but all too often it doesn't. It can get quite annoying.

[pat the magnificent]

personally, i just thought it was kind of anti-exploiting mechanism.

that is to say it only calculates the chance for something happening once, so if you fail and the reload to try again, the game repeats the same result as before.

like one of the previous posters stated, if you go and do something else after you reload to try again its easier to get a different result, presumably because the game has cleared out some memory.


(i'm not sure i'm explaining what i mean to say very well, somebody else clarify if you understand what i'm trying to say)

of course that doesn't explain your family members having only child of a single sex. (though i might speculate that the game is simulating genetic variations, IE some couples just don't seem to be able crank out babies of a particular sex.)

[JCoyote]

It's not a counter exploit, because there are ways of exploiting it. If you work it just right, you can use it to assassinate or accomplish 5% feats every time.

[pat the magnificent]

maybe its just not a very good one.

[seneschal.the]

Agree with Neoncat. The random seed can sometimes get stuck.. I had a general get something like 3 points of arse in 3 turns (4% chance each turn to increase 1 point) and I've had generals whose drink never went beyond the first level for 40 years. (4%).

And about the "anti-exploit", as it is very easy to avoid it. If you try to assassinate someone with a 5% chance and fail, you will always fail in that situation no matter how many times you reload. But, you reload and then do another action. Diplomacy, battle, spying etc. Then the random % for the assassination is assigned a new value, which might just be a successful one and you will kill that 10 star enemy king. If not, rinse and repeat until the random assigned value of your assassination passes the skill-test.

Civilization 4 uses the same system. If you save before attacking an enemy unit with your own, it will always be the same battle result if you reload. Because the random seed is not randomized every time, to make things smoother the CPU already knows what will happen (every action for every one of your agents on the campaign map in M2TW is already determined if it will success or fail.)

In Civ4 Ffiraxis hadded an option in the patch that allowed players to have the game randomize the seed after each reload. Meaning if your supertank got killed by an enemy longbowman, you could reload and it would get a new chance.

that said, somehow seeds seems to "get stuck" in M2TW.

It sure brings a new perspective to the game for some, learning that the outcome of their assassin mission is already determined but they will have to try to see the result. :)

[Snoil The Mighty]

I havent been able to replicate anything buggy. The controlled experiment I tried was done using 3 spies and 3 merchants. I didn't use any assassins or other agents. I used 50 reps each agent. Every reload, I used two other agents before using the test agent to insure I was using a fresh generator. This took quite a while and got pretty boring but here is what happened.

Merch1- Mission success % was 95%. Actual result 45/50= 90%.
Merch2- Mission success % was 44%. Actual result 27/50= 54%
Merch3- Mission success % was 13%. Actual result 5/50 = 10%

Spy1- Mission success % was 93%. Actual result 43/50. = 86%
Spy2- Mission success % was 79%. Actual result 32/50. = 64%
Spy3- Mission success % was 50%. Actual result 26/50. = 52%

The biggest deviation was 15% for Spy2. His mission was to spy on a nearby assassin. I'm not sure if stealth units are supposed to add any 'unpredictability' to the displayed mission % or not but even so, these results are clearly within an expected range in the fat part of the bell curve.

So far as I can see, there is no bug in the random generator.

[Neoncat]

Snoil The Mighty:

When you do those things after another it will look like you get right results, because thats how random generators with seeds are supposed to work. For example when you use 100 next seeds to test something you will get nice results, but when you use 1 seed now and wait for few turns use another seed after waiting and repeat this you can get really strange results. This is because random generators in programs aren't really random for now. If you would have totally random generator, this type of random generator might work, but for now it just can't work. Between those seeds can happen everything that can affect to next seed. I got 70% with 50 tries with 95% success rate thats just too low result not to mention that family tree.

There were a thread on official forums about this bug, but looks like its nailed or something. One another programmer agreed too that random generator in mtw2 have some probs because of 'static' seeds. He preferred to take seed from time every time, because thats what you usually do in games.

So play like you usually play and do lots of stuff between tries and check results.

[Zenicetus]

people are complaining that they fail missions with 95%, but its only 95% not 100% theres still a chance that the mission can fail.

This isn't necessarily a problem with the random number generator (although that could still be involved). As far as I can tell, the success estimate doesn't factor in the effect of enemy spies, or ancillary traits. I've had a top-ranked assassin fail every time while showing 90%+ success, only to find out later that the city had four enemy spies in it, or a family member had several of the anti-assassin ancillaries (guard dogs, or whatever). This isn't reflected in the success estimate. It does make the game a bit more unpredictable, so it could be an intentional design instead of a bug... it's hard to tell without a comment by CA.

And there could still be problems with the generator getting "stuck" or otherwise misbehaving, underneath all that.

[Snoil The Mighty]

In the next couple of days I'll just play through 5 turns normally, I'll use every agent I can that entails a mission success expectation of under 100%. I'll catalog the results here. Hopefully will be able to do it tonight but no guarantee on that, what I come with i will post here.

Just as an aside, IRL, I had 5 nieces before my for my first nephew was born, and back in school a friend of mine was the 5th of 7 brothers-no sisters. Unusual chains of events can and do happen.

[Snoil The Mighty]

Ok, started a new campaign as England since they start with several agent types and I could get to tracking results straight. I am starting turn 33. I have left the game loaded and running for over 48 hours straight, have not reloaded once. I also currently have several spots where I have recorded results from initial trials, so I can go back and try them again, just to check. The first crusade was just called so I thought I'd chime back in on how this is proceeding.

First things first, until I get something-anything-that can show me there's a bug specific to the random solution generator I have to assume it works and that's what my numbers are telling me. I can only track events that have specific success-failure chances so Dip actions are out of the calculations. Princesses do get some %chance missions on the marriage thing so I had a few of those mixed in. Mostly it's been spies of course., but merhcants have their fair share of incidents as well. A few priest vs. heretic actions. Only got assassins just recently and a few of them mixed in as well. The total number of incidents I have results for is now at 134 and counting. However, as a side note, 32 of the missions were 100% success rate (32/32 on those oddly enough) so it terms incidents with a stated chance of failure, it's at 102. I am using unmodded, unedited version purchased the day of release, patched to 1.1 version.

Current assessment- there is no bug that I have seen at this point but 102 incidents to track is not enough to warrant any conclusions. I have kept track of certain sets of numbers like the 93% and above missions. Have had 22 missions so far in this range (mostly new spies on a first mission @93% of course but there was a 95% merch mission and 2 other incidents higher than that but less than 100) and have failed 2 times. Both were spy missions. Kept track of what I call the coin-toss range too (Anything showing as 45-55%) where I am 8/19, a little low but certainly not out of order.

I am not calling anyone out, but I can report only what I see and I have yet to see a definitive piece of evidence that can reasonably claim to be a smoking gun, i.e.- something that says "See this result? This conclusively determines the random solution mechanics are bugged." In fact everything I have recorded so far indicates it is working perfectly and as advertised. I will continue to post results from this campaign as it continues since 102 is not enough of a sample but so far.... Adam and Jamie are saying 'This myth is busted'.

[Carl]

@Snoil: If I understand the descriptions of the bug correctly, the bug simply won't show up in the kind of testing you do. The bug boils down to this: every time the game calculates the chance of somthing happenning it fires a new seed. This means that if you do a save and reload the same result will come up every time.

Whats happenning however is this:

You use a spy with a 93% chance of sucsess. He fails. Normally it should switch to a new seed. However a lot of the time it apparently stays stuck on the same seed, so no mater how many spies with the 93% chance are tried, the seed ensure they all fail.

Your testing means your doing diffrent things between failiures and thus it will never produce one of the situations in which it gets stuck. What you need to do is produce about 30-40 spies, find a target they all have a 93% chance sucsess rate against, then try one. If it sucseeeds, do somthing else elswhere, then try with another spy until you fail. When you fail, immidietly try every other spy with a 93% chance without doing everything else.

If the bug happens they should ALL fail on you.

It must be the same target they are used against BTW.

[Snoil The Mighty]

@Snoil: If I understand the descriptions of the bug correctly, the bug simply won't show up in the kind of testing you do. The bug boils down to this: every time the game calculates the chance of somthing happenning it fires a new seed. This means that if you do a save and reload the same result will come up every time.

Whats happenning however is this:

You use a spy with a 93% chance of sucsess. He fails. Normally it should switch to a new seed. However a lot of the time it apparently stays stuck on the same seed, so no mater how many spies with the 93% chance are tried, the seed ensure they all fail.

Your testing means your doing diffrent things between failiures and thus it will never produce one of the situations in which it gets stuck. What you need to do is produce about 30-40 spies, find a target they all have a 93% chance sucsess rate against, then try one. If it sucseeeds, do somthing else elswhere, then try with another spy until you fail. When you fail, immidietly try every other spy with a 93% chance without doing everything else.

If the bug happens they should ALL fail on you.

It must be the same target they are used against BTW.

Ok, I'll load up on new spies and try at the 93% level but that's going to take me a few turns to crank out a decent number, and my weekend just ended. Will post results here. And just to be sure, I'll figure out a way to test it at least 10 times though this may take a while.

[Carl]

Yeah, just be aware that if you use a save and reload you'll probably get the same results regardless as I think thats the prupose of seeeds.

[seneschal.the]

It seems it creates a new seed after you reload, but only once. So if you fail the first, reload and then fail again, you will always fail no matter how many times you reload. I've onlt tried this up to like 10 reloads, but...

[Snoil The Mighty]

Well as the reload from a savegame is well known I am not testing that as it is not a bug, Civ battles played that way too- once the random result is calc'ed it saves with the savegame.

I did get a chance to see if it got stuck on my test game and was unable to replicate. Had a group of 7 noob spies with 93% chance to get into Paris (thing I hate most about this test is the spies train up and I lose my batches! Grr!) but the first guy failed, THEN the second guy failed as well! However third and subsequent spies all made it and with the odds of two spies @ 7% failure hitting twice in a row at basically a 1-in-200 chance this is going to happen sometimes. So once back in the Imperial chair later on tonight, it's back to testing........

[ransompendragon]

Agree with Neoncat. The random seed can sometimes get stuck.. I had a general get something like 3 points of arse in 3 turns (4% chance each turn to increase 1 point) and I've had generals whose drink never went beyond the first level for 40 years. (4%).

And about the "anti-exploit", as it is very easy to avoid it. If you try to assassinate someone with a 5% chance and fail, you will always fail in that situation no matter how many times you reload. But, you reload and then do another action. Diplomacy, battle, spying etc. Then the random % for the assassination is assigned a new value, which might just be a successful one and you will kill that 10 star enemy king. If not, rinse and repeat until the random assigned value of your assassination passes the skill-test.

Civilization 4 uses the same system. If you save before attacking an enemy unit with your own, it will always be the same battle result if you reload. Because the random seed is not randomized every time, to make things smoother the CPU already knows what will happen (every action for every one of your agents on the campaign map in M2TW is already determined if it will success or fail.)

In Civ4 Ffiraxis hadded an option in the patch that allowed players to have the game randomize the seed after each reload. Meaning if your supertank got killed by an enemy longbowman, you could reload and it would get a new chance.

that said, somehow seeds seems to "get stuck" in M2TW.

It sure brings a new perspective to the game for some, learning that the outcome of their assassin mission is already determined but they will have to try to see the result. :)

This is my first post. Yay.
This seems to be my experience. I can save/reload and reload and reload...but the outcome is already determined. However, if I quit completely and reload then I almost always succeed.

I applaud the people who have been doing multiple trials (>10) to test the probablilities. My trials of ten trials showed the actual probabilities to be much lower than what was displayed. Having read this thread now I think I understand what it is all about.

Glad to see I am not the only "chronic reloader."

cheers.
ransom