Damaged artillery bug: major

  • Scenario:

    - Single artillery is attacked directly and loses x% health (from 10-90)
    - Artillery survives attack, and is added to nearby stack of, for example, 20 artillery units
    - Even if artillery health is at 90% AND 100 GM is used to heal artillery, the entire stack gives a fraction of the damage (30hp per round -> less than 5hp per round)

    I'm on Petruz's discord server and showed SS of events on there. Users such as GB were saying it was a commonly known bug (I couldn't find a similar thread/report despite searching) and the only solution is to kill the damaged artillery. At that point it was already in a stack of 30 and very difficult to single out, especially in heat of war.

    This bug had a major impact on the map in question, and if it's as common as S1914 discord server users were saying, I would imagine a lot of people have lost maps because of this.

    Alliance: March of the Wolves
    Role: Officer

    Wins:

    Gold Rush (500p): 1 (coalition)
    World in Flames (100p): 1 (coalition)
    Team up (100p): 1
    Supremacy 1914 (31p): 1 (coalition)
    [Tutorial] The Great War (31p) 1 (coalition)

  • I don't think this is actually what happens. In reality there's a random number generator and that affects the outcome not the single damaged artillery. I think there is a superstition that some guns can not hit at all but I've taken 10 guns in a stack that were missing repeatedly separated them all and sent them to fire at infantry to make sure each and every one of them was able to accumulate more kills.

    It's very hard to collect enough data or I would probably reverse engineer the formula (or some sort of close approximation) for how the random number generator impacts damage here but casual observation tells me that the damage in 1914 never exceeds the damage potential. My best guess is it's capped at somewhere around 80-90% of the potential. A stack of 20 artillery (in your example) has 30 potential damage and when you add the damaged artillery that potential goes up not down. the damaged unit won't add the full 1.5 so you might have 31 potential now instead of 30 (or 31.5 if everything was healed).

    What is actually happening is the RNG has a range and that range starts at 0. It's my perception that the random number is weighted towards the low end of the scale meaning more 0% than 80% (or 100% if there is no cap and I just don't pay close enough attention). The problem is as humans we have a biased filter that fixates a lot more on the times we got "unlucky" than the times we got "lucky". Even though I'm aware of this and I'm really good at math (800 on the math SAT's 8 times) I can not be certain I'm able to logically suppress that bias without actually collecting a bunch of data that is a bit painful to collect. This means the RNG might actually be linear and you're just dealing with a wide amount of variance because there's a huge difference between 31 damage and 0 damage.

    So one of the problems here is there is no way to detect the actual granularity on the random number being generated. What if I generate a random number between 0-1 multiple by 100 divide by 25 and round down. Now my granularity is in 4th's and 0% damage happens 25% of the time no matter how many guns are shooting at the target. In this case getting 0 damage twice in a row would be 1 in 16 and would happen many times per game. If you take a straight random number (0-1) and multiply by potential damage then you get 1% granularity. Given how often I see two zero's come up in a row I sort of suspect that the damage is weighted so we're not using a true random number (linear) but computed "random factor" that is weighted or at least has a granularity in chunks as big as a tenth of the range so that double zero would be happening 1 time in hundred.

    Back to the human brains bias. Get two 0's in a row and you start to think something is broken. Someone came up with the idea that some of the artillery is "bugged" but knowing a lot about php and programming I'm guessing that's not the case. You're just getting some low rolls which happens to everyone and if it's linear there's simply a ton of opportunity for variance.

    All that noise aside, there is something you can do to cut way down on variance. If your guns are not in danger (i.e. taking fire) break them into smaller stacks and shoot at the target multiple times. it's easy for 20 guns to roll a zero and miss that one infantry but break them out into 10 stacks of 2 and that infantry is going to die almost every time because you're not going to roll 10 zero's in a row. You get a lot more "average" results and almost never see the extremes. Roll 11 die 10 - 10 and compare those numbers to a single die 100. One is a bell curve massively weighted towards the middle and the other is a line with a roughly equal distribution and a large amount of variance. Variance is why even if you make all the right moves in poker you still lose a percentage of the time. That's why pro poker players will "run it twice" or let other players bankroll part of their stake in a tournament, to cut down on variance.

  • In all honesty, I completely agree with everything you're saying, however, I have finally come to a very simple conclusion over the years of working with damage dealt through bombardment and bombing.


    That conclusion is at least 40 to 50 percent and sometimes as many as 100 percent of my artillerymen and my bomb loaders are loading marshmallows instead of regular armaments. In fact, I KNOW that some of my bombers loaded up Graham crackers while others loaded up chocolate squares because they thought the enemy might enjoy some Smores. (I write it off to at least half my army are pacifists.) SIGH ;(;(