"all champions have the same drop rates"......(photo)
gannicus0830
Member Posts: 652 ★★★★
Iron patriot approves this message.
This Gwenpool event just adds more weight to what I have already suspected for a long time. Kabam insists all champs have identical drop rates, but does anyone actually expect them to tell us if they don't?
Just watch the patterns and use your head.
21
Comments
There is a certain 5% chance in that crystal. How dare they!
Do make a vid of that epic Rulk kill 😂
So how did I get this and another person got IMIW? Should I submit a ticket to get IMIW?
Just bad luck, my alliance got pretty lucky with them.
It gets annoying when people keep saying this like their conspiratorial beliefs are supposed to be extremely self evident. For the record, three people in my battlegroup pulled 5* IMIW (I wasn't one of them). That beats the two people in your alliance that pulled IP. If I use my head, that's statistical variation. If I jump to conclusions watching patterns, that's Kabam hating your alliance more than mine.
You mention polls. Although the polls aren't scientific because they are self-selected, for the record the current top occurrence in the forum poll for the cinematic 5* crystal is (as of the most recent time I checked) Stark Enhanced Spiderman, 42 out of 440. Iron Patriot is in second place with 41, and Wasp is in third place with 38 pulls. The lowest pull is Yellowjacket with 14, and Star Lord is second to last with 17 pulls.
And that’s just out of that small sample too. I imagine it would average out to 5% for each champion if we had a larger pool to analyze.
I would wager good money that the issue wouldn't be brought up if the pulls were shown as mostly IMIW.
It's the same rate for every Champ.
I can only assume by looking at your avatar that you didn’t learn and you’ve rolled the same again
Yeah, the calculator pic was pretty offensive, apparently.
As your roster progresses, what is an “improvement” to your roster becomes narrower and narrower. So no, drop rates are not varied, just your perception of what a “good” champ is.
Hey there @Frnkielo . Would you mind providing some proof to that claim? You seem pretty convinced that statement is real. I would love to see factual evidence from you since you seem to have figured out the system. I'll wait right here for you.
4* medusa (3 times)
5* Magik (once)
5* Void (once)
5* Starlord (once and it was to awaken him finally)
But I’ve also pulled
4* Daredevil Classic (once max dupe)
4* OG Captain America (once dupe)
5* Antman (once and second dupe)
6* beast (once and BOOOO)
4* IMIW (once and not duped)
4* Winter Soldier (awakened)
4* Ms Marvel (the blonde) (once and neither she nor her clone are duped)
6 great pulls to 6 meh/bad pulls. My personal pulls have kinda disproven the idea of unequal champ odds to me.
The problem is that people only want THE best champs currently forgetting that the amount of trash-ok champs outweigh those we claim as god tier by a lot. Most of the time I’ll get ok champs that are great in a vacuum but are simply outclassed (look at cosmic class and tell me there isn’t power creep and situationally better champs).
The odds of any specific class awakening gem being the same if you have a 5* blade you want to awaken and no other worthy 5*s in any other class than you only have a 1/6 chance of being satisfied. Amplify that to the growing roster in the game and you start to see why “rigged pulls” aren’t really a thing.
The only reason you’re upset at 3 back to back IP pulls is because it seems specific. But take a 20 sided die, roll it 3 times. Any sequence of numbers you roll has a one in 20^3 (or 20*20*20 or 8,000) chance to happen. The triple iron patriot is only noticeable because we assign value to that and relatively few other odds. I believe there is a philosophical area based around the misconceptions of gambling (or chance in general) that could better explain this topic than myself
I'm certain it would. The variability in the data is consistent with normal random variation and the inevitable oddities that result from self-selected small sample sizes. When the reddit has done expansive data collection to determine the drop rates of things like featured GMCs, the data never seems to show evidence of these statistical anomalies that humans seem to be able to find with just eyeballs and a couple of crystals, and their data generally converges on results consistent with the published odds.
A statistical anomaly a human being can just spot in a few pulls is such an incredibly large skew that it would be essentially *impossible* for any actual testing of drops rates to fail to find. Testing cannot prove that the drop rates are exactly what are claimed, but they have conclusively demonstrated that the skews people claim exist do not exist. Not "probably don't exist." Do not exist.
whilst overall they have an even chance.
the reason for this is that RNG is still a mathematical calculation. it is a series of calculation based off a random seeded number.
something like this -
seed * x * y + a / r * e * x = champ given
a lot (most) RNG programs generate the seed every milisecond or so therefore the seed is based off time.
2 crystals opened at the exact same split milisecond would infact be the exact same.
so given this it is possible that the seed number generated at 7:30am is quite similar to the seed generated at 7:32am resulting in a higher drop rate of similar champions.
the seed number generated at 15:42 however would be a much different result.
if this is the case it would be theoretically possible to know what champ would be given at any time of the day and perfect your opening to determine your result.
to do this however you would need to be exact with your timing to the exact split milisecond (impossible) know exactly how the seed number is generated, know the exact calculation the applied to said seed number, and the value that is assigned to each champion.
this is of course practically impossible to do without a lot of knowledge and some serious programming wizardry, however it does give an insight as to how it likely works.