98% chance to do something!!

I don't know if everyone has this happen (I wont call it a problem) to them....but often when I am crafting or skinning or mining I fail and sometimes 2 or 3 times in a row!! That is despite the fact that I have a 98-93% of success. I think they need to revisit what's going on here. I am not going to say its a problem or an issue but it just doesn't seem right. Somethings wrong I think but I can't tell you what.

 

This sounds familiar. The maths experts here (I am not one of them!) have pointed out that a player will have to perform millions of crafting attempts in order for the overall success percentage to average 98%. With each individual crafting attempt the dice are rolled anew, so it is quite possible for you to have several failed attempts in a row even at success rate of 99.99%.

 

If you have a 98% to fail, it's virtually guaranteed that you will see failure streaks given enough attempts

 

there was a bug that if your crafting tool breaks during a bulk build, it will continue doing the que and failing till que ends.
And currently the durability of the tool doesnt update on the crafting screen.
Something that may have happend?

 

Math time!

Let's take 95% success as an example => 5% failure. It means that at least 1 of every 20 attempts will fail. Now, you saw it 3 times in a row (I saw it numerous times). It means that the probability of it is 5% * 5% * 5% = 1/20 * 1/20 * 1/20 = 1/8000 --- you will likely to see 3 failures in a row once per 8000 attempts.

What does it mean? Imagine that you make ingots at 95% success rate. If you make 100 ingots per every single day for a year, you will see 3 failures in a row about 365/80 = ~4.5 times a year. Since the persistence I logged ~420 hours. I already saw 3 failures in a row multiple times. Hence, I think there is something fishy about RNG. Hence, my posts.

PS: All numbers assume that the reported probability is accurate, and when SotA tells us 98%, it is 98%, not 70%. If numbers are not accurate --- it is a reporting bug. RNG can be fine in this case.

 

You're only counting your "instances", in reality the RNG isn't used for just your instances. It's used for everyone's instances. According to a Blizzard poster a long-long time ago when looking at RNG results you need to look at everything the RNG calculates not just your own anecdote. The complaint at the time was that someone rolled a 1 out of 100 5 times in a row and stated that meant the RNG was bugged. The developer responded that because the RNG isn't bugged it's virtually guaranteed that streaks like this will be seen by players.

 

It doesn't matter. Any subset of a fair random output is still fair, and exhibits the same randomness, provided it is not guided by the output itself.

Example of a guided selection: all positive outcomes are assigned to me, and all negative ones to people I don't like.

Example of an unguided selection: all odd examples are assigned to me, and all even ones are assigned to somebody else.

In the latter case, no matter what impartial rule we use, both sequences would have the same fairness, and same distribution, as the original one.

 

This is the way I look at it in a simple way. I may be wrong but I believe they are doing a random number between 1 and 1000 in order to account for a 93.33% chance to succeed. So if you rolled a 1000 sided dice 9333 and below is success and 9334 to 1000 is fail. So the chance of you rolling a fail and streaking a fail can and will happen, a lot.

 

it's a lot of time that maths of SOTA seems strange (1 or 2 years ago.. .things are much more strange) .... on skills for example a 77% i fizzle 3 times the spells when casting... 3 times? i understand that random is random... but why there are a lot of fizzle+Fizzle+fizzle+cast ? i can understand that a 50% can be something like Fizzle+cast or Fizzle+fizzle+cast ... but a 77% ... not.
probably there are other factors to add to these %.... no idea...

 

That "fizzle-fizzle-fizzle-..." is the worst offender for me, at least perceptually. When it starts, it can go into double digits for me, as if RNG is "stuck" somehow. My second offender is mining: it shows >90% of success, yet if it fails, it frequently fails in pairs. My third offender is making an exceptional item. With >20% every 5th item should be exceptional, yet my current streak is about 20 with no exceptional item across 2-3 weeks --- multiply the failure rate of 0.8 20 times to see the real probability of that (~1%). It is not supposed to happen, yet here we are.

 

Try a DnD based game with visible rolls. Even try Yhatzee. I've rolled three Yhatzees in a row once. Physics' RNG get stuck IRL?

 

I have the same problem that it has three or four times fizzle in a row for magic spells that have over 90% successful rate. And this situation is not rare. Also, many other cases are that the % successful rate and the things actually happened just doesn't right at all when I just use my common sense. I think something must be wrong behind the scene.

 

The problem with comparisons like this is that the OP said this happens "often", and this triple yahtzee happened "once".

If we get a once-off jackpot where we fail a 98% chance craft 3 or 4 or 5 times in a row, that's fine.

If we fail a 98% chance craft 3 or 4 or 5 times in a row every few times we attempt that craft, something is wrong.

 

Real math time!

What you calculated is the probability that out of 3 trials, all three will be failures, and then you calculated the expected number of such a result out of 8000 sets of 3 trials evaluated independently. So for example, the sequence SFFFSS obviously has 3 failures in a row but isn't counted in your model since it's broken into two sets (SFF, FSS) that are then evaluated independently.

The actual thing you need to calculate is that out of several hundred(or thousand) trials, what is the probability of seeing a run of three failures. It turns out this is really really complicated, and I can't be bothered, but feel free to do it yourself and let me know what you get.

 

Of course, you can't be bothered! (I decided to skip "real real math time!" --- anything with "real" in it sounds tacky, like "Real Housewives of SomeOtherPlaceWoodVilleState").

In my model I don't count any predecessors nor successors, nor sets (why you brought up sets? 3 trials per set to detect 3 identical results??? what does it even mean? why not 1 trial per set to detect 3-in-a-row???).

In my model just N in a row, whenever it starts --- I don't see how boundaries are relevant in this case. I don't cut them in trial runs, just treat as a never-ending succession of independent trials. I don't "forget/reset" previous trials after some arbitrary number, so boundaries do not affect it. As soon as I see a failure (or success) I count subsequent repeats --- what is wrong with this approach in your mind?

BTW, if I "lost" failures on boundaries, does it mean that N-in-a-row failures are even more frequent than I think?

Please note that I am not interested in "what’s the probability that in 50 coin tosses one has a streak of 20 heads?" (taken from your link) --- I don't think it is relevant in this case. Yes, with a long enough observable sequence some probability of rare occurrences will rise with the overall length. So what? We need to be practical. The real question is how it corresponds with what we see?

Example: in 1,000,000 some exotic sequence can be encountered twice. Yet it will not correspond to observations, if we had only 1,000 runs, and saw it twice (or more) already. "Yes, but the rest of 999,000 trials will be all kittens!" --- doesn't work here, because again it is a probability we can count again.

So, when you can be bothered, and insist on limited trial runs, take some reasonable number that corresponds to a realistic week, or a month, and calculate the probability of 3 failures in a row at 5% each. It is still be around 1/8000. Just to remind again: I assumed an year-worth run of ~36,500 trials (100 per every single day), and with a rough entry-level math I got an estimate ~4.5 times for 3-in-a-row 5% failure. With more accurate math it will be a different number of the same order of magnitude. I invite to count it, and shame me, if you get "once a week" or something like that.

Thanks in advance, Arlin!

 

i made a small app that records how many triple fizzle can be with my % .. and seems that i have to see a triple fizzle every run on deep ravenswood (25-30 min) ... the problem is that i see these triple fizzle 3 or 4 times ... on a Deep session... today .. i have see a QUADRUPLE FIZZLE on summing pet (69%)

 

Are you making each item one at a time or are you using the ability to make up to 20 items at a time? I was personally quite surprised when I stumbled across the section in the instructions that indicates you CANNOT crit when you are making multiples of an items. The only way to crit is if you make the items one at a time.

 

I make one at a time. I am familiar with this restriction, and don't mass-produce one-off items. But it is a good catch in general --- I didn't mention this important detail. Thank you.

 

If results are verifiable, I'd be willing to pay out 17:1 on (my) ingot failures at 1k gold per roll to anyone that wants the action.

 

Try Yhatzee. Try tabletop DnD. With thousands of players playing at once, you'll see a lot more unbelievable streaks than you seem to expect.
Rolling a 1 is less likely than rolling less than 2 but even that happens in streaks.