Edit: There is a second set of data that also includes tracking of streaks of rolls halfway down the post. It’s a smaller dataset, but unbelievably uniform to expectation. Tempering is not weighted, and no obvious behavior involving manipulation of streaks occurs regularly enough to appear in this sort of data.
This is preliminary data, limited to a single category on a single class.
Caveats:
Different categories could be weighted while this one isn’t
I could have misclicked occasionally - almost certainly not enough to change the data
Items could be somehow weighted INDIVIDUALLY, meaning that this isn’t a big enough data set to account for short stretch variability
Weighting could change on each particular roll relative to the prior roll
Weighting could be extremely low impact values
I rerolled 500 single hand melee weapons on Rogue using the cutthroat category, which has 4 options. If equally weighted, each option should have a 25% chance of rolling. I used all 6 rerolls on every item, for 3000 total rolls.
Total rolls: 3000
Vulnerability damage: 818 27.3%
Cutthroat Damage: 738 24.6%
Cutthroat Critical Strike Chance: 696 23.2%
Cutthroat Attack Speed: 748 24.9%
Over 3000 attempts, this is close enough to the expected values that there is almost no chance that values are weighted to any significant degree. It is possible that they could be weighted within the range of a couple of percent relative to each other, but there essentially no reason to do that (and given that vulnerability is the “desired” stat in this category, if their aim was to decrease the likelyhood of getting the desired option, this data skews in the opposite direction).
If you would like to repeat the experiment for yourself, this is the (very crude) AHK script I used to count rolls. Please note that it takes a LOT of rolls to overcome short streaks; if you only roll a couple of hundred times or less it proves extremely little. This is even more true on larger categories.
This script in no way interacts with D4; it is purely a script that counts the number of times you press the buttons 1,2,3,4,5 and expresses those counts as a percentage of total buttons pressed. If you cannot understand what it does by looking at the code below, and are not comfortable using it, don’t use it.
Edit: I have updated the script to write each roll to a logfile called temperslog.txt so you have access to the order of rolls if that is something you’re interested in.
#NoEnv
v1:=0
v2:=0
v3:=0
v4:=0
v5:=0
v_total:=0
1::
v1+=1
Iniwrite,1,temperlog.txt,Tempers
gosub,tally
return
2::
v2+=1
Iniwrite,2,temperlog.txt,Tempers
gosub,tally
return
3::
v3+=1
Iniwrite,3,temperlog.txt,Tempers
gosub,tally
return
4::
v4+=1
Iniwrite,4,temperlog.txt,Tempers
gosub,tally
return
5::
v5+=1
Iniwrite,5,temperlog.txt,Tempers
gosub,tally
return
tally:
v_total:=v1+v2+v3+v4+v5
v1t:=round(v1/v_total*100,1)
v2t:=round(v2/v_total*100,1)
v3t:=round(v3/v_total*100,1)
v4t:=round(v4/v_total*100,1)
v5t:=round(v5/v_total*100,1)
tooltip,Total: %v_total% || V1:%v1% %v1t%`% || V2:%v2% %v2t%`% || V3:%v3% %v3t%`% || V4:%v4% %v4t%`% || V5:%v5% %v5t%`%,1,1
return
3000 rolls, while not absolutely positive proof, should be enough to put doubt aside from all but the most tin-foil wearing of conspiracy theorists that there is almost no chance that any of the caveats above apply.
Here are the two reasons people perceive a potential weighting of categories where one almost certainly doesn’t exist:
-
You reroll every time you see a negative outcome, and never when you see a positive outcome. This means that you actually see every negative outcome on a given item, and you only see one positive outcome on every individual item, even if behind the scenes, the likelyhood of every outcome is the same.
-
Pure and simple, confirmation bias.
I realise that a great many people will say “of course the values aren’t weighted, why did you even bother doing this”, but to those people I say, go and read any of the many, many threads on tempering and you will see handfuls of people who are convinced on the matter.
Edit: This is a post from ~190 posts into the thread containing a second set of data on sorcerer frozen mastery category, including the raw data and streak counting. It’s only about 10% of the size of the first test, but frankly even at that point the data is so normal it would hard to be moreso. There’s nothing here; the bugs and behaviors people are ascribing to the system simply don’t exist in these data sets. Either they are confined to specific characters or accounts, or else they are so deeply malicious that they only occur in very specific circumstances to screw with very specific people at specific times (ie: they don’t, take off the tinfoil.
Anyway, this is the post:
Ok, another ~276 rolls in I am stopping, because there is absolutely zero indication that ANY form of bug or streak manipulation is occurring. If there were some form of indicator that there might be, I would probably be inclined to keep going, but there simply isn’t.
https://docs.google.com/spreadsheets/d/e/2PACX-1vQADvlALisXIuI1l0EF4FkAUwzTDyTQUD_f2SVmBoLZ3rB4Zpo_ffMo6j0VUDf9gnB034u6L69Nom9Y/pubhtml?gid=0&single=true
There is the spreadsheet with the raw data and the streak calculations. You can look at it yourself.
Here are the final numbers after 276 rolls (46 items, somewhat ironic given that guy’s claim he did 47 some time yesterday :P)
Total rolls 277 %rolls
1: 2x Frost Bolt 74 26.71
2: 2x Frozen Orb 70 25.27
3: 2x Icy Shards 67 24.19
4: Blizzard Size 65 23.47
Streak of 2 47
Streak of 3 11
Streak of 4 5
Streak of 5 2
Streak of 6 0
Streak of 2 on single item 46
Streak of 3 on single item 9
Streak of 4 on single item 4
Streak of 5 on single item 0
Streak of 6 on single item 0
TL:DR; on these 46 items, I rolled streaks of 2 of the same roll 46 times, 3 of the same roll 9 times, streaks of 4 4 times, and zero streaks of 5 or 6 of the same roll on the same item (or even across multiple items). These are all counting mutually exclusively (ie: a streak of 2 doesn’t get double counted when contained in a streak of 3). There’s simply nothing here. Could it be still bugged and I just got really lucky for 46 items worth of data? Sure. Is it likely? No. Not a single thing about this second data set stands out as untoward in any way given an unweighted distribution and no funky streak functions or bugs.
about tempering and making claims of weighted blah blah.