test content
Logo
What is the Arc Client?
Install Arc
Options
A statistician's view on current RNG
minotaur2857
Member, NW M9 Playtest Posts: 1,141 Arc User
I've thought there was something mildly wrong with this game's RNG for a long time. Every double RP I kept stats, and found I was consistently using 20-30% more pres wards than I should. Then after a while, I found I was getting about what I should so I stopped keeping those stats. I have a statistics degree so tend to look at things like this.
Recent events have told me something has gone drastically wrong.
Examples: I kept stats of everything I crafted to make new mastercrafting stuff (everything at 75% success chance).
I'm 146 successes from 219 attempts so exactly 2/3. So how far out of kilter is that ? Well since we have more than 40 attempts, we can use an approximation with some useful properties. 219 attempts x 0.75 = 164.25 is the expected number of successes, so we're 18.25 below that. The most useful bit of this approximation is that 95% of results should be within 2 standard deviations of the mean, and 99% within 3. The standard deviation is the sqrt(np(1-p) where n is the number of attempts and p is the probability of success so this is sqrt (41.06) = 6.4. So we are well outside 95% of the results and almost outside 99% (one more failure would have been outside).
I have tried to make a number of finished products in the last few days, only 1 worked first time. (and this is compounded by the new mastercrafting being INCREDIBLY mean on what you get back from a fail)
I had to go out and get some more effervescent water due to the fails so I ran some cold run explorers charts. I didn't see anybody else doing them when I was to muck up the RNG. Past experience suggested I was getting maybe 40-43% Eff water, 40-43% native iron, 15-20% raw rubellite, this time I got (60 nodes) 17 Eff water, 4 raw rubellite and a massive 39 native iron with 2 or 3 runs of all 6 nodes giving iron from a chart.
Suggestion: allow people the option of using a number of charts at a time, to get ingredients in their expected proportions, so 5 charts will give you 2 water/iron and 1 rubellite x your multiplier from each node if my subjective numbers are about right. This would also relieve the boredom of running a LOT of charts and put the people sitting by nodes to try to second guess the drops out of business.
Then the final insult was I tried to make some R13 bondings:
I went in with 2xR11, 3xR12 bondings and 365 preservation wards, I should have needed just over 200 pres wards on average.
What happened ? well the 2 5% chances took 143 and 42 attempts, and the 3%s the first took 85 and I still haven't got the second to go after 98 unsuccessful attempts.
Something I noticed without keeping numbers in the past is that when the servers are laggy (and they are at the moment), success rates plummet. My strong suspicion is that the RNG is not generating new values some of the time when lag bites.
An old Cryptic game used to display the RNG result in the logs if you enabled that option, would be very useful in tracking down if this was happening.
Recent events have told me something has gone drastically wrong.
Examples: I kept stats of everything I crafted to make new mastercrafting stuff (everything at 75% success chance).
I'm 146 successes from 219 attempts so exactly 2/3. So how far out of kilter is that ? Well since we have more than 40 attempts, we can use an approximation with some useful properties. 219 attempts x 0.75 = 164.25 is the expected number of successes, so we're 18.25 below that. The most useful bit of this approximation is that 95% of results should be within 2 standard deviations of the mean, and 99% within 3. The standard deviation is the sqrt(np(1-p) where n is the number of attempts and p is the probability of success so this is sqrt (41.06) = 6.4. So we are well outside 95% of the results and almost outside 99% (one more failure would have been outside).
I have tried to make a number of finished products in the last few days, only 1 worked first time. (and this is compounded by the new mastercrafting being INCREDIBLY mean on what you get back from a fail)
I had to go out and get some more effervescent water due to the fails so I ran some cold run explorers charts. I didn't see anybody else doing them when I was to muck up the RNG. Past experience suggested I was getting maybe 40-43% Eff water, 40-43% native iron, 15-20% raw rubellite, this time I got (60 nodes) 17 Eff water, 4 raw rubellite and a massive 39 native iron with 2 or 3 runs of all 6 nodes giving iron from a chart.
Suggestion: allow people the option of using a number of charts at a time, to get ingredients in their expected proportions, so 5 charts will give you 2 water/iron and 1 rubellite x your multiplier from each node if my subjective numbers are about right. This would also relieve the boredom of running a LOT of charts and put the people sitting by nodes to try to second guess the drops out of business.
Then the final insult was I tried to make some R13 bondings:
I went in with 2xR11, 3xR12 bondings and 365 preservation wards, I should have needed just over 200 pres wards on average.
What happened ? well the 2 5% chances took 143 and 42 attempts, and the 3%s the first took 85 and I still haven't got the second to go after 98 unsuccessful attempts.
Something I noticed without keeping numbers in the past is that when the servers are laggy (and they are at the moment), success rates plummet. My strong suspicion is that the RNG is not generating new values some of the time when lag bites.
An old Cryptic game used to display the RNG result in the logs if you enabled that option, would be very useful in tracking down if this was happening.
4
Comments
Over the two years I have been playing I have noticed this anecdotally a few times. During those periods I just stop refining and wait a month or two before going back to that.
I'm trying to make a lichstone sash (c1.5M on the AH), all tasks at 75%
Ingredients:
Lichstone enamel
2xobsidian shard
4xChultan silk
1xJute macrame
To make the lichstone enamel costs 1 lichstone, 4xartisan's enamel this costs c500K per attempt and 500 GMs, I failed once and succeeded once so got 1 and 3.
To make obsidian shard 12 obsidian 1 red rouge, makes 4, I got 4
To make jute macrame costs 15 samarachan jute, makes 3, I got 3
To make the chultan silk costs 4 chultan silk thread, 1 potash, 4 spider silk, success gives 2CS, fail gives 2CST
To make chultan silk thread costs 12 silkworm coccoon, 6 effervescent water success gives 4CST, fail gives 4SC
and this is where the problem really arises:
I went 4/7 on the CST so got 16 CST for 72 net SC and 42 effervescent water
I then went 2/4 on the CS and got 4CS and 4CST from that
Then I failed the final craft, getting back 2 CS
So then I used my 4 spare CST to try to make CS and that failed.
So my expenses are (at AH prices, I actually paid guild marks for charts, but had to buy things like the fartouched):
1M for the lichstone enamel, I have 3 left
140K for the jute, I have 2 left
38K for the obsidian, I have 2 left
830K for the chultan silk, I have 2 chultan silk and 2 chultan silk thread left
So basically I'm not that much closer to making the belt and am already down more than it's worth.
The problem is not getting anywhere close to 75% on the tasks, the small ratio of how many ingredients are used to how many you get of the crafted product and the incredible meanness of what's returned on a fail.
I could live with it if you were getting say 4 chultan silk thread for 6-8 resources rather than 18 or you were getting 2/3 of them back on a fail.
Compound this with that for me I'm only getting 1/3 to 1/4 effervescent water from cold run maps when before this module it was around 40%, it's been a very expensive exercise.
@nitocris83 also with the grotesque mishandling of the whole MC4 bug, really mastercrafters should be being given some sort of break/compensation for this. Of course the items like relic iron ingot went up from 20K-135K each on the AH when it was fixed (they're down to 55 or 60K now) as everybody scrabbled to finish their mastercrafting.