Simulationcraft fake results every time i use 'copy' line?

[ztn]

I see this for years.
1. take any char you want.
2. sim it and receive XXX result
3. add a line

copy=test

4. sim again and receive YYY number that lower then XXX by several K dps...

every time you use copy line to compare different items you receive another results then if you just change this item and don't use copy line at all... and please don't try to talk here about natural spread of results... you can simulate char 100 times and every time it will be XXX(+-200dps) and then add 'copy' line without changing sinle item and receive 100 time (XXX-3k)(+-200dps)...

if they don't fix this for YEARS then may be it's me don't understand some silly thing that obvious for everybody except me?

[DeadSwordz]

How many iterations are you doing because I just tried this with 25000 and got 196 dps difference the first time and 132 the second time

[Kuni Zyrekai]

Not sure what you've got set up on your end, but working as expected?

[Lothrik]

Do you not understand sample sizes or error margins? With 10,000 iterations my character has a 636.2 / 0.093% DPS error margin, which fits your description exactly. If you want a smaller error margin, increase the number of iterations in SimC.

https://en.wikipedia.org/wiki/Margin_of_error

[Puri]

I just tried it to disprove the OP, but I can confirm his findings.
If I sim without a copy line (my example:

head=mask_of_second_sight,id=138378,bonus_id=3514/1472/1813
neck=belerons_choker_of_misery,id=140899,bonus_id=3516/1487/1813,enchant=mark_of_the_claw
shoulders=shoulderguards_of_second_sight,id=138380,bonus_id=3514/1482/3336
back=cape_of_second_sight,id=138375,bonus_id=3514/1808/1477/3336,gems=150crit,enchant=200agi
chest=tunic_of_second_sight,id=138376,bonus_id=3516/1487/1813
wrists=wellflattened_wristguards,id=140857,bonus_id=3516/1492/3336
hands=gloves_of_synchronus_elements,id=140905,bonus_id=3516/1808/1487/1813,gems=150crit
waist=cake_carriers_girdle,id=140858,bonus_id=3444/1482/1813
legs=biornskin_leggings,id=134194,bonus_id=3510/1547/3337
feet=bleak_underworld_treads,id=137324,bonus_id=3536/1552/3337
finger1=spellblades_gemmed_signet,id=140895,bonus_id=3514/1808/1487/3337,gems=150crit,enchant=200crit
finger2=sephuzs_secret,id=132452,bonus_id=3529/3530,gems=150crit,enchant=200crit
trinket1=kiljaedens_burning_wish,id=144259,bonus_id=3529/3530
trinket2=convergence_of_fates,id=140806
main_hand=twinblades_of_the_deceiver,id=127829,bonus_id=719,gem_id=143525/133684/140825/0,relic_id=3454:1487:3337/3510:1542:3337/3516:1487:1813/0
off_hand=twinblades_of_the_deceiver,id=127830

I will reproducably get results between 5915000-5918000, so everythings fine and in the margins.

If I add

copy="Test_Con" trinket2=convergence_of_fates,id=140806

The result with the original trinket is about 1k less than before.
This is reproducable, it will never get as high as when I sim the trinket alone.

Result as documentation: http://imgur.com/a/y6YdP

[Lothrik]

Do a fixed time simulation.

Adding additional units is likely compressing the duration of specific burst windows (ie: "my class does 20% more damage to targets above 90% health" or "my class does 50% more damage to targets below 35% health") meaning your simulated character is getting (just as an example) a shortened execute phase.

Fixed time simulations can prevent this as it will force enemy health percentage distribution to be uniform. This means it doesn't matter how many actors are present, if your execute phase was 45 seconds long solo it will still be 45 seconds long even with 500 actors present in the simulation.

Keep in mind this can inflate the value of mechanics that increase damage dealt during specific windows (ie: you may get some pretty wildly unrealistic results if you do this as an execute-heavy class and the simulated execute phase is significantly longer than your "real" execute phase). If all you're doing is comparing trinkets, this shouldn't matter a ton.. but then again, a 1K DPS difference shouldn't matter either.

If you're really worrying over a 1K DPS inconsistency with players doing 700K+ in Nighthold when comparing trinkets, you're splitting hairs for no reason and probably already know which trinket is better without SimC spelling it out for you.

max_time=300
vary_combat_length=0.0
fixed_time=1

Edit: You may also be interested in the single_actor_batch=1 option, which simulates each character individually, then combines them at the very end. This was added in 7.1.0, release 3 as a beta feature.

[Puri]

Adding

single_actor_batch=1

is indeed resolving the issue.

[ztn]

it's interesting - even i already set vary lenght option to 0


both fixed_time=1 and single_actor_batch=1 options solve my issue. thank you! with fixed_time=1 i have another results like +20k dps but it's exactly same with or without copy line.

p.s. fixed_time=1 is not usable at all for smth real if your char depends on boss hp.
exactly same options: leather belt leg(+30% damage if target have >90% hp)
1. with fixed_time=1
Symbiote Strike 47140 6.2%
2. w/o
Symbiote Strike 16946 2.3%

so single_actor_batch=1 is the only real fix

[Deleted]

That was extremely useful, thanks for sharing