Quote:
Originally Posted by DeathsSilkyMist
[You must be logged in to view images. Log in or Register.]
On longer parses, we start to see a more consistent drop in damage as we increase AC. We also see an increase in Minimum hits, and a decrease in maximum hits.
It looks like the softcap is in effect well past 217 worn AC. The issue is simply that the softcap returns are quite small, which is expected for Shamans and Rangers.
My previous data sets with 400 hits per set were noisy enough to mask the subtle decrease in damage. I am thinking that is what happened in the Ranger thread as well. I don't think they were generally parsing around 1000 hits per AC value.
It is easy to see why people accused Haynar of having a hardcap on AC at high levels. He said he was running 3 hour parses to test his changes, and most people aren't gathering that much data per AC value.
This does show that AC has large diminishing returns past the softcap for at least some classes, which is why people don't feel like it does much. In the data above, going from 207 worn AC to 386 worn AC reduces damage by about 6.5%. It may be a bit higher if I ran even longer parses, as we see the 386 worn AC parse was basically identical to the 300 worn AC parse damage-wise. Considering that the 386 worn AC parse had like half of the max hits and 12 more min hits, that may just be unlucky RNG.
|
Good analysis. Nice to see you appreciate how the noise in a 400-hit parse can make it hard to see the impact of small effects. The screenshot/spreadsheet I posted earlier has the number of hits I parsed in the ranger thread; you can see there's two high-quality parses plus a bunch of noisy couple-hundred parses.
Just like it's possible that the 300 and 386 AC parses you took could be similar due to fluky RNG, it's also possible it's an accurate representation of some underlying mechanism. One of your earlier parses showed exactly 1 hit for max damage; for any metrics relying on the count of max hits that's clearly going to be too noisy to be of use. Not a criticism! I think you're applying the appropriate skepticism, I'm just riffing on questions of data quality.
This is also why I like to first skim the data looking for any fluky RNG that might make analysis unreliable. By doing that before further analysis I can avoid any subconscious bias to keep data that supports my hypothesis or throw out data that contradicts it.
One thing I'm very interested in seeing is some parses at those same AC values for your shaman against a different target. I'd love to see you do some parses against Shiel or some other level 40 mob, because then we can overlap our parses and also see if there's any mob-specific effects like the squelch point hypothesis from the other thread.