I’m using AMR and have been for a long time, but it’s hard to balance it sometimes against other guides. Pretty much every other website says that SFP is the best legendary for Shadow Priest bar none, but AMR says I should go a different Soulbind and run my 235 Eternal call to the Void over a 291 Shadowflame Prism.
What is the reasoning behind AMR being so far and away different from conventional wisdom?
Shadowflame Prism has been valued differently for the whole expansion - I’ve tested it on my priest at least a dozen times to make sure our simulator is doing the right damage.
I think it comes down to use case/preference. Shadowflame Prism is more controlled and less random, so you can often time it up to get really solid damage. We go with an “on paper” estimate that looks at total average damage for items as our general rule - in order for the optimizer to really work across any possible combination of gear, we have to follow the simulation data we have. Custom-tweaking the data can cause all sorts of unintended results.
Especially when it gets to the end of an expansion where players are at the top of the scaling curve… things get difficult and our advice can diverge. Our recommendations will always be good and keep you in the top tiers of damage (assuming similar skill) - but we won’t say for sure that you’ll get a number one parse against similarly skilled players running builds tailored to specific play styles. The game is too variable to give general advice that works in all cases.
The number of cases where we diverge from “conventional wisdom” is pretty small. Keep in mind that you can use the “customize” section in the UI to add a multiplier (positive or negative) to the value of specific legendary items if you would rather follow the meta. There’s nothing wrong with that - we built it into the site for this very reason.
At one point we were going to hand-curate “meta” builds as pre-canned presets where our data didn’t match what was popular - but we didn’t end up going that route. Maybe next expansion we’ll build out that feature depending on demand - but it’s a little hand-wavy for our theorycraft style.
Ideally, we’d like to team up with some players who play at a high level (mythic+, raid, etc.) who would be willing to give our default builds a “fair shake” out there in the game. That would mean playing with our advice for an extended period of time to overcome some of the game variance and see how good they would actually stand up in the cases where there is divergence from what is popular. It is extremely difficult to tell in many cases if our suggestions diverge because we have a bug, or if our suggestions would actually be just as good.
I can say from experience earlier in shadowlands that I was able to do comparable DPS to top tier warlocks using off-meta builds that our optimizer found for me.
Please define “high level”
Mostly people doing mythic raiding, but not in a race for world first scenario where they are trying to beat it under-geared.
High keys also (like higher than 15) - but that tends to be its own very specific meta-game that will favor builds that aren’t generic. Our “multi-target” build is roughly based on a mythic+ scenario, but it tends more towards a lower key done by less skilled groups scenario. It doesn’t assume people rounding up huge pulls and timing up every cooldown throughout the dungeon, treating it as one large encounter. I would expect high-end mythic+ players would need to customize our build advice. I’d be interested to know how far they can go with the generic advice, and where they feel it needs to be customized for specific dungeons in order to time them.
I’m running M+ as a protection warrior with quite uncommon team setup. Well, it was, now it’s just less awkward (Warrior Protection, Holy Priest, WW Monk, BM Hunter). Me and Priest are using MrRobot recommendations mostly, unless it gives inconvenient mixes. We were doing around 20-21 last season and we are expecting 23-25 this season (well, 18-20 right now). We tend to use our own routes based but not strictly aligned with quite popular ones. We are trying to use some MDI things also but with having in mind we are not that cool at pack control (CC and Interrupts). Also I’m a tank in raid team, usually we got our CE but closer to the end of season than to the start.
I can’t say Robot recommendations are wrong but some models are too synthetic and the closer the fight itself to a model the better results you have (obviously).
For example: normal raid tanking is not two targets and quite a lot of damage comes not from physical direct hits but from magical or bleed effects, this should value IP a bit more. Moreover usually you can be an active or inactive tank. Where active get a lot of hits and inactive got some nasty dot or sidestrikes. As a recommendation (while keeping things synthetic) I would like to have a second slider between magic and physical damage beside the TUF-ness itself. And usually we don’t have second target to hit in a raid fight, having a model with two targets gives more dps rating to things which hits more targets. Erm. I meant for pure sole following your recommendations give me less orange in parses than following popular opinions.
Mythic+ - your model are cool for small packs and shines on weeks when you can’t pull a lot. It gets worse when you consistently pull more than 5 targets, doesn’t include packs with huge magic hits and doesn’t rate long cooldowns good enough just because their second use is out of range of simulation or can be aligned with pulls.
Healing wise our Holy is pretty convenient with gearing but with a huge difference for legendary. A legendary which gives you a holy light buff is so different than in model. Your model involves you are out of mana by the end of fight but for m+ a fight is a 30 minutes run and the less time you spend drinking the better your dps (and WO made a huge difference for last season). And the fun thing the higher we key run the higher damage income we got but in a real at some point only tank gets damage because everything else is basically one-shot or near it. So it starts to have a spiky nature and will rate haste more.
As a dps, I’m not a good dps player but let’s say bearable (60 to 80 parses over logs) your model works only for fully buffed person. Like runes, food, flask, pots on cd and hero on cd, having something out may broke some “magic” relations and just a plain following recommendations gives you more consistent result. I know that the difference is not huge but that just the feeling I have to share.
Anyway, saying this I wanted to say if I can help MrRobot to became better I’m up for it.
Sorry for some weird language constructions I’m not a native speaker.
Thanks for the info.
Adding another raid-tanking strategy that focuses on a purely single target fight with more magic damage is certainly something we could look at doing if people think it would have good value.
I haven’t heard anyone talk about the fully-buffed nature of our recommendations changing things before. I was under the impression that most advice you find out there assumes you are buffed up. We could probably adjust some of the recommendations based on missing particular buffs.
Skolex is great example of a tank fight.
You are actively tanking or chilling with rend. And taking Riftmaw sometimes.
Dausegne is almost pure magic fight, light hits plus magic dot
I tried to make a single target raidboss model to check that it’s not just me but miserable failed 
Thank you for looking into it too.
Are you considering the shadow priest 2 piece tier effect? The dark thoughts proc gives an instant cast mind blast increasing mindbender uptime and damage. I have routinely seen uptime doubled in raid with mindbender being 13-16% of overall damage. Here’s a log from a very great shadow priest (not me 
) Shadow Priest Log highlighting mindbender damage
Yes, we are certainly considering the 2 piece tier effect. I’ll mess around with the rotation a bit today and try running more data, see if I can get something that looks closer to what people would expect in the optimizer.
I posted an update. Looks like the optimizer likes shadowflame for the multi-target now, by default. Still won’t pick it single target by default - but that still makes sense to me, no matter how much I try to massage the data/simulations.
I’m curious what the conduit and talent setup is that it prefers Talabar’s for single target.
I don’t see how Talbadar’s wouldn’t be the obvious choice for single target. It was already the strongest choice before the tier sets, and now the tier set has synergy with it as well - making it the easy choice for single target damage. I can’t think of a talent/conduit setup that wouldn’t favor Talbadar’s for single target.
The synergy with shadowflame makes a lot of sense for multi-target - but it’s not going to compete numerically with Talbadar’s for single target.
I’ve spent some time trying to play mage in M+, can’t say all of them were high M+ but 15+ for sure. Also, have some experience as a tank for higher keys and know what usually happens near 20-22 (not top keys now though). So now I want to share some thoughts about models (not simulation data or rotation things)
5 minutes sims and heroism / bloodlust
I have a feeling (and some guides also mention it) that heroism shouldn’t be included by default. It lasts 10% for 5 minute simulation but in real life situations it is below 5% and I would even say it should be ignored for PUG routes
Pull size
Pulls are different - normal pulls which are about 4-5 mobs, big pulls about 10+ targets and huge pulls. Normally people await big to huge pulls and use normal pulls only in some cases in extremly high keys. This can drasticaly change rotation and stats
Cooldown aligning
Big pulls are possible because you align them with your big cooldowns. The more pulls are aligned the more bursting classes are better and differ from sims. Imagine - 20 targets in 30-40 seconds and then minute or two of 1-2 targets and then 20 targets again.
Maybe it’s worth trying?
I liked the idea for tanks / heals (the idea is definitely brilliant) to have a slider which adjusts offense / defense ratio.
What if we could have a slider to adjust ratio between amount of targets?
<aoe> - <cleave> - <single> | <huge> - <a couple> - <one>
And also quite an interesting view
not going to discuss correct or not everything else but the idea from here quite short burst sims for about 2-4 minutes with almost endless hp on targets.
