Saturday, February 21, 2015

The danger of sims

People put way too much stock in simulators.  Or, perhaps more accurately, they refuse to acknowledge how often simulators are incredibly bad at modelling the way things actually work.  My example today is WOW Paladin talents for Retribution, the dps role.  The two talents that I am comparing are Execution Sentence, which is the one I see all the geared ret paladins using, and Holy Prism, which is what I use and which I feel is drastically superior.

Execution Sentence is a 10 second debuff that does 40k damage, with most of that damage coming at the end.  It has a 1 minute cooldown.

Holy Prism is an instant cast that either does 13k damage to a single enemy and a 14k heal to 5 nearby allies, or an 19k heal to a single ally and 10k damage to 5 nearby enemies.  It has a 20 second cooldown.

Simulators tend to think that Execution Sentence is the best for a few reasons.  It comes out of the gate better and has a very high damage per cast time value.  This is relevant because although Holy Prism also does 40k/minute it does so over 3 casts and that has the potential to interfere with other ability uses.  So if you build a simulator that models a stationary boss that just stands there for 5 minutes while the paladin beats on them then Execution Sentence looks better.  This, however, is a wretched way to model actual fights.

First off, fights often have lots of opponents to defeat.  The ability of Holy Prism to blast 5 mobs at once is hugely useful.  This is especially true because it is instant and so it makes it easier to quickly blow up dangerous enemies.  Execution Sentence is only useful when the target will live 10 seconds and often that isn't the case.  Execution Sentence also requires planning around times that the boss will go invincible, vanish, or otherwise render damage over time effects null.  It may seem like Execution Sentence goes off once a minute but because it actually has to sit there for a full 10 seconds there are going to be many times when it won't be usable or will have terrible performance.  Holy Prism never suffers from this problem.

The other critical thing is that even in the situation of a stationary boss with no extra targets Holy Prism adds a really considerable amount of healing.  The numbers above are my unbuffed values and given that a healer is probably putting out about 15k healing/sec.  (All numbers will rise drastically with gear and/or raid buffs.)  A single Holy Prism will heal 5 targets for 14k, for 70k healing.  That is nearly 5 seconds of healing, and it happens every 20 seconds!  It clearly isn't replacing a healer but randomly splashing 20% of a healer's output onto the tanks and melee is a serious benefit.  When using the version that heals a single ally and damages 5 enemies it would obviously be targetted on a tank and 19k into a tank is a nice side benefit to a strong AOE ability.

The fact is that simulators are good at modeling simple encounters but rubbish at figuring out the value of flexibility.  They also don't have any good valuation of healing output for a dps class and default to only worrying about damage dealt.  That doesn't mean that the simulation is wrong but it does mean that you have to actually think rather than just rely on what excel spits out.

I have gotten into fights over this in the past with people who insisted that the value of healing delivered by a dps class was zero.  I didn't buy that argument then and I don't buy it now.  In this case though you don't even need to accept that to believe that Holy Prism is just superior because its instant cast nature and AOE capabilities make it a better choice even if it didn't output a useful chunk of healing as a byproduct.


  1. This is why you need a spreadsheet that you paste your actual combat log into to evaluate how useful each ability is. Right?

  2. Don't make me spend one hundred hours building such a spreadsheet. I will do it, dammit.

  3. I don't even think I have my old priest spreadsheet anymore. I used to do just that, but it was never perfect enough for me to post on Elitist Jerks (it was miles ahead of what everyone else was doing since the priest forums were a bit of a wasteland, but it wasn't good enough for me).

    I think parsing logs line by line is probably only important for healers in raids because that was the only place where it's genuinely really hard to tell what happened. In particular there was that old discipline talent that stacked extra healing if you kept healing the same target that had a 50% chance to apply with one point and 100% with two and the confusing thing was that a 50% chance *also* meant you had a 50% chance to keep your stack if you cast one heal on someone else before switching back. I needed to actually have the parser run the hypothetical scenario on real logs to see how useful the second point was.

  4. At one point in time I do think healing from a DPSer was worth zero, or close to it. 40 man raids with healers running mods that heal low people at the click of a button? Your healing is pretty much worthless. Though to be fair your damage was often completely worthless too. The raid would win or lose even if half the raid went and made a sandwich instead of fighting.

    I don't know how things are now. Blizzard keeps claiming they're changing healing to be more interesting than full-almost dead-dead and if that is true then DPS healing can be rather valued. Also in a 10 man raid replacing a healer with a DPSer who does some healing can be clutch.

  5. Yeah, 40 man raids really very little mattered. However, I think healing from a dpser matters now. I don't pump out a huge amount but I do enough that it helps I think. Certainly people do hang around at medium life for awhile so any extra I put out there does something.