This is the first post in what is hopefully going to be a series examining what I call the “Healer Problem” – the way in which the healing role changes as players gain additional power. In my opinion, healing gameplay becomes diluted as we progress through an expansion, or even within a tier, resulting in a less engaging experience for our role.
The Healer Problem is multi-pronged and complicated. In brief, I’d suggest it is related to the following things, all of which are intertwined:
- Gear-related power increases for the healer
- Gear-related power increases for the raid
- Encounter design that disproportionately rewards faster kills and damage avoidance
- The CD and AoE ability arms race
Now seems an opportune time to talk about it. Legion development and testing is underway but it’s still early, and we’ve seen a lot of signs of things being very much mutable and open to feedback and influence. We’ve also seen some hints of how several of these systems will be operating in Legion, so it’s a good time to talk about how the new systems will affect the Healer Problem, for better or for worse.
I want to start with analysing the first of the above points, gear-related power increases for healers. As a healer gears up, he or she gains power in the following ways:
- Increased primary stats (Intellect and Spellpower);
- Increased secondary stats (Critical Strike, Mastery, Haste, Versatility, and Multistrike); and
- Increased availability of mana (accumulation of Spirit, trinket effects, tier bonuses, etc.).
All of these power increases affect our throughput, and the effects are multiplicative. This leads to rapidly increasing healer power as we gain more gear. And this is a part of the Healer Problem, since, after all, our healing effectiveness is ultimately constrained by player health pools, which scale only with one primary stat, Stamina. But more on this in a later post.
The primary and secondary stats simply add raw healing power to all of our spells in roughly the same way, and are fairly easy to evaluate and understand. Mana, however, is a different story. Unlike the primary and secondary stats, gaining mana does not add raw healing power to our spells. It simply allows us to change our casting patterns. It’s been historically very difficult to quantify just how much of a difference some additional mana can make to our healing performance.
Mana is a particularly interesting starting point because the mana system will be changing in Legion. Currently on the Legion Alpha, Spirit does not affect a player’s mana regeneration rate. There is no Spirit on gear, and players simply regen a flat amount of mana every 5 seconds. This has potential to shake up the Healer Problem – but by how much?
It really depends on the value of Spirit, especially compared to the value of the primary and secondary stats that aren’t going anywhere. So before I can look at just how big a difference the Legion mana system is going to make to the Healer Problem, I really need to get some idea of how our Warlords Spirit growth has increased our healing throughput. And this means I had to take apart everything I already know about Spirit and put it back together again.
Spirit Modeling: The Basics
Evaluating the strength of Spirit as a secondary stat has always been hard. So far the best picture we’ve had of it has come from Hamlet’s series of Healing Theory posts (specifically Part 6), and I’d strongly recommend reading those for the detailed background if you haven’t already. For the quick & dirty breakdown, I’ll summarise here:
- A purely theoretical model of Spirit evaluates the best strategy to turn the least mana into the most healing.
- Thus, any model of mana usage will use the most mana-efficient (highest HPM, or healing per mana spent) spells as much as possible.
- These spells, being so powerful, tend to be cooldown-limited.
- So the best use of our mana is to spend it first on using our cooldown-limited heals as much as possible, idling between casts.
- Once we can cast these spells on CD, we spend excess mana on lower efficiency, spammable “filler” heals, with preference toward the most mana-efficient of these heals.
- As we gain mana, we cast these filler heals more frequently, increasing our throughput by reducing the amount of time we’re idling.
- Once we have enough mana that we run out of time in the encounter before we run out of mana, we switch to our least mana-efficient, but highest-throughput filler heals.
The overall mana strategy was illustrated in the below graphic. I hope Hamlet won’t mind me yoinking so I can reproduce it!
This model was incorporated into HealerCalcs as well, which used this theory to predict how much HPS a player would gain from adding a point of Spirit to their gear. It would determine at what point a player was on the blue line of the above graph, and the value of that player’s next point of Spirit would be pretty much equal to the slope of the line segment the player fell on. It was the best and most advanced of all the Spirit evaluation tools yet, and it worked pretty well. The results typically showed that Spirit was stronger than most other secondaries, and that felt viscerally true given the Warlords Spirit model of limited availability.
But. The model necessarily makes a lot of simplifying assumptions, something that Hamlet and I were both pretty cognisant of as we tested and as he refined the tools. It uses pre-programmed rotations that tightly constrain healer behaviour in a way that no healer would really behave in any real setting.
The dashed red line on the above graph is meant to attempt to correct for that somewhat. You can see that, because it deviates from the model’s core concept, the healing throughput of “realistic” behaviour is lower than that of the model’s prediction. But for a significant part of the graph – the lower mana availability section – the slope of the red line is steeper than it is for the blue line. This suggests that a point of Spirit is more valuable to a real healer than it is to the model healer.
We were never really certain if this was true. After all, the red line was just an illustration, not even meant as an estimate. How big is the difference, really? How much was the model underrepresenting the value of Spirit? For just how long does the realistic line’s slope stay steeper than the model line’s slope? And how on
Azeroth Draenor could we ever really know?
These are the questions that keep me up at night. And they kept me up a lot over the last two weeks as I pondered how to quantify Spirit’s role in the Healer Problem. Most people would probably be happy to just use the model’s predictions and move on. But I struggle a lot with abstract versus concrete, with prediction versus reality, and I wanted to take a look to see if I could answer, or at least start learning about, all these questions.
We know that real healing deviates from the model’s predictions, even though the model’s predictions would generate the highest possible raw HPM. It’s worth briefly going over all the reasons why – they are things to keep in mind over the course of the rest of this post.
- Mana is not actually the primary arbiter of our casting decisions.
- Sometimes the most apparently mana-efficient spell would not be fully effective, and therefore isn’t as mana efficient as its raw HPM would lead us to believe.
- Player mistakes – ours or others’ – often interrupt our mana-saving plans.
- At any time, we are actually trying to maximise our chance of success against the encounter, and this may lead to necessary inefficiencies.
- Encounter mechanics and damage patterns often trump all other concerns.
- Damage can often ebb and flow within an encounter, so maximum theoretical healing is not often required.
- The best healers are anticipating the next 10-20 seconds of events and saving/lining up their most powerful cooldown-limited spells for optimal usage.
- Movement and tricky mechanics can also interrupt our mana-saving plans.
- Intrinsic and extrinsic factors like talent choices and gear/set bonuses can have drastic effects on our behaviour.
So what do real healers actually heal like? How much do we deviate from the predicted spell priorities? How do we play differently as we gain more Spirit? This seemed like a set of answerable questions, and I turned to log analysis to start to get some answers.
First I had to choose a class to focus on, to act as sort of a model of a mana-loving healer. After a bit of thinking, I chose on Restoration Druid, as their high mobility meant that encounter mechanics and movement requirements were a smaller influence on their spell choice, so hopefully the effects of increased mana pools would be more obvious. Additionally, Restoration Druids have had good mana gameplay for the entire expansion, unlike some other classes I won’t name here. e.e
I collected data for eight fights throughout the expansion that I felt represented “healthy” mana fights – ones where healers had plenty to do, and felt constrained by their available mana.
|Healthy Mana Fights|
|Encounter||Average Fight Length||Average Healer iLvl|
|Heroic Imperator Mar’gok||11:55||656|
|Mythic Twin Ogron||6:43||662|
|Mythic Beastmaster Darmac||9:27||682|
|Mythic Hellfire High Council||7:43||712|
|Mythic Tyrant Velhari||6:10||719|
(I should quickly acknowledge that Mythic Butcher is a much, much shorter fight than these other fights, and it will appear as though players had more mana during this fight than their Spirit should otherwise allow if you don’t take this into effect. It is easier to feel mana-rich even when you are Spirit-poor in a short fight, because your starting mana pool is such a large proportion of your available mana.
Similarly, Heroic Imperator Mar’gok was a very long fight early on in Highmaul, so the opposite applies. Our mana regen became much more important of a factor than our starting mana pool, and since we were typically a little undergeared while fighting Mar’gok, it was easy to feel very constrained by our mana availability.)
I looked through logs of early kills of these bosses – where players were likely to be close to the intended gear level and Spirit level the fights were balanced for – and recorded information like the iLvl of each healer I was studying, the duration of the fight, and the number of times each healer cast specific spells. I tended to choose logs where the healer in question spent most, if not all, of their mana. Where necessary I excluded certain spells or certain logs due to talent issues (particularly Dream of Cenarius, which breaks the mana model), ending up with an (admittedly small) sample of 5 unique healers for each fight.
I used this information to work out how much mana healers spent during the encounter, to examine spell casting behaviour changes over the course of the expansion, to determine how many times healers really cast their powerful CD-limited heals, etc.
Then I fed this information into HealerCalcs’ model to generate comparisons to what the model would do given an equivalent amount of mana and fight duration.
This model used the following assumptions:
- Wild Mushroom and Wild Growth used on CD
- Rejuvenation as our primary, mana-efficient filler
- Regrowth as our inefficient filler
- Lifebloom cast before the fight begins, and refreshed with Clearcasting Regrowths
- No talents or glyphs included
I left out talents and glyphs primarily because it would muddy the data analysis more to try to model each of the popular Druid builds.
Swiftmend and Healing Touch casts are excluded from the model; Swiftmend is simply too mana-inefficient for a Druid without the Soul of the Forest or Rampant Growth talents, and Healing Touch is about the same HPM as Rejuvenation but is a less appealing filler.
So let’s see what happens when we compare the Empirical data set (the averaged output of the 5 log reports for each fight) against the Theoretical data set (the behaviour that HealerCalcs predicts given the same mana and fight length).
Restoration Druid Casting Patterns
First I looked purely at what spells the Druid decided to cast during each fight, and compared it to what the model suggested they should have cast for maximum HPM:
Keep in mind that Mar’gok is a nearly 12-minute fight, while Butcher is just under 4 minutes; these fights are two ends of an extreme spectrum between mana scarcity (Mar’gok) and mana abundance (Butcher).
These graphs show a huge discrepancy between the ideal cast behaviour and reality. First of all, players do not cast Wild Growth anywhere near as often as the model predicts. While the model casts 100% of all possible Wild Growths in a fight, players range anywhere from 19% (on Mar’gok) to 42% (on Kromog) of all possible Wild Growth casts.
Secondly, note that real players end up casting a much wider variety of spells, and lean heavily on Rejuvenation. A large part of this is probably that Rejuvenation is a more flexible spell than Wild Growth since it can be used for AoE or single-target damage, while Wild Growth is really only optimal for heavy AoE damage. As for the spell diversity? Well, this goes back to the previous section and why mana isn’t always the only factor in our healing decisions. For example, Swiftmend is a terrible thing to cast from both a HPM and a HPCT metric – ranking second-last and third-last, respectively – but it is a large, instant heal, and sometimes you just need a large instant heal to save someone’s life.
There is, however, a big piece of the puzzle that is missing from the above graphs: the amount of time that healers spent actually casting these spells as a percentage of the total fight duration. So I created a second set of graphs to illustrate another effect of accumulating Spirit – you can cast more frequently throughout a fight. Rather than count absolute spell casts, these graphs show the percent of fight length that healers spent casting each of these spells.
This rather neatly illustrates Spirit’s behaviour in regards to healer activity. With the exception of Butcher, a fight whose uncharacteristic briefness allows healers to cast more freely (and thus for an uncharacteristically high proportion of time), the general trend is that players can spend more of an encounter actually casting spells the more Spirit they get.
Let’s compare this now to the time-standardised prediction our HealerCalcs-derived model gives us.
The overall pattern of spending more parts of the fight casting if you have more mana still holds, and Butcher is still an uncharacteristic outlier. But compare the Mar’gok and Ogron bars in the Theoretical graph here to the same bars in the Empirical, Standardised graph above – look at how much less time the Theoretical model spends casting! It’s no wonder players prefer to deviate from this model, particularly early on in the expansion – reducing the number of cooldown-limited spells they use may lower their theoretical max HPS, but it increases their responsiveness to the fight’s damage pattern, resulting in both higher effective throughput, and probably a more engaging and satisfying gameplay experience.
But honestly, if you kind of lump all the purpley-blue bits – the CD-limited spells – together, and lump the green bits – the filler spells – together, other than the Mar’gok/Ogron discrepancy the behaviours look fairly similar. Real healers spend less time on their CD-limited spells and more time mixing in inefficient non-Clearcasting Regrowths or cheap but low-throughput Healing Touch fillers. Real healers have to re-cast Lifebloom because life isn’t perfect. Real healers know that they can’t respond to every damage pattern with a Wild Growth, and sometimes need to save someone’s ass with Swiftmend. But on the whole, the HealerCalcs model and reality are remarkably similar.
All that being said …
There’s still going to be a discrepancy in the value of Spirit for our model healer and our real healer. It’s the very same issue we already discussed theoretically, but now we can see it depicted very clearly in these graphs. Real healers spend about the same time as model healers do casting our cooldown-limited spells, but we do not funnel all of that time into Wild Growth the way the model does. The other spells we cast in that time are less efficient than Wild Growth, but some are more efficient than Rejuvenation, so assuming that every new point of mana we add grants us additional healing power equal to Rejuvenation’s mana efficiency would be inaccurate.
As real healers gear up, the mana they gain goes towards casting all of their spells more frequently, as well as in shifting some casts of efficient spells into casts of inefficient, higher-throughput spells. As a result, every new point of mana we add grants us additional healing power equal to some mixture of the mana efficiency of all the spells we aren’t currently using maximally. The exact number is going to vary based on exactly what spells the healer adds, and in what proportion, but it’s very likely to be a larger gain for real healers since they will gain casts of their cooldown-limited spells too. So we should expect to see that, as we add Spirit, our HPS rises more steeply in reality than it does in the model.
Describing the Discrepancy
To check whether this was true, I compared the HPS of what the model predicts to the HPS of the casts made by the Druids in my log studies for these 8 fights. I again ignored talents in both data sets, and used the same Int/Spellpower/secondary stats for all calculations, to isolate the HPS growth that arises from changes in the healer’s casting behaviour. I then applied a trendline to illustrate the approximate tendency of each set of data:
(If these numbers seem excessively low to you, remember that this model is excluding all cooldowns, talents, glyphs, gems, and enchants, and is generated using iLvl 656 gear.)
This data supports my thesis pretty well. Real healers do less raw healing with their mana than the model healer does, but they start to catch up as they get more gear and more Spirit and can act a little more like the model.
At some point, as you can start to see with the extrapolation beyond 720 iLvl, the two lines look like they’re going to intersect, possibly even cross, and players will begin healing for more than what the model predicts! This shouldn’t happen in actuality. Added mana should start to lose some value once healers are able to fill all of the available casting time with spells, and both lines should sort of taper off. We just haven’t actually gotten to that point in Warlords yet – at least, not before iLvl 740, anyway. The expansion has not yet left the “Zone of Healthy Spirit Scaling” from Hamlet’s graph at the start of this post.
In this post I’ve compared model behaviour to player behaviour to see just how much of player behaviour is really being described by our models. The answer is, unfortunately, “it kinda depends”. But we have confirmed our thesis that models tend to undervalue the throughput potential of added mana, since the model assumes perfect conversion of mana to healing. Real healing situations demand a more dynamic response to damage patterns, and we can clearly see that this results in real healers deviating from the model.
Note that the very last graph I’ve presented is starting to give us not just a visual representation of the differences between reality and the model, but an estimation of the size of the discrepancy. If all of the changes in player behaviour from one fight to the next are related to mana availability, then the pink line ought to be a valid measure of the value of added Spirit for real healers. We should be able to chart the increase in HPS as a function of iLvl or Spirit and come to some conclusion about how much throughput each point of Spirit provides.
|HPS Potential From Spirit|
|Data Set||Starting HPS
The only sticking point here is that assumption that all of the casting decisions are made based on mana/Spirit. I’m not really comfortable with that assumption after such a limited review. If some of those spell choices are made for other reasons, then Spirit’s value is probably a little lower than the slope of the pink line would suggest. I’m still quite confident the value is higher than the slope of the blue line, though. So in future posts when referencing the value of Spirit and comparing it to the value of additional secondary stats, I’ll be using the midpoint between these two lines as an estimate. As I’ve shown in the table just above, that midpoint gives us a 61% increase in healing done over the course of the expansion, solely from increasing our mana availability.
There’s still a lot of room for us to further improve our understanding of the mana-to-HPS relationship. For this particular study, I specifically chose a mana-loving class to analyse; while primarily this was because I felt Druids were likely the most sensitive to mana constraints, this choice is almost definitely biasing my results to a higher value of Spirit. If I were to apply this data towards the analysis of another class, like Holy Paladins, for example, I’d certainly be introducing error to the analysis.
Additional log studies covering other classes, a more rigorous collection of data, and a larger number of data points would go a long way towards helping us refine the HPM-maximising mana model. It’s the best we have so far, but perhaps there is an opportunity here to make it even better – just in time for Legion to change our relationship with mana.