Monday, June 21, 2010

Gearing Up

Recently there has been a good deal of discussion about how gear will be handed out in 10-player and 25-player raids in Cataclysm. Blizzard representatives have waffled on the issue a little bit, which is natural since they are still in the alpha and things could change dramatically between now and release. We know that 10- and 25-player raids will give the same loot, but the goal is that 25-player raids will be more efficient in gearing up the raiders to reward the extra effort of organizing a larger raid.

How can this be accomplished? One suggestion is to have 25-player bosses drop more "badges" to buy gear with, another is to have 25-player bosses drop slightly more loot per person but with the same loot table.

What is ignored in this discussion is that 25-player is more efficient than 10-player at gearing up the raiders with precisely the same loot per boss. Intuitively this is actually pretty easy to understand. Suppose you live in a simple raiding world where a boss can drop five different items, you want one of those items, and you can fight the boss by yourself. Each week you have a one in five chance of gearing up. Suppose instead you have to fight the boss with four friends to defeat it, but the boss drops five items, rolling on its loot table five times. Assuming that your friends want different items (or that some of your want different items at the very least) you can now hand the loot out to the people who need it. If the boss drops item 4 and you want item 2 then at least someone gets item 4. And when you go back next week you know that you have five chances at item 2 and if it drops you get it because the other four people don't want it.

That's a very simple example, but it shows the principle. If we got a roll on the loot table for everyone on the server and then handed out loot to people who need it, almost everyone would get what they wanted, and those who didn't would be virtually guaranteed of getting it the next week.

How does this work in practice? Well, raid loot is pretty easy to simulate. Here is the gearing progress for 10- and 25-player raids fighting bosses that drop 1 item per 5 participants, 12 bosses a week. Tanks are selected randomly from all tank specs, same with healers and dps. 10-player raids are 2 tanks, 3 healers, 5 dps and 25-player raids are 3 tanks, 6 healers, 15 dps. There is one of each type of item in the mix (1 mail shoulder with intellect and another with agility, for example) and the chance of each item dropping is proportional to the number of people who can use it (there is five times as much plate itemized for dps than there is for intellect, for example). Finally, I assume there is only one useful trinket available to each raider despite having two trinket slots (this is how things seem to always work). These assumptions are not what the reality would look like, but they work to show how the curves look. Here is 8 weeks of raiding:

For reference, the bars show the number of shards per raider. Of course because the 10-player raids have less useful loot they have more shards per person. The percentage shows the total percentage of slots filled in the raid. Each person has either 14 or 15 slots, depending on whether they use items in both hands or use a single two-handed weapon.

So after 8 weeks, 70% of 10-player slots are filled. That means the typical 10-player raider has 10 or 11 pieces of gear, leaving 3 to 5 upgrades left in that raiding tier. After the same amount of time a 25-player raider will have 12-ish pieces of gear, meaning they need 2 or 3 more upgrades.

Here's the same chart for 15 weeks of raiding:

You see that both 10-player and 25-player raids are plateauing at this point. Extended infinitely both would tend towards 100%. But for realistic length of time, 25-players raids will tend to gear out their raiders about 95% of the way while 10-player raids will only tend to gear out raiders about 90% of the way.

So 25-player raids don't need more "badges" or more drops in order to gear up more efficiently. They gear up more efficiently by just rotting less loot. In fact, even if 25-player raids get less loot per person, they can still come out on top. This is 15-weeks of raiding with only 4 drops from 25-player bosses:

On this chart we see that while the 25-player raid starts out pretty even, it pulls ahead as time goes on.

What we have here, then, is really a problem with innumeracy. People want there to be an incentive to raid 25-player because it is harder to raid 25-player. Finding nine friends you like to play with or nine people who are really pretty good is not nearly as hard as doing the same thing when nine becomes twenty-four. It turns out, though, that if we actually understand how the math works out, that incentive exists with the same rewards per person for both raid sizes.

What should the developers do with this information? In a perfect world I'd say set the drops per boss and "badge" rewards to the same per person for 10- and 25-player raids. In reality, the amount of time they would spend explaining why this makes 25-player raids more efficient is probably pretty prohibitive, and their current plans of giving more rewards to 25-player raiders is probably the better one from their perspective, even if it is a tragedy from mine.

Appendix A:

Appendix B:

Monday, June 14, 2010

signed long int Cash_Money

The statistics in the statistics tab are not exactly reliable. Some of the errors they contain are bizarre (deaths in arena categories you've never played in, for example) but some are all too predictable.

It seems that the developers used 32-bit integers to store gold, both in your backpack and on the statistics page. 32-bit integers are pretty big, and record numbers a little over 2 billion, but there are two problems.

First of all, 2 billion is a really big number if you are talking about gold. We are not talking about gold here, however, but rather about copper. The game stores your money as a number of copper and formats it into gold/silver/copper for display only. 2 billion divided by 10,000 is only 200,000. In fact, the number in question is 214,748.3647, the familiar "214k and change" that someone will say if you ask "What's the gold cap" in trade chat.

Now, if you are part of a very small subset of people who has memorized the powers of 2 but is not familiar with computer science, you'll think, "What a minute, 2,147,483,647 is not 2 ^ 32 - 1, it's 2 ^ 31 - 1. Where did that last bit go? It went to keeping track of whether your cash-on-hand is positive or negative. Just in case you have negative cash, I suppose.

When a number can't be negative, you can use an unsigned numerical type to release that extra bit and use it to store more number instead of storing useless information. This is probably rarer practice than it should be. If the developers had done this then the answer to the trade chat question would be "429k and change" and we'd have a lot fewer people at gold cap. This, of course, wouldn't really make much of a difference to anything.

But while your money has an error message to protect it from overflowing its alloted bits, your statistics page has no such protection. If you earn 215,000 gold over your character's career, the statistics tab with faithfully report your lifetime earnings as just under negative 2 billion copper. Having overflowed the 31-bits of money storing memory you've flipped the 32nd bit and moved to negative town. If the developers never thought anyone would collect 214k gold when they were doing the alpha for classic WoW, they can be forgiven for that assumption. If they thought, however, in the Wrath alpha that no one would ever earn more than 214k gold over the entire course of their character's career, that is significantly more bizarre. What is much more likely is that in both cases they simply didn't think about it at all.

Thursday, June 10, 2010

Too Soon?

We've got talent previews for four classes for cataclysm at this point. As a priest there is a lot to be excited about. As a resto shaman there is nothing to be excited about. In fact, the attitude of a resto shaman I know was closer to "pissed" than "excited." Not only are there virtually no changes to the tree, but some changes that we pretty much received specific promises about didn't happen. There is still a talent for 5 points that is worth 2% extra healing per point.

Discipline Priests have 5 new talents, 6 talents moved around between tiers and quite a few substantial changes to the way they work. I tried to make a discipline build and if their goal was to give me interesting choices and to create a "cookie cutter" talent build that left 10 to 15 points left over to spend how I want then they were successful. They were probably even more successful at this in holy where I can see many different ways to spend my last 10-or-so points. In the resto shaman tree there two new talents, one talent has been removed and everything else is exactly the same as it was. My attempt as a resto shaman build left me feeling like there were very few places to spend my points, and ended up looking almost exactly like a current resto shaman build.

From Purification and Nature's Blessing I hope I can assume this tree is far from finished. Seems like they should have previewed three classes instead of two; there might be a lot fewer worried resto shamans out there right now.

Friday, June 4, 2010

Near Misses: Bosses that could have been great

I was looking at a list of the best bosses in Wrath of the Lich King today. I expect that since we are in pre-expansion season people will be compiling lots of lists of their favourite and least favourite parts of this expansion. There were some very good bosses in the expansion as well as some pretty mediocre ones, but I don't want to dwell on the good or the bad. Instead, when I looked at the list, I found myself thinking about the bosses that could have, or should have, been great but that weren't. I don't want to force anything in this list that doesn't belong, so I've only got four of them, but that's not to say I necessarily got them all. There are plenty of fights that didn't work out, but these are the fights where them not working out is a significant let down.

Ignis the Furnace Master: It might be a bit of an exaggeration to suggest that this fight could have been great but it at least could have been pretty good. It turned out, though, that the easiest way to win was to completely ignore the fight mechanics, have your tank stand in the water and tank the boss and all the adds, and just kill him before you got enough adds to be a problem. When we were doing this fight the "right" way it seemed pretty interesting. When we decided not to bother with all the fancy stuff and just burn him out it became a simple tank and spank. To fix this they really needed to make shattering the golems a more important part of the fight. Rather than shattering golems purely for the sake of getting golems off your back, if golem shatters actually advanced your cause in terms of winning the fight (for example, shattered golems drop an item that can be used to do massive damage to Ignis) then people would do this fight by shattering golems and we'd get a fairly good fight.

Twin Val'kyr: Importing Ikaruga to an MMO was a great idea, but unfortunately I feel like it just didn't work. Ultimately our successful strategy was for most of the raid to spend most of the time standing in a pile right behind the boss. It was a tank and spank where every 90 seconds there was a 50% chance they had to do something other than keep attacking. For the healers playing goalie with the white orbs it was a little more interesting, but it was still mostly just spamming raid heals. I truly wish this fight would have been awesome, but the black and white switching mechanic was kind of clumsy and the clumsiness of it meant that they didn't make you switch very much. This fight would need quite a significant redesign to live up to its promise, but a starting point would be to find a good technical way to let us switch colours on the fly. That would have opened a world of possibilities of encounter design they had to ignore.

Sindragosa: This fight seems like a lot of fun and is in many ways well executed, but it has two rather awful problems. The first is Chilled to the Bone, the mechanic that punishes auto-attacking characters. She punishes casters as well with Unchained Magic, but no matter how many casters you bring, she only punishes one of them. Each caster past the first is a full time dpser. Each auto-attacker, no matter how many you bring, is affected by Chilled to the Bone. That means stacking melee is bad. But much worse than this is the way the debuff is applied. The faster you attack the faster you get stacks, so rogues barely get to play. If you look at the top dps on World of Logs, rogues are very visibly absent from Sindragosa's charts. There hasn't been a fight in recent memory that so badly punished a particular class.

The second reason this fight didn't work out is because the initial phase of the fight is long and relatively easy and then it becomes very hard. After a few tries to get the hang of it, you rapidly get into a mode where you spend the majority of your boss fighting time doing boring stuff you have proved you can do dozens of times and a small minority of your time actually practicing the part you need to beat.

Why do I say Sindragosa was close to being a great fight? They could have made this fight a lot more fun to learn by doing a few simple things. Rearrange the Chilled to the Bone mechanic to make it apply more evenly to all auto-attacking dps. Reduce her health, enrage timer, and the length of ground phases, but leave the transition point to the end phase at the same amount of health. Basically by tweaking just a few numbers this fight could have been great instead of boring and frustrating (especially for our rogue who we refused to bring while we were trying to get our first kill).

Anub'arak: This fight is the grandfather of disappointments. Basically I think this fight is brilliantly designed and really unique but that it is effectively spoiled by three things, and also harmed by a missed opportunity on the developers' part.

The first is the limited attempt mechanic and the fact that you basically only got one attempt per week. This is really the fault of limited attempts not of Anub, but Anub had the misfortune of being the final boss of the one attempt instance, and that will forever mar my memory of this fight.

The second is the fact that a 50% healing debuff was mandatory for this fight. There are three classes that can provide this debuff (and warriors can't do it while tanking, which they might need to do), so in 25-player this was presumably never a problem. In 10-player it could be a big problem. This wasn't a case where it was really hard to win or one class was over-emphasized. When we attempted this fight without a mortal strike debuff we simply weren't doing any damage to Anub once the swarm was out. We'd hurt him, but he'd heal back up while we were killing adds. You may be nuts if your raid without the 13% spell damage debuff, but it isn't literally mandatory for any fight. This requirement was way out of line.

The third fatal flaw with this encounter is the need for a shield tank. Since the damage multiplier applied by the adds applied *after* blocking effects, a well geared block tank could take no damage at all while tanking them, where a similarly geared bear or death knight would be taking nearly 14k dps. We had the good fortune to be using a paladin and a warrior for our main tanks though ToC, but if we had been using a DK and a druid as we were in Naxx we basically wouldn't have been able to win.

Finally, this fight was a missed opportunity. The developers decided in ToC to drop the concept of multi-step hardmodes. In the future we won't have a fight like Yogg, Sartharion or Flame Leviathan, where you get to choose how hard a hardmode you want to do. It will just be hard or easy. Anub was a great chance for a multi-step hardmode. His hardmode was harder in three very different ways that could have been separated out: 1) Frost Patches don't respawn, 2) Adds spawn during the swarm phase and get shadow strike and 3) Swarm deals 20% instead of 10% health per second. I love multi-step hardmodes and the opportunity to practice at one level before moving on to the next. I have to assume they were abandoned because it was felt they took too much developer time, but it's a real shame. Especially on a fight where it would have been so easy to implement one.

And what about Anub's problems? All are extremely simple to solve. Get rid of the concept of limited attempts, reduce the healing of his swarm by half and make him immune to healing reduction effects (at least on 10-player), and change the way the stacking debuff works from the adds so that there isn't a 14k dps difference in the damage taken by shield and non-shield tanks.

I can't really blame them for trying limited attempts once to see how it worked out1, but the other two are so glaring as to seem unforgivable. This fight should be something I remember with awe the same way I think of KT, Yogg and Lich King. Instead it just seems like a nuisance.

1. I can't forgive them for using limited attempts again in ICC. It was terrible, terrible, terrible.

Wednesday, June 2, 2010

DoT Leveling

Since my main is a priest I am clearly going to level by casting DoTs on things and then ignoring them until they die. I used to love smiting my enemies down, but those days are well and truly gone.

Of course the thing that matters when you are killing swaths of enemies with DoTs is how much damage you can do with one or two DoTs. This typically meant that for DoT classes the only thing that mattered was spell power when leveling (and at max level as well, I suppose). With hasted DoTs adding ticks to the end of DoT length, we now have the possibility of increasing our damage through haste rating as well.

If you are wearing two-piece tier 9 then your Vampiric Touch ticks seven times1. That means you would need in 14.29% haste or more to add an extra tick to the end of your Vampiric Touch. If you have less haste the ticks will come faster but you won't actually get any more damage out of them. At level 80, that would require 469 haste. Here's some assumptions we'll use to figure out how much haste increases our damage:
  1. We have about 3700 spell power
  2. Darkness will be going away to be replaced by our 8% passive talent tree bonus
  3. Shadow Orbs will have no effect on our damage if all we do is cast VT and SW:P
With 3700 spell power, Vampiric Touch will do nearly 15400 damage. That 469 haste would give us an 8th tick, increasing the damage to 17600. 469 spell power, on the other hand, would increase the damage by only 1794. Haste wins! Well, haste wins as long as you have exactly the right amount of it, more or less doesn't affect your damage at all until the next multiple of 469. If we are casting Shadow Word: Pain as well then spell power beats out haste at 469, but if we get 547 haste - the amount required to get a sixth tick of Shadow Word: Pain - then haste pulls ahead again, despite the extra 78 haste having no impact on our Vampiric Touch damage.

But we all know that the amount of rating required to get 1% haste is going to go up massively between now and level 85. Here is a list of how much rating it would take to add one tick to your Vampiric Touch as you are leveling.

LevelRating for Extra VT Tick

While one extra Vampiric Touch tick might do more for you than 469 extra spell power, it certainly does not do more for you than 1928 extra spell power. In fact, at that rating conversion, haste is worth less than 30% of spell power, and it is only worth that much if you have the right amount. 1800 haste rating, of course, is completely wasted.

What's the lesson? While leveling, ignore everything but intellect on gear. A piece of gear with haste may be better at level 80, but it will become obsolete in no time. Unless something is done about the rating conversion, haste will stay obsolete the entire expansion, just like all ratings will for all classes.

Side Note: Will single- or double- dotting be sufficient to kill things?
According to modelling tools, level 80 characters in exclusively quest greens can do about 2400 dps without raid buffs. That means the typical level 80 monster is about 5 to 6 seconds worth of modelled dps. Assuming our dps doesn't drop between level 80 and level 85 - which isn't necessarily a safe assumption, but lets run with it - that should put us up against mobs with around 30k to 35k health. We would be able to two-DoT for that much if we can gain around 1900 spell power - the resemblance between this and the amount of haste we need to get one extra tick on Vampiric Touch is surely coincidental.

1. If you threw away your tier 9 gear, run some heroics and buy yourself some gloves and shoulders for leveling and farming. Even at level 85 I doubt there will be anything better for mowing down large numbers of mobs.

Tuesday, June 1, 2010


Time to start complaining about things that don't even exist yet.

Things are still in alpha so it's a little too early to complain about the particulars of anything from Cataclysm, but its the very best time to complain about the general structure of things. Basically however mastery looks when we get to the beta is probably how its going to look when we get to release and when we get to the next expansion. The numbers will get moved around, but the basic premise of how you get mastery will take a huge amount of effort to change.

When we first heard about mastery it was a passive bonus attached to every talent point. More talent points means more bonus. Since then we've seen how this concept has been refined. Basically its refinement has dealt with some fairly obvious problems.

Problem: People will be encourage to spend all their talent points in their chosen tree regardless of whether they want those talents just to get the passive boost. A retribution paladin will, for example, want to spend all 71 points in retribution since those give a damage bonus. In this way, mastery will lower choice instead of increasing it.
Solution: Put a cap on how much mastery bonus you can get from a tree. Only the first 51 points give you the bonus.

Problem: Classes that have multiple trees that perform essentially the same function will be able to "double-dip" mastery bonuses. Any mastery bonus you could give a rogue, for example, would be good for a rogue of any talent spec.
Solution: Only give people mastery bonus from the tree they put the most points into.

Since then people have raised some other problems. The mastery system will mean you need to put 51 points into a single tree and so will essentially kill hybrid specs. Another issue is from a new player perspective. You hit level 10, you say, "I want to smite people with awesome holy power" so you look at retribution. Sure, the talents don't appear to make you better at smashing faces, but that's okay, there's that mastery bonus at the bottom of the tree. So you click your point and you get... 0.156% extra damage and 0.392% extra holy damage. You might be a little unimpressed.

The solutions proposed for these appear to be to say, "Who cares about hybrid specs anyway?"1 and to not realize the second issue is a problem at all.

I have a two stage solution to both problems.

The first stage is to put mastery on a non-linear scale. Since you only get mastery from one tree, and only the tree you've spent the most points in, you don't have to get the same amount of mastery for every point you put in. Instead of the first point giving you just under 2% of your final mastery bonus, it could easily give you around a third of it. Imagine if that first point gave you 3% bonus damage, healing or damage reduction. At that point you wouldn't wonder what it did, you'd know what it did. Your talent choices would help to define your character's abilities from the very first point. This would also prevent the system from punishing hybrids excessively. If you get to 80% of your final mastery bonus for putting 30 points in the tree then you will be closer to a full master bonus if you split your points three ways.

The more extreme solution, and the road I think they should probably take, is to build on what I've said above but base mastery bonus on level rather than on talent points spent. Talent points would only determine which mastery bonus you got. This would completely eliminate the senseless penalty to hybrid specs. It would also provide a very meaningful and noticeable way to reward advancement because mastery bonuses could be awarded in steps at certain levels rather than in tiny increments. You could even put achievements in (achieved rank x mastery) to let players know that their mastery bonuses had increased. The mastery system, since someone dreamed it up, has gotten further and further from adding a passing bonus to all talents, might as well go whole hog.

I would put the probability of my second solution at extremely slim, but there is some hope for the first one. Suggestions forum, here I come!

1. If hybrid specs don't matter and specs aren't really specs until you put in that 51st point then why does Titan's Grip need a 10% damage penalty?