A Balance Proposal

Greypelt

Squire
Squire
Joined
4 March 2018
Messages
6
Reaction score
2
A Balance Proposal

Contents

  • Abstract
  • Potential Proposed Solutions
  • Theoretical Examples
  • Concluding Statements
  • External References
A special note: About 15 people from around the AoC community have already seen this in various stages, giving a lot of useful feedback. Initial reception has been very mixed, many with opposing views have offered the most in terms of initial edits. I hope this can lead to productive discussion.


Abstract

Intrepid wants to create a complex web of augments, gear set bonuses, and 64 classes based on 8 archetypes. In the May 24th, 2017 livestream Intrepid stated that they want to balance to avoid cookie cutter builds with backend solutions. ( http://www.aocwiki.net/Ashes_of_Creation_Kickstarter_Livestream_2017-05-24#.5B.5D_Will_classes_be_balanced_to_avoid_cookie-cutter_builds.3F ) What sorts of solutions could Intrepid use to produce this end?



Potential Proposed Solutions

  • Make balance changes more often - much more often - if the class with the highest theoretical DPS rotates each day, it would increase the difficulty of chasing that top spot. In the same way if you adjusted class balance every ten minutes, it would become difficult to take advantage of being on top by stacking that class/spec if you didn’t present that info in a deliberate player facing manner since it would take that long to test the different available options. Number tweaks on abilities would need to be able to be changed live on the server.
  • How could this be done with an automated system without accidentally making the good choices even better? The devs should already have access to metrics that show What primary archetype / Class / abilities / augments each player is already using. With access to a large player base, the balancing system can be set to increase the overall primary archetype dps if the class represents less than 12.5% of the playerbase, to represent 1/8th of the playerbase playing each of the 8 archetypes, and decrease the classes overall dps if it represents more than 12.5% of the playerbase. Small incremental changes could be made to say increase the lowest represented classes dps by 0.05% per hour so that over the course of 20 hours the highest represented class will decrease it’s dps by 1% and the lowest represented class can increase it’s dps by 1% - As long as the goal is to maintain a ratio of primary class archetypes for dungeons/raids primary tanks/healers should be fine slotting into this system - if 20% of players attempted to play a cleric the system would tick along until some of those players were encouraged to play something else. The same logic can then be applied within each archetypes choice of secondary archetypes to encourage a balance of Mage/Rogues and Mage/Rangers. If players feel like one class offers better utility than another the system will make adjustments until players roughly feel like the utility to dps quotient has been balanced based on player subjectivity indicated by player behavioral changes.
  • For primary archetype Z, abilities that damage for x or heal for x, x can be raised or lowered automatically by 0.05% per hour until they are used y% of the time where y = 100/N where N is the total number of abilities for primary archetype Z. For non-damaging/non-healing abilities, either the system can increase/decrease the duration of the effect, increase/decrease the cooldown of the ability, adjust the resource cost or generation, or be completely ignored by the system if an ability is intended entirely for cosmetic purposes. Once the primary ability has been balanced for use, individual augments can be balanced in a similar fashion, perhaps accounting for the percentage of players that have access to an augment.
  • The balancing system could be made more intelligent. A Machine Learning Program could be made that predicted the impact of it’s changes to make smaller or larger increments of change or attempt more proactive changes based on things other than throughput, this could make the system more granular, help prevent over-corrections, and predict player reaction to balance tweaks. Modifications could be made to say lower the intended % use of augments intended to be rare. Upper and lower bounds could be set on the system. If an ability drifts 20% away from it’s intended dev set balance value, it’s possible that there are serious issues that need to be dealt with in regards to it outside the scope of what an automatic backend solution can handle. Updates could be made more or less often or on a partially random interval.
  • The system could be made with progressive balancing so that the top and bottom archetypes get adjusted by 0.05%, the 2nd highest and 2nd lowest represented archetypes get adjusted by 0.03%, and the 3rd highest and 3rd lowest get adjusted by 0.01% - This could help prevent changes that are as jarring.
  • In a survey of 2,000 MMO (WoW) players only roughly 38% of them listed ‘Fun’ as the primary reason to make a build decision (See ‘Side Effects of Cookie Cutter Builds’ Below). If player perception of fun is actively included in balancing, it should result in a larger percentage of players making decisions based on fun, with less of those players being seen in a negative light by performance focused players. Players making decisions based on fun can benefit from players making min/max decisions and visa versa. As players choose unfun specs it increases the damage of fun specs, and as people choose fun specs it increases the damage of complicated/awkward/bland specs.
  • If people were to try chasing deliberately unfun things to play, for performance benefits it should be self-defeating as other players perceiving them as being powerful would then bring them back in line. Most people should just settle into whatever they like with a small percentage of performance chasers.
  • Intrepid seems to want to obfuscate some performance numbers with a lack of DPS meters, so balancing based on player perception might be particularly relevant. This obfuscation would make it harder for players to notice live balance updates within a play session. Players would still be aware that the system exists, it just wouldn’t need constant player facing adjustment.
  • Balancing servers separately from each other could make it harder to follow a generic guide. A weighted system could be put into place in which say the population of the server counted for 50% and the overall player base counted for 50% of counted characters
  • What characters should be counted though? For Class count I’d recommend counting characters of any level, played for at least 190 minutes in the last 168 hours (About 3 hours, in about the last week) for augment/ability balancing I would recommend only counting max level characters as people will likely level with different abilities than they intend to use at maximum level - Exact details on this information should ideally be hidden from players, and ideally fluctuate slightly over time in order to increase the difficulty of attempting to game the system. Lower numbers here make the system more reactive, while higher numbers make the system more resilient.
  • A margin of error should probably be set on the system, for example if all primary archetypes represent 11-14% of the player-base the system could be set to not make any adjustments up or down. Adjustments up and down should be paired together so that the game doesn’t suffer from overall power creep.
  • The system should have a fail-safe for dropping a class to baseline rapidly if a large group of an underplayed class shows up working together.
  • Would the system handle pvp and pve separately? I figure PvE and PvP together, since the game is being balanced to share gear between PvE and PvP I figure things like monster health/dmg output will be balanced in such a way as to have the same things be appropriate for player vs player as well, but if there are hidden pvp stats the system could be set to adjust those separately.
  • A system of live balance updates might have a certain amount of pushback from players who want a static meta to exist.
Theoretical Examples

Explanation: This example follows the version of the proposal in which the most played class loses 0.05% dps per hour, the 2nd most played class loses 0.03% dps per hour, and the 3rd most played class loses 0.01% dps per hour. The three lowest dps classes respectively gain the same. All adjustments cease if all classes represent 12-13% of counted characters. Only population from a single server is counted in this scenario. Initial numbers are made up but the system should arrive at the desired class representation in the end with whatever it has to do along the way. This scenario is presented not as a representation of future starting numbers, but as an example of how the system would unfold.


Assumptions: Assume Ashes server “Fangler” averages 9000 concurrent players, 18000 Total Active players, who play a total of 36000 characters who play for the minimum amount of time to be included in statistics. On the first four hours of servers being up 9000 total characters are counted. It is assumed that players are aware that there will be live balancing updates, but are only given partial information on the system. This would likely cause some players to react quicker than they otherwise would.





On hour five 200 more characters qualify to be counted, and the table is adjusted as per the explanation.



Things continue along the same trend, but we don’t expect to see immediate player behavioral response. Going into Hour 6 the table is adjusted as per the explanation.


6


7


8


9


10


11


12


13


14


15


16


17


18


Sometime later, call it hour 48

Later


Sometime Even later, about a week in

Later +


Much Later, about 4 weeks in

Much later



Concluding Statements

  • Actual total adjustment numbers could be higher or lower, largely dependent on balance going in.
  • By 4 weeks in even though ‘Mage’ as a whole might have a 4.32% damage reduction, a Mage/Fighter might have received a separate 5% damage increase, while Mage/Summoners might have received an additional 4% damage decrease.
  • The system should be able to function off of a script, but could be improved with AI
  • This could present a useful data tracking and determining which class needs attention, even if not automated as a tool for manual balance.


External References

  • What are the causes of cookie cutter builds?
    • From WoWwiki
    • “Cookie cutter refers to a talent build that is commonly used by the average player looking to make an effective build without too much customization. A cookie cutter build is generally a build that is recommended to newer players, or players that have just recently switched to a spec they are not practiced in. A cookie cutter build is not always bad, and generally speaking, a cookie cutter build will in fact perform adequately, if not above average, in most situations. The reason these builds are deemed cookie cutters in the first place is because they tend to have the best talents, which is how they end up being widely used.”
  • A great data driven series on cookie cutter builds written by zoopercat of the askmrrobot team - part 3 of which links to more great theorycrafting on Cookie Cutter builds
 
Last edited:

Majerin

Recruit
Recruit
Joined
31 January 2018
Messages
12
Reaction score
5
While this is quite fascinating and indeed balance is a key issue when it comes to games and mmorpgs in particular I believe that a sheer balance based around play percentage is a mistake. While it is true that more people tend to play stronger classes, that isn't always the case: famously in wow for example hunters and paladins make up a massive amount of the played classes yet none of the four damage specs those classes offer have been "OP" in quite some time (and protection and holy,the tank and healer specialization, have been okay. But tank and healer meta in wow is famously far more fluid). Therefore, basing something purely off play percentage would lead to an underpowered class just because so many people like the archetype or what not. (And this phenomenon can be observed in most games. In swtor the most played classes are the two base melee jedi/sith classes by a decent margin, yet the gunslinger/sniper is typically more valuable: usually only one melee spec is in the top 5 of parses whereas sniper/slinger would have at least 2. So based on play numbers again the already stronger class would get a buff because more people just wanted to swing a lightsaber)

On the flip side, this could also induce something of a secret hoarding in an effort to keep strong classes secret which seems rather opposite of what ashes seems to want with its community.

While I do think play percentage is an interesting metric, i dont think it's the right one.
 

Morashtak

Tanks with Fishing Pole
Legend Premium
Recruit
Joined
27 December 2018
Messages
24
Reaction score
24
AoC Class
Tank
Discord
Morashtak#6754
IS has stated that they are striving for group dynamics and group balancing rather than individual archetype/class balancing;

Game Balance;
"There will be match ups in 1v1s where one class will be superior to another; and that application should be a rock-paper-scissors dynamic. We want there to be counter-play between the different classes... Instead it's going to be a group focused balance, where as long as you have the diversity of classes present, that's going to be an equal level playing field. It's going to be very dependent on skill and strategy." – Steven Sharif

The rock-paper-scissors dynamic is going to be more of a weaker-than/stronger-than mechanic rather than the auto win that is in RPS.

Being there will be 8 archetypes with a total of 64 classes we are looking at a chart somewhere between these two;


RPS15.jpg



RPS101.png



While overwhelming at first if one divides the second image into 8 major sections each with 8 sub-sections with the thought of "this section is stronger than that section over there but weaker than the others in another section" then one starts to see that IS appears to be aiming for a general set of damage numbers rather than specific ones, at least it appears to me to be so.

With that in mind let's rebut respectfully;


A Balance Proposal
  • Make balance changes more often -
Players dislike change for the sake of change. Any re-balancing on the fly with any discernible changes is going rub players the wrong way. This could even lead some to worry that the raid group they have worked on to put together for days if not weeks suddenly and with little explanation becomes nonviable and they have to start over from scratch leaving them with a feeling of wasted their time.


A Balance Proposal
  • ... so that over the course of 20 hours the highest represented class will decrease it’s dps by 1% and the lowest represented class can increase it’s dps by 1%.
See above. While a noble goal this could lead to unnecessary frustration when setting up the raid composition - What works one week is unworkable the next.


A Balance Proposal
  • Once the primary ability has been balanced for use, individual augments can be balanced in a similar fashion, perhaps accounting for the percentage of players that have access to an augment.
"Ah crap. Bob, your ranger/rogue needs to repec to ranger/cleric." "Why?" "Because the boss is all but immune to your special now." "W.T.F!"


A Balance Proposal
  • A Machine Learning Program... .
Assuming that IS has the expertise or contracts for this technology how it interacts with the rest of the code is always a crap shoot - Even the smallest of errors could lead to unintended swings in numbers that make game play fun for one class but totally un-fun for many others.

Should the game succeed widely and IS has cash to throw at it then this could be something to look at in a future patch/expac.


A Balance Proposal
  • The system could be made with progressive balancing so that the top and bottom archetypes get adjusted by 0.05%, the 2nd highest and 2nd lowest represented archetypes get adjusted by 0.03%, and the 3rd highest and 3rd lowest get adjusted by 0.01% - This could help prevent changes that are as jarring.
See above.


A Balance Proposal
  • In a survey of 2,000 MMO (WoW) players only roughly 38% of them listed ‘Fun’ as the primary reason to make a build decision.
Will have to dive into the numbers here as I am someone that feels a build is "fun" when it is viable for what it is intended to do - Does it help me level faster? Does it let me tank small group content better? Does it allow me to tank raids?

Some players like "flashy" and feel that is more fun than viability. Some will even make a build without knowing that a specific build is not as efficient as another but simply that because it since looks more flashy it feels more fun.

Again, will have to look deeper into the numbers.


A Balance Proposal
  • If people were to try chasing deliberately unfun things to play, for performance benefits it should be self-defeating as other players perceiving them as being powerful would then bring them back in line. Most people should just settle into whatever they like with a small percentage of performance chasers.
To be a bit redundant - Some players perceive certain mechanics to be more fun due to how flashy the mechanic is rather than how efficient it is.


A Balance Proposal
  • Intrepid seems to want to obfuscate some performance numbers with a lack of DPS meters, so balancing based on player perception might be particularly relevant.
Doesn't this go against your first few bullet points? Perception does not play into a Machine Learning Program, the program merely does what it is programmed to do not what others perceive needing to be done.


A Balance Proposal
  • Balancing servers separately from each other could make it harder to follow a generic guide.
Totally agree here - A raid party going up against a fire demon in an enclosed space is going to have a different composition than one going up against an ice dragon that is in a relatively large space. But the raid leaders need the numbers to have a certain consistency in order to ensure a measure of success between the time they choose the raid members and the first pull of the boss.


A Balance Proposal
  • Exact details on this information should ideally be hidden from players, and ideally fluctuate slightly over time in order to increase the difficulty of attempting to game the system.
Never underestimate the ingenuity of "hard core" gamers. While exact numbers may be hidden with a large enough subset of recorded damage numbers players that wish to will crunch the numbers and quickly figure out which class is better at a task than another.


A Balance Proposal
  • A margin of error should probably be set on the system, for example if all primary archetypes represent 11-14% of the player-base the system could be set to not make any adjustments up or down. Adjustments up and down should be paired together so that the game doesn’t suffer from overall power creep.
After reading through this a couple of times it appears to say that should a servers population have a near equally played set of archetypes (all eight) played and near equally played a number of hours than balance between all archetypes have been achieved.

Would this take into account of how they are being played? Ex. A large number of Rangers out gathering is not going to be played the same way as a equal number of Mages who are in different raid groups even if by some weird quirk both groups are being played the same number of hours and dishing out similar damage numbers.


A Balance Proposal
  • The system should have a fail-safe for dropping a class to baseline rapidly if a large group of an underplayed class shows up working together.
<solo player clearing a cave> "I wonder why no one plays Ranger/Rogue? It's great for exploring and clearing caves. I should grab some health potion first. BRB."
<meanwhile on the other side of the map> "Hey guys, everyone grab their Ranger/Rogues, we need to clear trash before we take out the boss."
<we return to the original player> "Okay, I'm back with the potions I need to clear out the last mobs in this cave so... <zap, pow> Why am I dying? It was going so well before! WTF IS!?"


A Balance Proposal
  • Would the system handle pvp and pve separately?
Am glad that there apparently won't be any "resilience" gear and all that other nonsense Blizzard implemented in their bid to balance PvP. Damage is damage whether applied to a player or a mob.


A Balance Proposal
  • A system of live balance updates might have a certain amount of push back from players who want a static meta to exist.
It's not so much about being "static" as being "reliable and transparent". Should changes be made they should not happen on the fly but be part of patches and included in the patch notes. This allows players to plan and coordinate rather than chase constantly changing numbers.


Just my two coppers.
 
Top