## Mathematics of XP

##### August 08, 2006

What is the best mathematical principle
upon which to base an experience system? To explore the mathematics of XP we
must first look at why such systems can be useful, and then examine the
properties of the mathematical functions upon which they are most commonly built.

Experience systems have their origins in
tabletop war games, but became more significant when they were used as the
central progression mechanic in *Dungeons & Dragons* back in 1974.
Since then, their significance to tabletop role playing games has diminished to
irrelevance, but they remain of central importance to computer RPGs where they
provide a centralised progression mechanic effectively brokering the time the
player spends with the game into rewards. By relating a character’s power level
to experience (and hence indirectly to time spent playing) cRPGs allow any and
all players to progress in return for their time commitment.

This can be seen as the central transaction between the player and the game in the context of cRPGs: “Continue to play me, and you will continue to progress.”

Compare this to the deal offered by
adventure games (“Solve my puzzles to progress”) or tactical and strategic
games (“Try to beat me!”) and the strength of the experience system (and its
equivalents) becomes more apparent.

Central to the implementation of an experience point (or XP) system is the gearing of the progression mechanic – that is, the relationship between the XP required for one level and the next. This gearing can be arranged according to a number of different mathematical models, with different implications.

It is important to note that we will be
examining solely the XP values for levels in this discussion (because this data
is readily available) and not the XP rewards the player receives for, say,
killing monsters. This data is harder to come by, but we can estimate the
effect of this element, as we shall see shortly.

**Progression Ratios**

In examining these differences, we should
consider in particular the *progression ratios* – which are determined by
dividing the XP required to reach one level by the XP required to reach the
previous level. There are two such ratios of note:

- the
*basic progression ratio*: [XP difference between level (n+1) and level (n) / XP difference between level (n) and level (n-1)] - the
*total progression ratio*: [XP for level (n+1) / XP for level (n)].

Note that these ratios are rarely fixed
values, but usually *decrease* as the game progresses as a consequence of
the mathematics of the XP systems that are used.

In general, the lower the basic progression ratio the less difference between the time the player must spend to reach one level and the next, that is, the less steep the progression curve of the game. It should be remembered that the basic progression ratio need not be constant, and in fact generally recedes as the game progresses. This doesn’t mean that progress gets faster as the game proceeds, however, owing to the cumulative effect of multipliers over a number of levels.

The total progression ratio on the other
hand shows how the *total XP* required for a level increases. If this
value is large, it implies that the total XP the player requires for a level
increases significantly as the game progresses. To a reasonable approximation,
this indicates that the rewards for killing monsters (the usual basic source of
XP) increases significantly between levels – with the consequence that players
will progress much faster through the game if they choose to stay “ahead of the
curve” (to fight monsters much tougher than themselves).

We are now ready to examine some specific mathematical choices in the context of XP.

**1. Exponential Progression**

In an exponential progression, the relationship of XP for level (n) and level (n+1) is a fixed number e.g. in the sequence 100, 150, 225, 337 the exponential factor is 1.5.

A quirk of these exponential systems is
that (in general terms) the basic and total progression ratios are the same.
For instance, if the total experience increases by a factor of 1.5 per level,
the basic progression ratio will also work out at 1.5. Similarly, if the
experience difference per level increases by 2.0, the total experience required
will rapidly approximate to 2.0.

The original *Dungeons & Dragons*
mechanics were based on a broadly exponential progression with the exponential
factor changing slightly by class, but being around 2.0. This element of the
rules was lost when Wizards of the Coast/TSR revamped the system in 2000.

Exponential progression might be falling
out of fashion in games design, in part because of the high total progression
ratio it implies. In order to give the player a smooth experience, the rewards
must also rise exponentially as the player’s level increases – often meaning
that the game can be ‘rushed’ by staying ahead of the curve. The extent to
which one can remain ahead of the curve determines how rushed the game can be –
in the worst case, the player can dip into a future area, kill a monster for
phenomenal XP and level up through a large part of the game in seconds. This
problem can be ameliorated by various means, however.

Where exponential progression does survive
in cRPGs, it tends to do so by having a very low exponential factor. For
example, in *RuneScape* (Jagex, 2001) the exponential factor is a
relatively low 1.1 e.g. total XP required = 83, 174, 276, 388, 512, XP differences
= 83, 91, 102, 112, 124. With progression ratios around 1.1, the difficulty curve
of the game is relatively gentle.

**2. Polynomial Progression**

In a polynomial progression, the XP
required for a level is proportional to level^f, where f is the polynomial
factor e.g. 2 for squared, 3 for cubed etc. It is a property of these systems
that both the progression factors decline as level advances, but that total
progression ratios fall more slowly than basic progression factors.

An example of such a system can be found in
the Dreamcast game *Armada* (Metro 3D, 1999) which uses the formula of 8 x
(current level)^3 to determine the XP requirements.

Polynomial systems generally have gradation
in their early levels akin to an exponential progression, and with similar
problems. The difference comes in the effect at later levels. In an exponential
system, the XP values become astronomical – beyond any reasonable expectation
of achievement. But in a polynomial system, the difference actually gets smaller as the
level increases; at higher levels, the basic progression ratio gets closer and
closer to 1.0 (the total progression ratio does the same, but the effect is
slightly less pronounced).

These systems are broadly speaking an improvement over strict exponential progression, but the flattening out of the curve in the upper reaches tends to lead to a flattening out of the progression process – there becomes no appreciable difference between the actions required and time taken to advance a level once the level becomes sufficiently high.

A related problem is that the lowering total progression ratio implies there will be little difference in rewards in fighting stronger foes – in effect, the player would do better to fight weaker foes (which presumably die faster) than stronger ones, as the XP difference between the two will generally not be significant next to the number of foes that must be defeated, and so racking up easy kills becomes a faster means of progression than fighting foes of equivalent level.

**3. Fibonacci Progression**

In a Fibonacci progression the relationship
of XP for level (n) and level (n+1) is the sum of the XP for level (n) and
level (n-1) e.g. 100, 200, 300, 500, 800, 1300, 2100 etc. In such a progression
system, both of the progression ratios are a fixed 1.618034, that is, *the
golden ratio* (also referred to by the Greek character ‘phi’).

I know of no game which uses such a
progression system, but we can speculate about such a game. Given the values of
the ratios are fixed, we can expect that such a game would remain relatively
constant in its rate of progress – the time taken to level up would be directly
proportional to the players choice of relative difficulty. If they stay ahead
of the curve, they will advance more quickly and vice versa. It would, in
short, inherit all the properties of a regular exponential progression
mechanism.

I am forced to conclude that Fibonacci progression, while initially seeming of interest as an alternative to exponential progression is in fact just a special case of it.

**4. Near Arithmetic Progression**

In cases of arithmetic progression (AP), the
relationship of XP for level (n) and level (n+1) is a constant value. This is
never used in practice, because it contains no variability at all, but a closely
related form does occur - what we might call the *near arithmetic progression*
(NAP). In this, the value is not constant, but increases very slightly between
levels. For example, in *World of Warcraft* (Blizzard, 2004) the
differences between level XP values are (for instance) 400, 500, 700, 700, 800,
900, 900, 1100, 1100, 1200 etc - i.e. a near arithmetic progression.

When NAP is used, the progression ratios
begin relatively high but fall gradually to just above 1.0. This may result in
the same problems as polynomial progression in the later stages of such a game –
a problem generally avoided by setting a maximum level (60 for *World of
Warcraft*).

However, arithmetic progressions are much
more forgiving than polynomial progression at lower levels, since their
progression ratios are always relatively low. This creates a much more gentle
curve than exponential or polynomial systems, and this gentle curve is becoming
popular with players because it delivers rewards relatively regularly, and
while advantages can be gained by pushing “ahead of the curve”, they are
generally limited. (In fact, *World of Warcraft* enforces various rules on
the acquisition of XP to limit the players ability to push ahead).

**Conclusion**

An examination of the basic patterns in the
mathematics of XP seems to show that it is not the choice of mathematical
system which is the key issue, but rather the setting of the constant values
that determine the steepness of the experience curve. Whether it is *RuneScape*
with its exponential system with a low factor of 1.1, or *World of Warcraft*
with its gentle near arithmetic progression, the focus seems to be on creating
smooth curves with low progression ratios. These low progression ratios mean that
there is a gentle pressure on the player to advance within the game space, and
that the gaps between levels are retreating only slightly – the next level is
always in sight, encouraging the player to play just that little bit more.

The fact that these progression mechanics can be fiendishly addictive to players – by virtue of creating a reward schedule which trains the player towards constant activity – is perhaps the reason we see so much of their use in modern games, and we can expect that XP systems will be with us for a long time to come.

*Feel free to share your perspective on
the experience systems of cRPGS you have played. Can you think of specific
games in which progression was too slow or too fast? Have you played games
where the struggle to advance eventually became insurmountable? I welcome your views! Those with a more mathematical bent are encouraged to comment on the mathematics, especially if I have made any errors! *

One has to take into account the relation between the XP required to level and the XP given by (for example) killing a foe. If I recall when at work, Sacred has a polynomial system but compensates by relating the XP reward to the relative level of player and foe. It only gives 1 XP if the foe is trivial to kill, making the standard approach of fighting easier foes much less appealing.

Posted by: Peter Crowther | August 08, 2006 at 01:06 PM

Yes, of course - I thought I had said this explicitly, but in fact I had only explained why I didn't use this data, i.e. it's harder to come by. :)

If XP rewards are normalised to level, as described above (and as happens in WoW as well, amongst other games) the shape of the XP curve is the only relevant factor, is it not, because variations in XP are being normalised (i.e. the normalisation system serves solely as a homeostat to rate of progress)?

Or it the case that the strength of the homeostatic effect drowns out any effect from the curve? This would suprise me, but it's certainly a possibility.

Or to put it another way: did the normalisation of XP in Oblivion ruin some people's enjoyment of the game because this effect was a more dominant component in the progression mechanics than the shape of the overall curve?

Anyone have any thoughts?

Posted by: Chris | August 08, 2006 at 05:32 PM

From the way my wife plays it, Oblivion is an exercise in grinding in much the same way as something like WoW. She selects what she does in such a way as to get the most benefit from grinding, but I have (for example) caught her putting a weight on the 'sneak' key and leaving the character close to a sleeping beggar for a few hours in order to grind the Sneak skill to 100%.

Levels appear to be related to improvements in the other skills, so careful control of the rate of acquiring those other skills appears to be a greater contributor to levels than (for example) straight combat.

This appeasr to have increased my wife's enjoyment of the game, as she feels she can "bend the rules" and "beat the game" somewhat in order to level faster - an effect that may be worthy of a post sometime, Chris :-).

Posted by: Peter Crowther | August 08, 2006 at 08:43 PM

Chris, I refraining from comment, because I don't feel I have much to add just at the moment. Hopefully more comments will spark a reaction other than "cool post". I am planing a post on XP/Renown in my system, but it's at least a week away at this point.

As a designer with an anti-grind, anti-level approach, you may well imagine that I aim to buck the tendencies you describe here.

Posted by: Corvus | August 08, 2006 at 09:06 PM

Actually I agree with Corvus. I prefer to think of the "mathematics of progression" in terms of transfinite sets, that is deep patterns in the system, rather than finite numerical progression. What I mean to refer to is the theory of infinite numbers developed by George Cantor in the 1900s, where you can have infinite sets that are greater than other infinite sets. Think of it as the player gradually coming to grok the more percolated patterns, and by adjudicant having a stronger intuitve grasp of the system.

That said, I really think you should go with the Fibonacci approach for Reluctant Hero. I was reading a conversation with Phillip K. Dick about his year-long angel encounters, and at the end the Angel showed him a shape whcih defines 8 and 13, which are the inflexion point on the sequence before it begins its hyberbolic acceleration. Not that that proves its system dyanmical utlity, but its interesting.

Posted by: Patrick | August 08, 2006 at 11:54 PM

Well this post serves as an introduction to the experience system for Reluctant Hero, which I'll post about today, time willing.

Thanks for the comments!

Posted by: Chris | August 09, 2006 at 08:19 AM

While not a designer - merely a player - I also have a distinct distaste for the whole "level" thing. In tabletop RPGs, I tend to prefer the Chaosium and GURPS approach of awarding skill increases or character points. However, I assume levels have the benefit that characters of the same level are roughly as tough as each other, so it's easier to tailor encounters to the ability of the player(s) in a computer game.

Posted by: Peter Crowther | August 09, 2006 at 09:39 AM

Yes, this is precisely the point, Peter. Like you, I don't like class and level in tabletop, but in cRPG design one of the key problems is game balancing which gets very complex if you don't have something central like level to help.

Posted by: Chris | August 09, 2006 at 09:52 AM

Chris, I just wanted to post a thank-you for the existence of this post. I'm designing an experience system and wanted to know if anyone had used a Fibonacci curve. I'm still interested in trying it just to see what happens (our experience system is more important in terms of unlocks and items, not PvP, so we can focus exclusively on the rewards), but this analysis was very helpful.

Posted by: Erin Hoffman | January 05, 2009 at 08:06 PM

Erin: glad to have helped! Let me know how it works out for you.

Posted by: Chris | January 06, 2009 at 09:14 AM

Maybe I haven't spent enough time considering the issue, but I don't understand why advancement of skills, advancement of levels, advancement of enemy ability and advancement of the XP-worth of an enemy are necessarily related.

To avoid the 'problems' you mentioned in the polynomial progression section, the higher levels could give progressively lower attribute boosts. Then the enemies' XP worth could continue to rise, both encouraging players to battle harder enemies and also allowing for a contiually decellerating advancement of skills, which you seemed to want.

Though just a browser game, Kingdom of Loathing has a notable levelling mechanic. No XP exists. Only 3 attributes exist - Muscle, Magic and Mojo (sneakiness/ability to dodge). After a battle, attributes are raised directly. Then, once the class-relevant attribute is raised sufficiently, level increases. The increased level allows access to new skills, but no bonus increases in the 3 attributes. Unfortunately, I can't remember exactly how the levels increase (though suspect it works in a polynomial fashion if I understand the term correctly) and couldn't easily find that info.

Posted by: Behrooz 'Bezman' Shahriari | January 06, 2009 at 04:47 PM

Bezman: I haven't read this in quite a while so I'm not at all sure of the details I wrote. That said...

"I don't understand why advancement of skills, advancement of levels, advancement of enemy ability and advancement of the XP-worth of an enemy are necessarily related."

They're not necessarily related, but they're usually related as follows:

The player advances by killing enemies. The difficulty of this depends on the ratio of player skills vs enemy skills. The rate of progression of player skills depends upon the levelling mechanic (usually). The rate of progression of enemy skills is set by the game design in some manner.

So in effect there is one progress curve, and anything that deviates from the central curve disrupts the basic pattern. This could be faster player skill development, or it could be increasing amelioration of the the XP reward for killing enemies etc.

"Only 3 attributes exist - Muscle, Magic and Mojo (sneakiness/ability to dodge). After a battle, attributes are raised directly. Then, once the class-relevant attribute is raised sufficiently, level increases."

Depending on how the attributes are increased, this is functionally equivalent to an XP system (but arguably more elegant): instead of gaining XP which then raises attributes, you raise the attributes and then this is converted into a level.

I enjoyed Kingdom of Loathing, but I didn't play it for long. :)

Posted by: Chris | January 06, 2009 at 05:46 PM