Resolution Methods
It is fair to say that most roleplaying games do not consider the ontological foundations of the core mechanic that they use - assuming that the system is sufficiently modern to even consider having a core mechanic in the first place. This certainly was not the case in early editions of Advanced Dungeons & Dragons; roll high on a d20 for attack rolls and saving throws, roll low with percentages for thief abilities, etc; these are well elaborated by a 'blog post by The Grumblin' Grognard [1]. Even more contemporary games like GURPS engage in an inconsistency like this, although not to the same degree; it's roll low for skills, roll high for reactions rolls, for example. Most of the time however there is competing desires in the minds of game designers to produce something that is (a) realistic, in the sense for average player-character activity is roughly simulated, (b) playable, insofar it is intuitively grasped for new players of the system along with all its nuances and modifiers, (c) incorporated in the game system in a systematic manner, thus providing both opportunities for additional ease of play through consistency, and for making use of the game's system in an advantageous manner for the player's objectives. In more recent years there have been attempts to bring die rolls as a very up-front part of the game with higher levels of potential manipulation. This is particularly notable in, for example, Greg Stolze's Unknown Armies and the One-Roll Engine.
Several years ago I conducted a quick poll on rpg.net on preferred dice mechanics [2]. At the time, and with the results of the poll, I differentiated between the following: (I) modified target number, roll under/low (e.g., RuneQuest/BRP, Hero System, GURPS, Hero System) 32 votes or 21.62%, (II) target number, roll above/high (e.g., Traveller, Rolemaster, D&D 3.x) 39 votes or 26.35%, (III) target Number, roll-and-keep, raises (Legend of the Five Rings, Seven Seas) 2 votes or 1.35%, (IV) positive and negative dice to base ability rating (FUDGE, FATE) 14 votes or 9.46%, or (V) dice pool with target number per die and degrees of success (e.g., White Wolf, Burning Wheel) 37 votes or 25.00%. There was also a hefty 24 votes or 16.22% for 'other' which was at least partially due to my ignorance at the time of many innovations in the preceeding several years (ORE was commonly cited). A common response in the comments was a desire for knowing the degree by which an action failed or succeeded, rather than a binary success/failure. Whilst this is of course possible in Type I and Type II some posters preferred the explicit results directly from the die, as occurs in Type V. Many expressed approval of the simplicity of Type I systems (Alternity received particular mention for roll under with some degrees of success), and there was evident confusion of result floor/ceiling issues between Type I and Type II systems.
The results were particularly interesting from my own perspective at the two resolution systems that I particularly liked (the positive-negative system of FATE/FUDGE and L5R's Roll-and-Keep) were the two least popular. Of course at least some of this is due to exposure; the dice mechanics represented by roll-under, roll-above, and dice pools, certainly represent close to 90% of market share of roleplaying games. The various implementations of the FATE/FUDGE mechanic were small then and remain so now, although games like "Spirit of the Century" do enjoy a small but enthusiastic following. Whilst the Roll-and-Keep games enjoyed significant backing it by the time the poll was taken this was largely over. Overall, like most "quick-and-dirty" opportunistic polling of this nature, they shouldn't be taken too seriously, with most insight coming from qualitative review of the comments, rather than the raw results themselves. The figure today if such a poll is probably not much different. Newer games such Eclipse Phase add to Type I, Dragon Age's 3d6+ bonuses to add to the Type II, along with D20-derived systems such as Dungeons & Dragons 4th edition and Pathfinder.
Distribution Curves
Of course, this results-based orientation is not the only way to construct a typology of dice mechanics. Another is to refer to the distribution mechanism. This resolves, for example, the debate between floor and ceiling in the "roll high vs roll low" false dichotomy. What is important is the distribution of randomness. GURPS and the BRP-derived games both have a roll-under core mechanic, but the distribution of results is different; the BRP games have a flat distribution, whereas GURPS is somewhat closer to normal distribution (it's not a bell curve in the statistical sense, but it is in a visual layperson's sense). Likewise Dragon Age and D20-type games have a roll-high core mechanic, but when it comes to distribution of die results, the effect is that Dragon Age is closer to GURPS and BRP to D20. Savage Worlds is an interesting variation insofar that it provides two flat distribution curves (at least for PCs and important NPCs). With the combination of open-ended rolls and the wild die for PCs and important NPCs the probability distribution is unusual. In some cases lower die abilities have a higher chance of success than higher die abilities [3]; this is less than optimal. The Savage Worlds mechanic is in some ways a variation on the roll-and-keep method, except (using an elaborated notation scheme) it can be expressed as ((1dx! + 1d6!)k1). For its own part the roll-and-keep games have are increasingly of the bell curve method, depending on the number of dice in the pool.
A particularly notable recent innovation is the dice mechanic used in the third edition of Warhammer Fantasy Roleplay and the Star Wars Edge of Empire by Fantasy Flight Games which utilise proprietary dice. In a sense it is a standard dice-pool system - a player rolls a number of ability plus proficiency dice of the character against with various difficulty and challenge dice set by the GM added to the pool. In addition to this there are setback dice and boost dice based on circumstances problems or benefits. However rather than standard dice numbers, FFG created propriety dice with faces showing various icons representing additional results that are narrated. On the very basic level, a character succeeds if they roll from their pool more successes than failure, however there can be more additional side effects based on the die rolls. In terms of the resolution mechanic is a dice pool (Type V). In terms of a distribution curve, it tends towards a normalised distribution.
There are some partisan attitudes towards the different distribution that models are even more intense than the difference between resolution methods. Probably due to prevalence, the partisanship tends towards either the flat distribution camp and the 3d6 distribution camp. Games like Traveller, which uses 2d6, tend to be ignored by both groups because results were in-between (some call this a "triangular distribution"). It should be pointed out that no dice mechanic can replicate a Gaussian normal distribution, because it is continuous, whereas anything with a finite number of dice will produce a discrete distribution. It is rare to see, for example, anyone arguing for a 10d6 mechanic, although that certainly be would be more normalised than a 3d6 system. In a nutshell, in any ydx system, the size of x represents a discrete degree of randomness and the scale of y represents a tendency towards clustering around a mean. An rather bizarre variation is represented by Cyborg Command with 1d10*1d10, which technically gives a range from 01 to 100, with a heavy clustering down the lower end of the range (it was a roll-under system) and with some numbers simply impossible [4]
In an attempt to be unpopular with both advocates it is argued that flat and normalised distributions have their merits in the right context. In general the better arguments of the two schools of thought can be summarised as follows; the flat distribution has realistic and consistent modifiers and is mechanically easier to play, whereas the bell curve distribution generates realistic results, and provides implicit educational value on probability issues. This said, one should not overstate the playability issue of flat distribution models. Playability is a real issue of course. One of the major problems with the roll-and-keep method, for example, is that the addition of results from a large dice pool is time consuming. Whilst an ideal is that the results are known when the dice hit the table, a little bit of calculation doesn't go astray, such as some fairly simple addition and subtraction isn't too onerous; d20 is quicker than adding 1d6+1d6+1d6, but the latter isn't that bad. Certainly not compared to K.A.B.A.L. where attributes are expressed in percentile range and modifiers are square roots of the stat, whereby situational modifiers are applied to the attribute, meaning a square root calculation every change in circumstances with attack chances based on ratios of these modified results.
Perhaps one of the more earnestly argued positions is that modifiers themselves are also subject to a degree of normal distribution. The argument goes that a person who is good at a skill is more capable of dealing with adverse circumstantial effects, where a person who is poor at a skill is simply not going to get any worse. However this is not what the normalised distribution provides. Normalised distribution (to use GURPS as an example) presents a situation where a skill 18, 12, and 8 ability characters with a modifier of -4, have a modification of -10%, -49%, and -24% respectively to their ability. In other words, the average characters are affected worst of all from circumstantial modifiers whereas, as educationalists will be quick to point out, relatively incompetent characters are seriously affected by circumstances, whereas experienced characters are much less so. This system of diminishing returns is difficult to incorporate in resolution mechanics, and is best expressed in the skill improvement system; Rolemaster was an early example of game that incorporated this into the development system and which is fortunately typical.
Degrees of Randomness and Actual Results
This brings into question of what does the random roll component of a skill actually represent? And what scale of randomness is appropriate? In the Cyberpunk games system, the typical components of a skill test was stat (1-10) plus skill (1-10) plus random (1d10), before adding equipment bonues. For d20, the normal range of stat bonuses plus skill bonuses was around (1-20) plus (1-20) for the skill roll. In other words, randomness contributed around 1/3 to 1/2 as much to results in Cyberpunk as it did in d20. In Amber, as the most well-known diceless RPG, randomness had no effect as all. From a simulationist perspective, the degree of randomness is the circumstances that are affect the result, but are not incorporated as a direct modifier. Depending on the resolution system being used, the gamist perspective will see randomness as making choices less predictable in their outcomes (think "chess with dice", rather than just chess). Something that is typically absent from most RPGs is the recognition that randomness varies according to context, even when the same skills are involved. A world-class sprinter can defeat an average runner on the athletics field, but in a chase down a busy urban street on a rainy night (think Deckard chasing Zhora in Blade Runner) the level of randomness could wreck the chances of the sprinter - if the GM doesn't decide that dodge (with a running augment perhaps?) isn't the most appropriate ability.
Finally, an issue almost universal to roleplaying systems is the confusion between success and skill application and the actual effect of results. Many early games conceptually distinguished the difference by having a use of combat skill rolls, and then a separate roll for the effects (roll "to hit", roll "for damage"). However this was usually limited just to that skill group. For other skills the successful use of the skill represented the success achievement of results and the margin of success represented the degrees of success. In an early playtest of the Mimesis roleplaying game, all skills were differentiated between skill use and skill results. A character could be very good at a particular task, but due to a lack of attributes or equipment, they could have but minimal effect. The Hero system can simulate this sort of difference as well, especially with powers that require a skill roll, common in say, Fantasy Hero where a wizard can have a high magic skill, but still a relatively lacklustre set of powers. The major argument against this distinction really is the use of a second set of die rolls to provide a final result.
The Mimesis Dice Mechanic
Thus, the Mimesis roleplaying game system, having reviewed a wide range of resolution methods, thus seeks to combine the best features of exiting products, avoidance of the worst, and adds some innovations from its own. It takes into account "reality checking", especially from studies in education and learning. On the first instance, the resolution method is consistent throughout the entire game system, from which HeroQuest is particularly influential. The resolution method is orientated towards a roll-high target number with both positive and negative influences, such as in FATE/FUDGE. This provides the sort of discrete normalised bell-curve similar to GURPS etc. However, rather than a flat number of positive and negative influences, the Mimesis system provides for varying degrees of negative circumstances, which can be offset by a varied level of skill from a core attribute level, with modifiers, with the character's willpower acting as a random positive influence of variable value.
Whilst a novel contribution in its own there is also a visceral aesthetic in having such a mechanic that is derived from Earthdawn which made conscious use of different die-types in its resolution sytsem. Results, of the effect of the skill test, are contested against either a static target number or against a variable level against a conscious opponent. The degree of success or failure is determined by comparison and, with a slight influence from Over The Edge, modified by the available equipment or other attribute etc. This is playable, aesthetically pleasing and provides and little bit of implicit educational statistics.
Expressed in standard die notation, the Mimesis RPG system uses attribute + skill/knowledge + spirit dx - circumstances dx, contested.
URL References
[1] http://grumblingrognard.blogspot.com.au/2009/02/what-is-d-old-mechanics-...
[2] http://forum.rpg.net/showthread.php?363034-Task-Resolution-Systems
[3] http://www.daemonstorm.com/role-playing/savage-worlds/Savage-Worlds-Dice... and http://www.insomnihack.com/?p=495
[4] http://anydice.com/program/2bb7