OSR Rules Families

So, here’s the thing: originally I wrote this like a fucking research article with a hypothesis, a methodology, and all that stuff. I’m not even a scientific research person. That’s my partner’s job. Not mine. So, instead of walking you through every single step I took, I’m going to take the journalistic approach and start with the big picture before I zoom into it and tell you about the little details.

I read, reviewed, and statistically organized 38 different rulebooks considered to be OSR or OSR-adjacent. These include four rulebooks from TSR-era Dungeons & Dragons, as well as ten rulebooks from the 2000s and 2010s. The remaining 24 rulebooks postdate the closure of G+ in early 2019. Please refer to the bibliography at the bottom of this post for more information.

After having collected and organized a dataset with ~90 variables, I found the statistical similarity between each pair of rulebooks based on that data. Finally, I ran an algorithm to determine clusters of these books based on their relative similarity. What I found… won’t shock you. But I hope you will find it interesting anyways!

By statistically analyzing these rulesets and organizing them into clusters, we can figure out what familial resemblances exist between them, as well as what historical developments tie them together. For example, how many games have attacking or saving throws distinct from ability checks? What games don’t make you roll separately for attacking and dealing damage? What character classes appear in which games? How many games have random encounters, or hazard dice, or no random event procedure at all? The fact that we can ask these questions about this dataset indicates that not only are these games similar in their subject matter, but they might all be considered to represent the same “game” with predictable variations.

The Big Picture

The goal of this project was to procedurally organize each ruleset into a “cluster” with other rulesets. Each cluster represents a high degree of relative familiarity between its members, or relative difference from non-members. If a ruleset is included in a certain cluster, that means that it is more similar to the average statistics of its cluster than to the average statistics of another cluster (even if it might be more similar to another specific ruleset from another cluster). Meanwhile, for each new member, the average statistics of the cluster is recalculated. Since the algorithm starts off with a random selection of rulesets, I ran the algorithm multiple times to find the average graph produced by each run. What was the result?

Well, I think it would be unfair to not show you the effect of different numbers of clusters. This is because the algorithm above depends on the user (i.e. me) selecting the number of clusters into which I want to organize the data. I’ll tell you in a second my favorite graph, based on the similarities and differences that I think are most significant, but first I will show you a sequence of graphs, each one having an increased number of clusters. One of the effects you will notice is that clusters do not split evenly. Sometimes rulesets will jump from cluster to cluster. This is because the cluster only keeps track of the average ruleset it contains. The algorithm does not directly compare each ruleset to other rulesets, but each ruleset to the average of each cluster.

Two Clusters

It would be silly if I started off with one cluster, which would just be a singular giant blob across the whole graph. When asked to determine two clusters, the algorithm neatly cuts off what I would consider to be “D&D-Like” from “D&D-Unlike” rulesets. Although the “D&D-Like” rulesets deviate from actual TSR-era D&D in significant ways, they are more similar to D&D than not. They all typically have character classes, character races, the six typical ability categories, rolls to attack opponents, damage varying by weapon, and simple procedures for different stages of play (one-versus-one combat, underworld exploration, overworld exploration, domain play). They also tend to have predefined spells organized by class and level, as opposed to having more interpretive or randomly-generated spells.

This is relative to the rulesets on the left which, as we will see, are extremely varied in their difference from standard D&D.

Three Clusters

This is almost a clean split within the “D&D-Like” cluster. The most faithful (semi-)retroclones are grouped with the ‘authentic’ TSR-era Dungeons & Dragons, and separated from the more modernized or customized D&D-Likes. Meanwhile, two rulesets jump ship from the D&D-Unlike side to the modernized D&D-Like side. This is because the average of that middle cluster has been recalculated. Without including the most faithful D&D-Likes, The Black Hack and A Dungeon Game are more similar to the average ruleset of that cluster than to the average ruleset of the rightmost Non-Derivative cluster.

Now we can talk about what distinguishes the ‘Faithful’ D&D-Like cluster from the ‘Modern’ D&D-Like cluster. The ‘Faithful’ games do not have a universal resolution procedure; they have specific subroutines to handle situations like combat, spellcasting, and avoiding danger. The likelihood of success at these tasks also depends on a character’s class and level, rather than on that character’s individual capabilities (e.g. their strength or intelligence). It is also notable that, except for The Black Hack, these games alone include turning the undead as a class ability for clerics.

Meanwhile, the 'Modern' games greatly emphasize character over class ability. They all have universal resolution procedures based on characters’ ability scores. They almost all replace a literal coin- or pound-based weight measurement scheme with an abstract slot-based one, although more often than not they still retain an encumbrance system where more weight carried corresponds to decreased character speed.

Four Clusters

Finally, we are seeing differentiation on the D&D-Unlike side of the graph. I am going to generalize broadly in saying that the bottom cluster tends to be structured after Into the Odd (2014), whereas the top cluster tends to be structured after Knave (2018). This holds true especially for the latter, which is made up of classless games with d20+bonus resolution systems. The bottom cluster, meanwhile, has a lot more variation within itself. I am going to hold off on discussing this in detail until we have six clusters, at which point I can point out some important and more specific differences.

Five Clusters

This is just the two editions of The Black Hack (2016 & 2018), which are evidently more similar to each other than they are to either the Modern D&D-Derivative cluster or the two rightmost clusters (previously D&D-Unlike). I think this speaks to The Black Hack as almost a transitional species between these two groups. Like many of the post-D&D rulesets, it highly emphasizes individual character ability. It also still has character classes and a spell list organized by class and level. Yet The Black Hack prefigures later, more experimental rulesets by depending entirely on character abilities for all situations and representing armor as a reduction in damage rather than as a modifier of attack likelihood. Over time, rulesets which took influence from The Black Hack seem to have further experimented in their own ways, simply becoming more distinct.

The Black Hack also has some features which are very particular to itself. For example, it does not have any specific dungeon or wilderness procedures, but it asks the referee to check for encounters every 15 minutes of real-life time. It abstracts the consumption of items through ‘usage dice’, where players roll when using an item to see if it randomly depletes [1]. Although these mechanics were popular during the mid-to-late 2010s, when everyone and their grandmother had a special take on The Black Hack, they do not appear in as many of the rulesets from the 2020s which were influenced more by other rulesets.

Six Clusters

A middle cluster emerges between the Knave and Into the Odd clusters, encompassing rulesets that were previously part of either of those previous clusters. As a result of this new split, the rulesets in the Knave cluster become more Knave-like, and those in the Into the Odd cluster become more Odd-like. Now we can more specifically talk about what sets these different clusters apart.

The Odd Hacks

Into the Odd has remained consistently, if not increasingly, influential. The ruleset massively simplified what player-facing rules for a dungeon fantasy game could look like: no classes, three abilities (dexterity, strength, and willpower), and no attack rolls. All you have to do is roll up your character’s abilities, and roll d20 less than or equal to the relevant score when your character is challenged. That is the essence of Into the Odd and its most faithful hacks, even at the expense of not covering other topics of play like exploration or encumbrance or whatever. Yet most of its derivatives do include play procedures for underworld and overworld exploration.

The reduction of ability categories tells us a lot about not only which abilities are considered useful in play, but also which ones are considered categorically distinct enough to include. The three ability scores of Into the Odd seem to align with the most common three character classes: dexterity for thief, strength for fighters, and willpower for magic-users [2]. Gone are constitution, wisdom, intelligence, and charisma; the latter two (or three) seem to be considered the constituent parts of the new ‘willpower’ category, which measures magical capability as well as personal gravitas.

The Knave Hacks

I hesitated at first to name this cluster after Knave since it includes an earlier ruleset by the author of Knave, the v0.3 edition of Maze Rats (which is, notice, very distinct from the v0.1 edition which was a straightforward fantasy clone of Into the Odd). These rulesets are set apart by replacing ability scores altogether with ability bonuses, which are then used in the modifier-based universal resolution procedure. This is a distinct development from the Modern D&D-Likes which, although many used modifier-based rolls, based those modifiers on characters’ ability scores. These rulesets cut out the middle-man, as it were, although some of them require the player to determine “ability defense” scores (10 plus bonus) which NPCs roll against. Oddly enough, this idea does not seem to originate from Knave but from a draft of Into the Odd except that, in the latter, scores are rolled first and bonuses are then calculated as a score minus 10 [3].

This cluster also emphasizes random character generation, not only in terms of ability bonuses but also random equipment, appearance, and personality traits. Even spells are randomly generated and interpreted, being made up of random words which determine the effects of the spell when cast. Rulebooks which derive from Knave often use these tables to provide information about the world’s setting.

Like Into the Odd, Knave does not include procedures for exploration, but its derivatives often do. This seems to be because both Into the Odd and Knave date to a period when there was an expansive culture of play centered on referee-side procedures, such that these rulebooks aimed only to provide a different player-character interface than the standard D&D Basic/Expert rules. The derivatives of both of these rulesets, however, tend to be more cohesive in providing rules for both playing and running the game. This is what ultimately distinguishes, in my opinion, the ultra-light games of the 2010s from their successors in the 2020s. One category aims to simplify the player side of a commonly understood 'game'; the other category wants to resystematize the 'game' altogether.

The Old-School Baroque

This is where the middle cluster comes in. At a first glance, these rulesets have little in common. They are idiosyncratic takes on the D&D formula, synthesizing influences from across the dataset to come up with something uniquely chimeral. Yet some common characteristics emerge: abstract measures of distance (e.g. hexes, rooms, zones); quantum items representing vague supplies or tools, which are transformed into specific items at will; and a more robust system for random events, including resource depletion or personal exhaustion besides just the typical wandering monster encounters [1]. These qualities are not common to all of these rulesets, but you might as well pick two out of three. The overall focus is on abstracting the minutiae of the game in order to focus on procedurally simulating specific activities, from dungeon exploration to hex crawls to long-term domain play.

One trend I have noticed in this cluster is a slightly increased number of ability categories relative to Into the Odd. Rather than strength, dexterity, and willpower, there are: strength, dexterity, intelligence, and charisma. This seems to correspond with a desire to include clerics or cleric-like characters, especially in cases like Errant or His Majesty the Worm where the charisma attribute is strictly related to clerics. Meanwhile, in Into the Odd rulesets, the cleric is rendered obsolete and is encapsulated by the magical attribute or class (if one exists). This development not only points to a resurgence of the cleric, but also to a reinterpretation of the class and its purpose. In D&D-Like games, the cleric is associated with wisdom (perhaps as divine Sophia or something). In Baroque games, the cleric is more associated with charisma, perhaps because of the significance of attracting followers to one’s cause. This in turn corresponds more broadly with the Baroque games’ interest in hirelings and followers. The return of clerics and charisma, now intertwined, thus seems to be overall related to the social or political dimension of Baroque games, which had been lacking in OSR rulesets for a relatively long time.

By the way, “Baroque” is a joke. It’s after the Renaissance and it’s kind of flashy and over-the-top and a reaction against Protestantism (which is too plain and simple). Do you get it?

Seven Clusters

A great split in the Modern D&D-Like cluster! The new bottom cluster is made up of closely-related games; in fact, you can read them from left to right as if to summarize their history. Whitehack is the ancestor of this group, originating a blackjack-style resolution procedure where the player must roll less than their character’s ability number and over some difficulty number. To be precise, however, only combat has blackjack-style rolls, where the top bound is the character’s attacking ability (e.g. from 11 to 20) and the bottom bound is the target’s armor class (e.g. from 1 to 10); the use of blackjack rolls for universal resolution seems to only appear in Errant, which can be found in the Baroque cluster described above.

These rulesets, however, have variable numbers of classes. Whitehack has abstracted or genericized substitutes for fighters, magic-users, and thieves, written to be more broadly applicable than their representations in fantasy rulesets typically are. The Vanilla Game has only fighters and magic-users. A Dungeon Game has no classes whatsoever. The Vanilla Game splits hit points into two resources, grit and flesh, representing luck and stamina versus actual flesh and bone. A Dungeon Game, however, has typical hit points. As per usual, on an abstract mathematical level these games were grouped because they are more similar to each other than they are different to everyone else. It just happens that these rulesets are similar and even related in ways that the dataset cannot detect. I think if you took the average of those two games, you would get something that ultimately resembled Whitehack.

What broadly distinguishes the top cluster, besides the lack of blackjack rolls or asking players to roll high and not low, is the lack of class-determined scores (or bonuses) for saving throws, and perhaps (though less so) for attacking as well. They also tend to have skill systems for some reason. Dungeon Crawl Classics and Into the Depths emerge as outliers in that both still have class- or level-determined likelihoods of attacking an opponent or saving oneself from danger; in fact, Into the Depths does not have ability scores at all. I think this cluster, although it seems to be aptly placed in between the most faithful D&D-Likes and the D&D-Unlikes, is ultimately made up of rulesets which miscellaneously deviate from the standard D&D rules. It might as well just be the “not-Whitehack” cluster.

Past this point, increasing differentiation is no longer helpful. For example, the hacks of Into the Odd split into ones with or without exploration procedures, and then Mork Borg jumps from the Knave hack cluster into the Odd cluster because it too does not have exploration procedures. FLEE moves into a cluster of its own because it’s that special (hi Emmy!). I am exhausted thinking about all this.

Conclusion

This was a really fun exercise because it seems to have revealed how OSR-style rulesets have developed over time, especially with respect to prioritizing class versus character capabilities. My favorite grouping was the one with six clusters because I think it picked up on enough differences between groups to be able to draw conclusions about each of them, without getting bogged down in differences which are (in my opinion) almost semantic. Like, for each of the groups it determined, I was like “Yeah! That makes sense!” and not “This is going to be annoying to explain.” So, let’s summarize those six, in total:

• Faithful D&D-Likes: The most ‘conservative’ of the dataset, being closest to the TSR-era D&D rulebooks. They are characterized by class-based distinctions between characters, and a lack of dependence on individual character abilities. Procedures for underworld and wilderness exploration are simple.
• Modern D&D-Likes: Similar to the Faithful D&D-Likes, except that character abilities tend to be more important than class-based ones (compare to D&D Third Edition and early d20 System retroclones). These have universal resolution procedures. It includes the Whitehack subfamily which is distinguished by having blackjack-style rolls.
• The Black Hacks: This cluster is populated entirely by the two editions of The Black Hack. The ruleset is notable for its simple math, dependence on ability scores, abstract item usage, and real-time encounter checks. The graph does not illustrate how truly influential The Black Hack was across the scene, with there being many hacks and variations in the 2010s.
• The Odd Hacks: Modeled after Into the Odd, a classless ruleset which emphasizes character ability scores and removes attack rolls (in favor of only rolling for damage). The most faithful derivatives only include rules for character interactions, rather than for play procedure or anything else. However, the cluster includes the Mausritter-derived subfamily which does include play procedures.
• The Knave Hacks: Similar to Into the Odd, Knave has no classes, emphasizes individual character ability, and lacks play procedures. However, it retains to-hit rolls separate from damage rolls, and emphasizes random generation for characters, spells, and aspects of the game-world.
• The Old-School Baroque: Although conceptually similar to the D&D-Likes, these rulesets each have their own idiosyncratic approach. They are influenced by other rulesets across the board, especially from the rules-light rulebooks, but are distinguished by a renewed interest in play procedures.

Yet something important to keep in mind is that these categories are not ultimately expressive of each individual ruleset. Categories in themselves are abstractions which elide qualities of things not considered by those categories. We have seen how much variation there is within each of these groups, like the splitting of the Modern D&D-Likes or the expansive hacks of Into the Odd. Moreover, the statistical analysis of each of these texts has reduced them into (literally) a bunch of ones and zeroes. This does not tell you anything about how well-written the text is, what varieties of characters exist, or what toys you can play with. Even on a formal level, the data matrix does not and cannot tell you specifically what defines each individual ruleset.

Gee, I wonder what else that all sounds like.

Anyway, on the other hand, the fact that these rulesets can be directly compared given the same set of criteria indicates that they operate within the same tradition, pull from the same sources, and influence each other. I feel as though there are a handful of rulesets here which, if you randomly generated permutations of them, would eventually result in all of the other rulesets. This is not necessarily a call for innovation in rulesets because it seems that everyone is quite happy playing just about the same game. However, with the increasing production rate and homogeneity of rulebooks, I think it would benefit everyone to become aware of the creative commons at our fingertips—and to produce new things.

Here is a link to my dataset. Have fun with that! I may write about my methodology later [4], or more specifically about emerging trends in rulesets, but this post is already kind of a long one.

Updates

I have posted these in reverse-chronological order, with the most recent at the top.

Update 3: I have posted a second blog post with an FAQ, a methodology, and updated results! Here is a link.

Update 2: I weighed the group variables wrong! Rather than dividing each of them by the number of elements in the set (e.g. if a game has fighters and mages, divide both of their 1-values by 2), I should have set each one equal to the square root of that percentage. This way, the total value of the category equals 1 unless all values are 0, in which case it's 0. Thankfully, this does not really impact the shape of the graph. It only heightens the differences already present, especially relating to classes and abilities. For example, the cluster for The Black Hack is a lot more tightly-knit. That's all! I have updated the datasheet with these changes. Below is a single screenshot of the 7-cluster step, to avoid crowding up this post. It also now includes Macchiato Monsters!

Update 1: Below is a slightly revised graph after correcting some mistakes in the data! I've also incorporated the actual The Black Hack, Second Edition alongside the booklet version (now indicated as TBH1.5). Let me show you each of the steps again!

Endnotes

[1] B., Marcia. 2022-07-26 “Usage & Hazard Dice: How to Emulate Bookkeeping with Dice”, Traverse Fantasy.

[2] In rulesets with three classes, it is more likely for there to be a thief than a cleric! Meanwhile, for any classed ruleset, the fighter and magic-user are always included.

[3] McDowall, Chris. 2011-08-30. “Project Odd: Abilities and Saves”, Bastionland. I seriously feel that this blog post is the culmination of character-centric developments in classic/old-school games, and the basis of almost everything that came after it. Knave may have been the wiser for shifting the focus from scores to bonuses, although I think it makes for a slightly more awkward dice roll. If you asked me, this is a fine opportunity to replace random ability determination with simple point assignment, or random determination within stricter bounds; for example, roll d6 for one out of six permutations of {+1, +2, +3}. Random generation is usually desirable because it’s fast but it seems like determining each ability individually is holding it back.

[4] Especially to discuss certain interpretive choices, such as counting some nominally classed rulesets as classless due to their classes being more like abilities or backgrounds than serving to organize characters into discrete categories. One noticeable trend is that as individual character abilities become more important, the role of classes is less to group characters together than it is to offer packages of unique character features. The classed games I have counted as classless do neither of these things, but this is (in my opinion) a good development.

Bibliography

Organized by year, and then by author. Acronyms used on the datasheet are indicated in parentheses after the title. Thank you to the authors who gave me a sneak peek of their work!

• Arneson, David & Gary Gygax. 1974. Dungeons & Dragons (D&D74).
• Gygax, Gary. 1977-9. Advanced Dungeons & Dragons (AD&D).
• Cook, David. 1981. Dungeons & Dragons: Expert Rules (D&D81).
• Moldvay, Tom. 1981. Dungeons & Dragons, Basic Rules (D&D81).
• Mentzer, Frank. 1983. Dungeons & Dragons, Set 1: Basic Rules (D&D83).
• Mentzer, Frank. 1983. Dungeons & Dragons, Set 2: Expert Rules (D&D83).
• Gonnerman, Chris. 2006. Basic Fantasy Role-Playing Game (BFRPG).
• Proctor, Daniel. 2007. Labyrinth Lord (LL).
• Raggi IV, James Edward. 2011. Lamentations of the Flame Princess (LotFP).
• Goodman, Joseph. 2012. Dungeon Crawl Classics (DCC).
• McDowall, Chris. 2014. Into the Odd (ITO).
• Mehrstram, Christian. 2015. Whitehack, Second Edition (WH2).
• Milton, Ben. 2015. Maze Rats, v0.1 (MR0.1).
• B., John. 2016. Into the Depths (ITD).
  
• Black, David. 2016. The Black Hack (TBH1).
• Milton, Ben. 2016. Maze Rats, v0.3 (MR0.3).
• Black, David. 2018. The Black Hack Booklet, Second Edition (TBH1.5).
• Black, David. 2018. The Black Hack, Second Edition (TBH2).
• Milton, Ben. 2018. Knave.
• Nieudan, Eric. 2018. Macchiato Monsters (MM).
  
• Nilsson, Pelle. 2019. Mörk Borg.
• (Skerples). 2019. Many Rats on Sticks, v2 (MROS).
• Treme, Nate. 2019. Tunnel Goons.
• Gal, Yochai. 2020. Cairn.
• Hunt, Leo. 2020. Vaults of Vaarn, Issues 1-3 (VOV).
• Mehrstram, Christian. 2020. Whitehack, Third Edition (WH3).
• Williams, Isaac. 2020. Mausritter: Expanded Edition (Mausritter).
• Anderson, Micah. 2021. Bastards.
• Boven, Emiel. 2021. DURF.
• Crawford, Kevin. 2021. Worlds Without Number (WWN).
• (CavernsOfHeresy). 2021. Rogueland.
• Rose, Noora. 2021. Unconquered.
• Sinclair, Jared. 2021. The Vanilla Game (TVG).
• Surles, Reese R. 2021. Crowns.
  
• Bisette, Chris. 2022. A Dungeon Game (Dungeon Game).
• Boven, Emiel. 2022. The Electrum Archive, Issue 1 (TEA).
• Hartranft, Tobias. 2022. Trespasser: Dark Fantasy Tactics (Trespasser).
• Islam, Ava. 2022. Errant.
• Linderum, Markus. 2022. Down We Go (DWG).
• Verte, Emmy. 2022. FLEE.
• McCroo, Joshua. (Unreleased). His Majesty the Worm (HMTW).
• Smith, W.F. (Unreleased). Prismatic Wasteland (PW).

I wanted to include the games below, but they were distinct enough that including them in the dataset felt misrepresentative. Congratulations on being different!

• Browning, Oz. 2021. OZR.
• McCoy, Sean. 2018. Mothership.
• Allen, Emily. 2021. Dungeon Bitches.
  
• Nogueira, Diogo. 2022. Primal Quest.
• Rolim, Lucas. 2022. MiniBX.