Savage Worlds

When I mentioned to a friend that I’d been looking at D&D 4th Edition, he mentioned he’d bought a copy of Savage Worlds (from Pinnacle Entertainment Group) which was — shock, horror — skill-based. There’s a fully playable free version of Savage Worlds available for download on the website — I recommend it to you. The thing which I really like about Savage Worlds is that — like ForeSight — it’s designed for shorthand, and you can play with “all the rules you can remember” and whatever’s written on character sheets. And, unlike ForeSight, you don’t need to remember quite so many rules.

Savage Worlds is one whole level more abstract than ForeSight, making it a lot simpler, and on top of that it’s designed to reduce pretty much everything (explicitly including settings and adventures) to bullet points. It’s easy to imagine you could design a character in five minutes, and prepare to run a module in ten. Here’s the nutshell version:

Attributes and Skills are represented by a die type, e.g. D4 is the worst, D6 is average, and so on up to D12 (this effectively gives a range of five possible values). All resolution rolls are open-ended — if you roll maximum, roll again and add on. So if you have a D4 ability you can roll 4, 4, 2 and get 10. A typical task requires a 4 to succeed, which means a D6 ability has a 50/50 chance of success. Each attribute covers a number of skills, and it’s cheap to buy a skill up to the attribute, and then more expensive to go beyond. So if Shooting is based on Agility and your agility is D8, then it’s cheap to buy Shooting up to D8, but gets more expensive to get it to D10 or whatever. Degree of success is measured by the number of 4’s by which you succeed.

Powers represent supernatural (or perhaps just zany) elements, they draw from a points pool (you start with a 10 point pool and get back one point per hour), and they’re abstracted out from settings. So a power called “blast” might allow you to fire off magic missiles, cones of fire, and fireballs (yes, powers are quite versatile) — or psionic stabs, psychic waves, and telekinetic explosions. What a power does is covered by rules, how it’s flavored is a matter of setting, GM whim, etc.

Edges and hindrances are Savage Worlds‘s equivalent of Hero System/GURPS Advantages and Disadvantages or Fallout‘s Perks. In general, I like this system better than any similar system I’ve seen in a paper game (although Fallout’s Perks remain supreme), although many of the hindrances are things I think a player should take on voluntarily rather than be bribed to do. What I like about Fallout’s Perks is that the “bummer” is generally tied to the “goody” in a sensible way (“I’m a gifted healer but I hesitate to inflict harm on others”), versus the “I’ll be afraid of spiders so I can buy expert marksman” tradeoffs you inevitably see in these systems.

Oh, and as in any sane system, armor blocks damage, skillful people are harder to hit.

From the game master’s point of view, things like hit points are pretty much abstracted out. Unless an NPC is major (e.g. the main villain or a named henchman), he/she can be alive, “shaken”, or out of the action (wounded, incapacitated, dead — doesn’t matter). An ability called toughness determines how hard it is to one-shot an “extra”, and how likely they are to recover from (or expire from) being shaken. Major characters (“Wild Cards”) only get “wounded” when other folks would be eliminated — each “wound” inflicts a -1 modifier to all actions. A typical NPC is all D6s except for anything requiring special attention.

And finally, players and the DM get “bennies” — the equivalent of James Bond 007‘s “Hero Points” — which allow extra rolls in tight spots (you always use the better roll). Each player gets three bennies per session (they are lost if not spent), and the DM gets one for each player.

How does it compare to D&D4?

Unlike any version of D&D, Savage Worlds is a game that lets you think up the character you want to play, build that character, and play it right now (modulo perhaps being more or less capable). Relative to D&D4, the amount of book-keeping required (by players or the GM) is hugely reduced. And the designers of Savage Worlds are aware of the criticality of delivering key information when it’s needed, versus the information overload in every aspect of D&D4.

Probably my only real misgiving about Savage Worlds is the rather crude resolution system, but it’s better to have a crude resolution system that’s at least pointing in the right direction, than a finer-grained system that’s simply wrong (e.g. AC makes you harder to hit).

Best of all, Savage Worlds is designed to fit into your setting, and be complete by and of itself, versus D&D4 which imposes its will on your setting (unless you’re willing to do a humungous amount of work) and requires you to buy a ridiculous number of over-priced rulebooks.

How does it compare to ForeSight?

ForeSight is a much more complex game than Savage Worlds, and it’s intended to be more “realistic”. Aside from that, the two games are quite similar in terms of the way they bind to narrative (the level of abstraction). There are some things in Savage Worlds that I would steal for the next edition of ForeSight — the general-purpose power framework is absolutely brilliant, for example, and the way edges and hindrances work is pretty good — and while I think that any “goodies and bummers” system is open to massive abuse by rules lawyers, Savage Worlds is not a game that really lends itself to rules lawyering.

Whenever I see a game that’s fundamentally well thought out and much simpler than ForeSight it makes we wonder what the complexity in ForeSight really buys me. One interesting point made in the article in which the designer describes how he went about designing Savage Worlds is that he was annoyed by the slow pace of combat in every game system he’d used, citing D20 system in particular. A fight involving 20 NPCs might easily take a couple of hours (I’ve personally been stuck in D&D fights that occupied an entire night, and they weren’t terribly interesting fights either). One area where ForeSight excels is speed. Even large, complex fights in ForeSight tend to be over with pretty quickly. (One reason for this is that ForeSight was designed very much for people who can’t be bothered with miniatures.)

Savage Worlds is way ahead of ForeSight in terms of its streamlining. There’s no question that ForeSight can express a much more detailed and complex thing than Savage Worlds — everything is described both in more terms and in a finer-grained way. But there’s a huge difference between a game that old married people play and games than college students and teenagers play. We don’t need fine-grained item definitions and skill progression if we’re only going to manage one gaming session every two months — all this detail is really only useful if you’re gaming often enough to care.

Finally, the rough-and-ready nature of Savage Worlds‘s game mechanics lends itself very well to a less-than-serious settings. ForeSight‘s detail and expressiveness lend themselves to more serious settings. In this sense, the two game systems complement each other.

What would I do differently?

Simply put, I’d replace the resolution system and then rescale everything to match the replacement system.

Because the system is so reliant on the crazy polyhedral dice, it doesn’t have a very large sliding window of probability to play with. How can you represent — for example — differences within similar populations with a large gap between them (e.g. normal people and superheroes?). What’s more, open-ended die rolls are a bit of a mess when it comes to analysis, and it’s even worse when each different kind of die has a different chance of an add-on roll. Also I hate D4s. They simply don’t roll.

But there’s a much more important and — unfortunately — lethal argument against using the existing system: it’s hard to make things that are hard for idiots to do but reasonably routine for competent people to do. Let’s suppose we decide that a computer system is pretty tough to hack, and give it a difficulty of 7. Then a “world-class” (D12) hacker has a 1/2 chance of success (less, presumably, if the hacker is getting a blowjob at the time…), while a complete idiot (D4) hacker has a 1/8 chance of success. If I wanted to give a D8 hacker a 50% chance of success I’d need to make the difficulty 5, which is only one notch above “typical”, and this gives a moron a 25% chance of success. This makes it virtually impossible to create a challenge that will probably be handled by a “pretty competent” character (D8 or D10) but almost certainly not by a gumby — yet this is the exact kind of challenge you need for adventure stories, otherwise why be good at anything?

Note: while revisiting this post I noticed there are also some pretty bad “poverty traps” (situations in which having more makes you worse off) in the resolution system. E.g. a D4 skill has a better chance of performing a difficulty 6 task (3/16) than a D6 skill (1/6). The fact that higher skill dice are less likely to reroll also means that having higher skills isn’t much of an advantage when attempting truly difficult tasks (e.g. the difference between attempting a difficulty 25 task with skill D12 vs. D10 is minuscule (0.64% vs 0.6%), whereas the difference for a difficulty 10 task is enormous (25% vs. 10%)). This is the opposite of what one would expect and is likely to make the game feel wrong.

I think that a better alternative would be to base everything on D10s (my favorite kind of dice — they roll better than D6s or D8s and, unlike D20s, can actually stop rolling on carpet) and use simple additive modifiers. E.g. attributes could range from 3 to 10 or more (i.e. ForeSight’s range), and be added to D10 resolution rolls. You could leave rolls open-ended, but you’d only have one lumpy distribution to worry about. It would follow from such a system that if (say) 2 is average for normal folk, and 4 is average for heroes, that a roll of 10 means success (a really nice round number :-).), a normal person succeeds 30% of the time, and a hero 50% of the time (very much like the rules as they are). This gives you twice as much “grain” to work with with respect to weapon damage, situational modifiers, character advancement, and the like, with no additional complexity. As for raises — make them 5 (another nice round number).

If we return to our hacker example, we give the system a difficulty of 12, say, which gives our “world class” (+10) hacker a 90% chance of success, and our idiot (+1) hacker a 0% chance of success and a reasonably competent (+6) hacker a 50% chance of success. Now this is starting to sound useful.

Bare success and failure are very useful concepts (dramatically!), and can be implemented by making any success where the roll was exactly the number needed a bare success, and “missed by one” a bare failure. Each is guaranteed to be something that happens 10% of the time at most, which is very nice.

The system of “raises” in Savage Worlds serves the role of “crits”, but there’s a question as to the (very lumpy) distribution. E.g. a D8 skill vs. difficulty 4 has exactly a 1/8 chance of achieving 1 or more raises, and in fact has a 3/64 chance of getting one raise, a 4/64 chance of getting two raises, and a 1/64 chance of getting three or more raises. This is just weird. Indeed, based on whether you can get a raise without a reroll, the distribution of probabilities fluctuates wildly. E.g. D10 vs. 4 has a 21% chance of one raise, a 4% chance of two raises, a 3% chance of getting three raises, and so on — so a D8 vs difficulty 4 has a better chance of two raises than D10 vs difficulty 4. This is just one example — open-ended die rolls will do this everywhere.

My alternative: going back to the D10 resolution system, if you roll a (solid) success, you roll for success again at -2 (cumulative) to get a “raise”. So if you’re a decent (6) hacker working on a pretty tough (10) network, you have 20% chance of failure, a 10% chance of a “bare failure”, 10% chance of a “bare success”, 60% chance of a solid success (which has a 40% chance of getting a raise). I haven’t put all this into a spreadsheet, but it looks to me like a system with legs, although the -2 modifier may need some fiddling.

Simple example: our +6 hacker vs. the difficulty 10 system has a 30% chance of failure, a 10% chance of a bare success, a 30% chance of a solid success, a 21% chance of one raise, and a 9% chance of two or more raises. (I’m assuming a bare success on a roll to “raise” terminates.) Increase the difficulty to 12 and the hacker has a 50% chance of failure, 10% chance of bare success, 28% chance of solid success, a roughly 11% chance of one raise, and a 1% chance of two or more raises. Now increase the difficulty to 15 and we have a 10% chance of bare success, 9% chance of a solid success, and a 1% chance of one or more raises. As you can see we have a pretty nice “rolling window” of probability (pretty easy to very hard is a range of six) with nice probability distributions all round (nicer than ForeSight’s, I’d say).

Critical failure is also a very useful concept, and here we do the same thing in reverse: if you fail solidly you roll again to succeed at +2 (cumulative), and so forth. The opposite of a “raise” we can call a “deuce”, and the more deuces the more horrible the failure. Since this is exactly the “dual” of critical success, we know that if we get the first mechanic to work just right, this one will too.

Next there’s the whole problem of open-ended die rolls. Consider for a moment the D6 skill attempting the difficulty 6 task. Success chance is 1/6. What’s the success chance against difficulty 7? Oh, it’s 1/6. What’s the success chance against difficulty 8? It’s 5/36 (a shade less than 1/6). And so on. If you draw the graph of this it’s pretty diabolical, and it means that if you’re shooting at a typical (difficulty 4) target while running (-2) you might as well use your off-hand (-4). If you want everyone to act like complete idiots — and you might — then this is exactly the resolution system you should use. But I’m guessing it’s not exactly what the designers intended. From a lot of points of view, open-ended die rolls are cute — they generate excitement, and extend the “long tail” of the resolution system, allowing people to do amazing things occasionally (hmm, not that occasionally). But it’s bad enough when there’s a 1/10 chance of zaniness every time you roll the dice (i.e. the D20 system where a 1 or 20 has special effects) — it’s going to be a lot worse if it’s 1/6.

Returning to the D10 replacement rule — a simple option is to replace any D10 roll with a new roll at +5 — if the player wants it. (If you’ve already succeeded, there’s no point.) Since this has exactly a 50% chance of exceeding the original roll it guarantees that the long tail will thin out (albeit rather chunkily). So if you think of our (6) hacker, the success probabilities as the system gets harder from 12 are 50%, 40%, 30%, 20%, 10%, 5%, 4%, 3%, 2%, 1%, 0.5% and so on. I think you’ll agree that this is enormously preferable.

With this “live organ transplant” I think Savage Worlds, or something a lot like it, could be a very solid system. And you won’t need any of those stupid D4s, D8s, and D12s.

The way skills have been defined is a wee bit silly. E.g. one skill — shooting — covers all forms of ranged combat, while another — fighting — covers all forms of melee combat, but there are three vehicle skills (driving, boating, piloting) and riding. This means to be Death On Wheels you only need two skills, but to be The Driver you need four.

I’m also a bit put off by the choice of attributes. Basically you get Strength, Smarts, Agility — fine so far — then Vigor and Spirit. Vigor is essentially used for avoiding from wounds — no skills are driven by it. Now, not dying is pretty important, but that’s kind of lame. Spirit seems to be Willpower (it drives guts, intimidation, and persuasion). OK, I’m fine with having a willpower attribute, but if you’re going to have five stats, I’d probably pick something other than “vigor” to be the fifth. I think I’d probably split combat skills into Unarmed, Melee, Shooting, and Archery.

I’d have four attributes: physique, coordination, intellect, and — ok, let’s call it spirit. Physique combines “strength” and “vigor”, Coordination combines “dexterity” and “agility”, Intellect combines “intelligence” and “perception”, and Spirit combines willpower and charm. Then you can use perks/edges to tweak the balance (make someone unusually strong vs. healthy, or willful vs. charming). There’s a symmetry to this, with there being — in essence — four attributes, physical and intellectual, power and finesse — with the finesse attributes tending to drive skills and the power attributes tending to drive output (provide mana pools, prevent you from dying, etc.). This also further streamlines sketching out NPCs (now you only need to worry about four attributes).

Finally, I have a (fairly minor) quibble with the way experience points are spent: characters are given an overall Rank (ranging from Novice through to Legendary) based on how much experience they have, which is in essence their point cost. But because of the way attributes and skills interact, it’s possible that a character who bought skills early and attributes later will be less capable with a given amount of experience when compared with a character who bought attributes first and skills later. The obvious solution would be to refund skill points when an attribute is improved, or reduce the character’s “seniority” by the “lost” skill points. Either is kind of clumsy. If attributes simply limited skill advancement (e.g. you can’t raise a skill more than one rank beyond its attribute), as in ForeSight (or James Bond 007 from which the mechanic was taken), this problem would not occur.


I really like almost everything in Savage Worlds — if only the resolution system weren’t broken.  But at least it’s pointing in the right direction. I think I’ll try to design a simple RPG along the same lines and publish it as a short pamphlet (much in the spirit of the very first version of ForeSight).

Post Script

I’ve done a lot of thinking and experimenting with the “live organ transplant” described above and come to two important conclusions: rerolling at -2 (or any such scheme) for criticals produces an absolutely terrible distribution); and the kind of open-ended roll which leaves holes in the distribution (e.g. on a D10 roll you can’t roll 10) is The Devil.

So I’ve come up with a new open-ended D10 roll (it works, by extension, for any kind of die) — on a “natural 10” you reroll at +5 and replace the original roll if the new roll is higher (and this is recursive, so if you roll 10 and then 10 + 5 the second 10 is rerolled at +5 and replaced if higher and so on). 1s go in the opposite direction the same way. This produces a pretty nice* distribution (e.g. the probability of rolling 9 is 10%; 10 is 5%, 11, 12, 13, and 14 are 1%, 15 is 0.5%, and so forth — vs. Savage Worlds’s system which is: 10% for 9, 0 for 10, 1% for 11, and so on).

Because this distribution has no lumps and groups probabilities, a “success by 5s” critical system doesn’t do anything horrible (indeed the probability of each extra level of success is one tenth the previous). In fact, the whole system appears to produce a nicer set of outcomes than James Bond 007‘s Ease Factor / Quality Rating system (of which ForeSight uses a modified form) and has the advantage of being completely open ended (so superheroes work).

The D6 version works just as nicely, but has the expected issue of being cruder in the middle. I’d suggest that open-ended dice (as described here) along with “success/failure by N” (where N is half the size of the die) is a very satisfactory resolution system for tabletop gaming, and you simply need to match the die size to the level of detail you desire, although — frankly — 3D6 is a nicer distribution if you don’t need a large or open-ended window for modifiers. The probability tails will always go 1/N, N/(2N^2), 1/N^2 … at the edges which is pretty darn nice.

As regards the D20 system, making D20s open-ended in this way, and then using “success by N” the criterion for critical hits would probably be a much better alternative to “any 20 that isn’t a miss is a crit”. (When fighting people who can barely hit you it’s not exactly comforting to know that if you do get hit it will be a crit.)