Saturday, October 13, 2012

Save-scumming is Perverse Optimization

XCOM uses a technique that Firaxis also used in Civilization games since Civilization III to prevent save-scumming: the seed used to generate psuedo-random numbers is saved along with your scenario, so when you reload and do the same actions again, the exact same results will occur.

It seems like a strange thing to do, and directly against a meta-game tactic that players may use to achieve desired results in a game. It merits an examination of save-scumming as a result of game design; when does it happen and why? How can we avoid it--and should we even try?

The practical definition of save-scumming is somewhat controversial. I think the safest way to describe it is "using a save-game feature to manipulate the outcome of events by repeating the same passage of play until you get the results you want." (If you think I'm missing a key part of the concept, please let me know in the comments.)

In order to predict save-scumming behavior in a given game, we must come to an understanding of why a player would even want to save-scum. Save-scumming isn't a particularly fun thing to do in a game: it consists of the administrative tasks of managing saves and, in many games, repeatedly sitting at loading screens, and repeatedly doing basically the same few actions in the same context. This is the kind of repetition that I think most gamers would say games should dispense with all together, yet gamers find themselves obliged to do it.

So why would anyone save-scum?

When the consequences of failure in a game are significant and may snowball into large amounts of player-time loss, save-scumming becomes a common behavior if the saving mechanics permit it. This is a natural defense mechanism, and actually fits the original purpose of saving your game, which is to prevent the time-loss caused by having to replay the game from the beginning every time you boot it up.

I argue that save-scumming is a reaction to poor game design. Game designers are responsible for the level of fun optimal play allows their players. If optimal play involves save-scumming, I believe the player experience is usually compromised by optimal play--I don't find the administrative juggling of save-games to be fun. The best games benefit from forcing the player to accept failure and work within the confines it may impose. This isn't easy to do in game design, though, and most mainstream games have no interest in even trying because it's just too potentially dangerous to marketability and accessibility. The Demons' Souls and Dark Souls have had partial but notable success in this area, which may begin to turn the tide.

Players will save-scum when they feel that is the only way to play optimally and prevent what they feel may be excessive time-loss. I would not fault players for this behavior, because it could be made impractical or avoided entirely by alterations in game design. Games should only allow save-scumming when it's designed into the content and systems. Failure can be an enriching experience that does a great job of contextualizing success, though, so I would not advocate designing save-scummy games just because it's easier.

Games that force you to save and reload frequently can seem save-scummy, but frequent saving and loading may be an important design feature in certain kinds of games. Some games are intense tests of the players' ability to execute maneuvers with a low margin of error. Super Meat Boy or I Wanna Be The Guy allow you to keep trying the newer-to-you parts of the game without having to ceaselessly replay hard parts that you've already struggled with and overcome. A save point can be a reward in such a case, and I hesitate to claim that the constant dance of failure and automatic reloading is save-scumming. It seems to fit the design of those games well, and, as such, shouldn't have a negative connotation attached to it.

Limited player agency may also push players to desire save-scumming. If you get a series of bad dice rolls in an RPG that causes some serious consequences, it's understandable to be upset and feel that the game is being unfair. When I play Madden games, I often have a strong desire to quit the game because some ridiculous event occurs during play that is so unrealistic and unpredictable to me that I feel it has compromised the representation of football. Sometimes it's just an emotional reaction to throwing a dumb interception, but most of the time it's from terrible dice rolls or omniscient linebackers who behave as if they can see out of the backs of their helmets. When a game disenfranchises me as a player, I don't have any qualms about reloading a save and trying again. If a game is well-designed, you shouldn't feel the need to reload in this fashion.

But what if you want players to accept the results of randomness and incorporate the variable nature of results into their strategizing? You can build in re-roll mechanics so that players can even take some agency in randomness and don't feel like they have absolutely no recourse against results that are obviously out of their control. You can also place a barrier to save-scumming by doing what XCOM did: store your random number generator seeds with the save-games to make it impossible to scum on that scale. A little push in the right direction can break the spell of save-scumming and let players have fun playing the game as it was intended.

Saving mechanisms in games provide a way to mitigate time-loss, but also let the player do some repetitive result-selection outside of the game mechanics that can lead to optimal play being boring. With careful design, I think we can remove save-scumming from the games we make, and continue to use saving for its intended purpose and not as a perverse optimization tool.


Azuriel said...

The only times where I feel "save-scumming" is worthwhile as a player is when the outcome variance is so ridiculously high. I remember reloading certain saves of Final Fantasy Tactics or Tactics Ogre over and over because an attack with a 10% chance to hit managed to land and completely changed the battlefield single-handedly. I suppose it could be argued that I wanted to avoid excessive time-loss, but on some level, I felt offended, like the computer was "cheating" by going through with an otherwise irrational act (attacking with a 90% miss chance) and somehow succeeding.

Now that I think about it, I wonder if fewer players would feel the urge to save-scumm if designers removed the sort of binary "chance to hit" mechanic. You could still have RNG in terms of damage ranges and the like, and I doubt many people who keep reloading to make sure their attack dealt 150 damage instead of 135. Unless, of course, that 15 damage difference was the margin between an enemy dying or getting another full round of attacks off.

Kdansky said...

@Azuriel: One could use the toHit-percentages differently. Instead of 80% chance of doing 200 damage, just do 80% OF 200 damage (e.g. 160). Arguably, this removes tension. On the other hand, a game should not rely on randomness for tension to begin with. Chess has a lot of tension, and zero randomness. It's not required, it's just the cheap way out. Keith Burgun has written a great article on the topic:

If we're talking about (very) current games, I want to mention Dishonored. It's a bit like Deus Ex, except with the caveat that "kill or disable" is one of the major "choices" in the game. Killing gives a worse score, but it's easier. I've played it on a somewhat fast PC with an SSD. Load times are less than a second. It's neigh impossible to suppress the urge to quickload every time you make a small mistake, go back in time and fix it. But one of the issues is that failing is very painful (you'd have to replay a 10-hour game just to seee the other ending) and not interesting, or even worse, you *must* load the game when you get killed (which happens quite easily).

evizaer said...

Randomness is not always a cheap way out. You're speaking particularly of "input randomness," anyway. Some games are not designed to be played competitively, and, as such, aim to be somewhat forgiving to newer players. They use randomness as a way of leveling the playing field and allowing players of more diverse skill levels to participate meaningfully. Burgun's theory is very narrow in what it allows and encourages, and I don't support his work for that reason. He would blast such a game as Settlers of Catan just because it doesn't have a high degree of competitive integrity; he'd totally ignore what makes it such a popular game that so many players enjoy.

The particular article you linked has numerous serious problems and takes cheap shots at designers of other games--I don't think it's a valuable and positive contribution to game design discussion. I prefer if you wouldn't link to such articles on my site in the future.

Thanks for your comment.

Unknown said...

@evizaer: I think a lot of the problems with save-scumming in games comes down to designers, consciously or not, designing their game around the fact that players can save/load any time.

I've worked (as a QA tester) on a AAA action/adventure series that only allowed saving in between missions. There was a lot of back-and-forth between designers and qa about mid-mission checkpoints - how many should there be and where - in order to properly balance challenge and respect the player's time/effort. Had the game allowed saving at any time, none of that discussion would have taken place, and the games would probably be overall much harder and cheaper (combat not as well balanced). Placing constraints on design usually produces better design.

@Azuriel: I'm with you on that. When I used to play the Fire Emblem series I played it emulated so that I could use save states, as I couldn't bear to lose any units (death = permanent loss for the rest of the game). Eventually I realized I was wasting my time and gave up on the series entirely.

I like the idea of removing to-hit % entirely and just going with a damage range. IIRC, Yggdra Union had a system like that, where it was all about smart unit positioning to maximize damage output and minimize incoming damage. Each battle almost became a puzzle, except you had your own unique set of units and cards that designers couldn't specifically plan for so there was a lot of good desicions needed on the part of the player.

@evizaer: I'm intrigued by your strong reaction to Keith Burgun. I discovered him relatively recently, and agree with a lot of what he says. I'd be curious to hear a more detailed breakdown of how and why you disagree with him, if you could bring yourself to do that.

It's interesting that you mention Settlers of Catan, because that's totally a game I'd save-scum if I were playing a digital version (and if I could bring myself to even play it). My attempts at Catan are chock full of "numbers that are supposedly high percentage never get rolled, or if they do the robber is standing on the resource I would have gotten". In my experience, the only people who claim to enjoy it and want to play it (or similar dice-heavy games like Risk or Monopoly) are those who haven't yet been exposed to newer and better board games that emphasize strategy and skill over output randomness (aka pure luck).

evizaer said...

After trying to have discourse Burgun several times and reading lots of his work (including his book), I stumbled upon this thread and ended up getting extraordinarily frustrated with his inability to write clearly and make sane/coherent arguments, notwithstanding his elitism and condescension. This all was more than enough to convince me that he's probably best ignored.

Unknown said...

Wow, I just spent far longer than I care to admit skimming through that thread. I'm not even going to touch that issue.

One more thought on the topic of save-scumming. You mention XCOM, but you didn't bring up the other anti-save-scumm feature of "Ironman Mode". One save file that autoupdates, no undoes or takebacks ever. I believe lead designer Jake Solomon claimed this is the "real" way to play the game. I think it was a smart move not to make this a default or (god forbid) the only option, but it's great that it's there. It's a great example of a developer having respect for their core audience while still trying to appeal to as many people as possible (they're a business, after all).

I'll admit I save-scummed the hell out of the original X-Com, but for my 3 playthroughs of the new XCOM I've (mostly) limited myself to save files at the very start of each encounter, and I think I'm ready for the Ironman Classic challenge. I'll likely fail, but it should be fun.

evizaer said...

I regret some of my behavior in that thread (and subsequently apologized for it), and I hope you don't think much less of me for the way I behaved.

I didn't mention Iron Man Mode specifically because it was optional. I anticipate that most players don't turn Iron Man Mode on. The seed-saving part is not optional, and the fan reaction to it indicated that people didn't understand why it was done, so I wrote this article about the topic to fully explore it.

Marco said...

Whoa-- Seeding the RNG as a counter to save-scumming? Do you have documentation for this claim? Did designers for XCOM and Civilizations actually come out and say this, or is this your assumption?

As a developer, seeding the RNG is a way to SIGNIFICANTLY simplify saving, replays, and replication. I can go into detail if you'd like.

Also, seeding the RNG doesn't actually stop save-scumming. The player just has to use that number elsewhere to get the next number and test that. They just need to continue throwing "numbers away" until they get something they like.

evizaer said...

>Whoa-- Seeding the RNG as a counter to save-scumming? Do you have documentation for this claim? Did designers for XCOM and Civilizations actually come out and say this, or is this your assumption?

"Preserving random seeds" has been a discussion topic in the Civ community since before Civ III. Civ IV Warlords has a specific reference to the fact that the random seed is preserved--it allows you to turn the feature off in pre-game options.

>As a developer, seeding the RNG is a way to SIGNIFICANTLY simplify saving, replays, and replication. I can go into detail if you'd like.

I know how it works. I'm a developer. I have seen the seed preservation referenced by players and people at Firaxis (don't have the interview saved anywhere, unfortunately) as a way to prevent save-scumming, and it does prevent a certain kind of save scumming.

>Also, seeding the RNG doesn't actually stop save-scumming. The player just has to use that number elsewhere to get the next number and test that. They just need to continue throwing "numbers away" until they get something they like.

The point is that it's a measure to make save-scumming harder. It eliminates a kind of save-scumming. You can still load an earlier save and play from there, making different decisions to avoid the mistakes you made last game.