Electron Dance

I Hate Playing With Myself

This is the fourth part of The Ouroboros Sequence, a series on puzzle games.

In 2008, a recommendation on Rock Paper Shotgun led me to Cursor*10 (nekogames, 2008) in which the player has to make it to the 16th floor - but the player's life only lasts a short time. With each new life, the player is accompanied by the ghosts of their previous incarnations, working side by side to reach a common goal. If you need to click a box 100 times, it’s a damn sight quicker if a previous life is there to assist you with the clicking. It was the first time I’d seen this sort of mechanic, but unlikely to be a world first: Braid (Number None, 2008) released later in the same year utilised a similar mechanic.

But I’ve seen this design pattern again and again over the years in puzzle games. Today my mission is to explain why I hate it.

While some games argue this is “time travel” puzzling I’m christening these "playback" puzzles based on how The Talos Principle (Croteam, 2014) more honestly presents them. In Talos, you turn on a device which records your actions which then plays them back through a ghost duplicate. The playback ghost will operate as if you were working with another player to solve the problem, making it local co-op for one. Can’t press two buttons at the same time? You could if there were two of you. Can’t reach that ledge? You could if you could give yourself a boost. Talos goes a little further than most because your ghost self has ghost copies of blocks and fans - which means Talos playback puzzles are more about doubling resources than just about being in the right place at the right time.

The Talos Principle: Playback ghost

Rose and Time (Sophie Houlden, 2010) is an interesting riff on the playback puzzle, which recalls those science fiction tales where protagonists are worried about “crossing their own timeline” and upsetting what they remember seeing - an idea stretched to delightfully absurd lengths in Spanish thriller Los Cronocrímenes/Timecrimes (Nacho Vigalondo, 2007). In Rose and Time, the player must retrieve crystals from each level - picking up a crystal will send you back in time and then you'll have to avoid being seen by your previous incarnation so as not to cause a paradox. After just a few levels, you are juggling multiple versions of the protagonist and quickly becomes difficult. While the initial inspiration was a "stealth game" where you have to hide from your past self, that's not really how it comes across: my experience with Rose and Time suggests it is more about forward planning, creating the ability to hide, rather than purely dodging your old self. Mistakes at the start can prevent solution.

Rose and Time

More recently, Induction (Bryan Gale, 2017) is a minimalist playback game which takes the time travel theme and allows you to mess around with the past, provided you don't ruin history completely. (I guess you can shoot Hitler, just as long as you don't kill him.) There are the usual elements - buttons to trigger, little barrels to push around - but the only way to get through each level is to go back in time and build on what you previously did. You can change what happened but you cannot create an outright paradox: your old self must still be able to travel back in time at the same place you originally did.


Now during my Talos days, every time a playback machine appeared I would groan and inevitably go check out other puzzles until I had no choice but to come back. At the time I just assumed my groan was due to playback puzzles being harder as they multiplied the possibilities at your disposal. But, over time, I came to an understanding that these puzzles annoyed me for very particular reasons.

When you’re working through a recording phase, you need to think about where your future self should be and how much time your future self needs to do things. If you move too quickly, you’ll deny yourself the time to complete certain actions or tolerate small mistakes. This imprecision often drives players to build a cavernous safety margin in which the future self gets to do their thing. So you put your block here then wait, I dunno, maybe 20 seconds? Then you pick up the block and move it somewhere else and wait, maybe, another 20 seconds. Here, in fact, is where Induction excels as it allows you freedom to not just rewind but also fast forward through waiting.

Rose and Time players also have the power to rewind the action whereas in Talos making one mistake with your recorded run means you’ll have to do the whole thing again. However, while rewinding is useful for small errors, these puzzles are always a Jenga tower of cause and effect and making a mistake at the base of the sequence means the whole thing will come crashing down. Rewinding is not going to fix that.


Action games seem a better fit when your past incarnations become your enemy - consider the superb platformer N++ (Metanet Software, 2017) in which multiple copies of the player can be in hot pursuit. Like Rose and Time, early access Echoplex (Output Games) appears to be a hybrid - where you need ghosts to do work for you, but they will kill you if they catch up with you. I haven't played Echoplex but this line from Brendan Caldwell's preview seems telling: "A lot of the solutions have to do with precise timing – using your old clone to do something you know you’ll need in the future." Indeed, playback puzzles are often about timing.

Naturally all puzzle games involve repeating actions and exploring consequences but the timing aspect often seems antithetical to contemplation. There’s no time to think once the recording starts, you’ve got to get moving, stay on track. While in theory it sounds like beautiful co-operation between multiple versions of the player, the reality is like working with an idiot who follows orders to the letter and has never learnt to improvise. And that's what makes exploring the possibility space so exhausting. What was supposed to be a cerebral challenge can sometimes border on an arcade, twitchy experience while expecting you to do a lot of waiting. Hang on, that really does sound like a stealth game...


The playback puzzle is a clever design pattern that injects complexity into otherwise straightforward tasks, much like the way “portals” screw around with your appreciation of space in puzzle games such as Cosmic Express (Hazelden & Davis, 2017) or Lazors (Pyrosphere, 2010). But I find the traditional playback puzzle aggravating - not because of the inherent difficulty but because they are particularly laborious and, ironically for mechanics often associated with time travel, time consuming.

Next: The fall and rise of Sokoban.

Download my FREE eBook on the collapse of indie game prices an accessible and comprehensive explanation of what has happened to the market.

Sign up for the monthly Electron Dance Newsletter and follow on Twitter!

Electron Dance Highlights

Comments (14) Trackbacks (0)
  1. Yeah, I enjoy playback puzzles in theory. In practise it’s exactly like you said. Timing issues, and the headache of accounting for double the resources. The playback puzzles were my least favourite of Talos, but I appreciated the solution once they happened.

  2. Hi Dave! Yes, I was always proud when I vanquished a particular playback puzzle in Talos but hated working through them. I was pleased to discover they practically disappeared in the (brilliant) Road to Gehenna DLC.

    Rose and Time was just too fierce for me to want to work through more than a few levels. The biggest issue is you cannot tell how far Rose can see (omniscient you can see the full distance) and it’s difficult to appreciate the full situation around you – it’s third person and you can’t really see behind you.

    I have renewed enthusiasm for Induction actually – it doesn’t suffer anywhere near as much from these issues and I’m going to keep plugging away at it.

  3. I too dislike these playback puzzles, and I think when you write “these puzzles are always a Jenga tower of cause and effect” you’ve capture the core of the reason why:

    Too often it’s impossible to solve the puzzle iteratively. Or to put it another way, many playback puzzles require me to develop a single global solution to the puzzle, instead of being a series of smaller, localised solutions. So in order to solve it, I have to hold all the variables in the entire puzzle in my head, and sort them into the right sequence, and then execute that sequence. While the need to allocate time for a later self to use (standing on a switch for 20 seconds for example) is tedious, I don’t find it frustrating except when I didn’t allocate enough. It’s juggling all the different variables of the puzzle in my head that’s the difficult, frustrating part.

    I have an aversion to Sokoban and similar block-sliding puzzles that I think is based on the same problem: their solutions tend to need careful sequencing to work, which means holding all the moves in my head and sorting them.

    It’s easy with many puzzle systems to create puzzles of this kind, of course, and designer blindness can make it hard to see. Take for example the custom Portal 2 maps I made ages ago: Island Hopping needs a global solution (and careful timing!), and while it was a fun process designing it, it’s frustrating to play, as I discovered when I revisited it long after making it, when I’d pretty much forgotten the correct solution. But one of the others, Three Cube Tango is essentially a series of much smaller puzzles, and solving each one makes the next one available to you. There’s little reuse of resources between the parts—I feel like that’s significant in its avoidance of needing a global solution.

  4. I don’t have strong feelings about this subject – maybe I haven’t played any games like this recently?

    I have a related pet peeve though! Games where you’re controlling several players with the same input (sometimes with some of them mirrored). It’s bad and awkward 99% of the time.

  5. Hi Andy

    I was making some mental connections between Sokoban and time travel puzzles while I was writing this but didn’t want to draw them out… just yet. But you’re right in the sense that these puzzles need to be solved as a whole and the process of moving towards that final state is laborious. Devil’s advocate: While it can be difficult to solve the levels in terms of constituent parts, designers would argue that you’re meant to use logic to prune the tree of possible approaches and use iteration to refine towards an actual solution. Indeed, this is the logic many Sokoban games require from their players. So, for me, Playback make this kind of tightly coupled puzzle solving very expensive compared to Sokoban which is fast and rapid. Still: I hadn’t really thought about the difference between puzzles that can be reduced into smaller components vs demanding a global whole, that’s interesting. Especially that is precisely why I hated a puzzlescript game Vacuum that was really interesting but in the end it pissed me off because the whole game was one long puzzle that you couldn’t waste moves in.


    I know what you mean, but struggling to think of examples at the moment. Can you jog my memory?

  6. Binaries is one example of the kind of game Alan is talking about.

  7. “you’re meant to use logic to prune the tree of possible approaches and use iteration to refine towards an actual solution.”

    There’s an enormous practical difference when iterating means resetting the level and doing a dozen steps, and when it means undoing the last step or two and trying them differently.

    But a Sokoban-esque game where you can pull the blocks as well as push them (thus giving you an undo) is a significantly different game to one where you can only push them. And the puzzles designed for it would be quite different too.

    This train of thought leads me to wonder what a playback puzzle game would be like if you had the ability not just simply rewind time, but to edit it like an audio or video track? To cut out a segment, or splice in a different one; to lengthen a “20 seconds of standing”, or shorten one; maybe even to be able to split off a track to create a new clone when needed, or merge it back into another track when it’s done its job. Hmm.

  8. Yeah Binaries is one example, I see it more often in the context of block-pushing puzzle games though. No examples are sticking in my mind, proving definitively that no game has ever done it well.

    (Actually I’ve kinda done it with my most recent puzzle game Hack The Net http://www.draknek.org/games/puzzlescript/hack-the-net.php and very briefly in Mirror Isles http://www.draknek.org/games/puzzlescript/mirrors.php but obviously I’m immune to criticism)

    (P.S. Vacuum is great and “the whole game is one puzzle” is always cool)

  9. “Games where you’re controlling several players with the same input (sometimes with some of them mirrored).”

    I was going to leave a comment pointing out that I’ve written a short game that relies on Joel’s pet peeve, and I got to this, and I’ve also written a short game that relies on Alan’s pet peeve. Here’s my IFDb page! The respective games are Faithful Companion and Terminator–that’s a mechanical spoiler for Faithful Companion, but the central puzzle remains unspoiled! These are the only puzzle-based games I’ve written (unless you want to count the game about making tea and toast, but insofar as making tea and toast is a puzzle that game will tell you how to do it if you thrash about long enough). The other two entries are both troll games, so I guess my primary game design talent is annoying people.

    In my defense the loop in Faithful Companion is short enough that I think the coordination problem may not arise quite as much–you never have enough time to build a Jenga tower of any size. Terminator is, well, kind of awkward (there’s some stuff in there where sometimes a robot will disappear from another behind a ridge or in a crater, and where those terrain features affect how fast you get caught in the sun, and it does not come acrossat all, because giving feedback on that in text form is hard to program). Both of those were kind of tech demos for code libraries other people wrote.

    Another game with the dual-control system is some parts of And Yet It Moves. Those tended to be the best parts somehow (and not awfully puzzly IMO). Also there are some puzzles in The Bridge that rely on this IIRC.

    Are the cooperate-with-yourself games you mention all realtime? That seems like it’d make a difference, since you have to guesstimate how to coordinate with the future self you can’t see yet.

    Andy, it’s not playback, but the editing thing reminds me of Cosmic Express–you have to start by drawing a continuous track, but you can cut it into pieces to edit it, which is nice when you’ve got something that works in one part of the level but you’re not sure how to get there.

  10. Andy
    Yeah, so I’ve had zara chiding me on Twitter in the last hour implying my comparison to Sokoban as less laborious is bullshit. Sokoban is generally hard work. I think we’re on to something here about the global solution, the tightly-coupled nature of the puzzle steps. Sokobanlikes are reknowned for needing every step to count which means you often have to revisit the first moves – which you make as steps into the unknown without much idea of what you’re doing – to make it work.

    Playback puzzles are pretty much this all over but Sokobanlikes have moved towards much more rapid iteration. Fast undo/redo, fast movement, freedom to tinker. Whereas Talos – and it did have a fast forward option on PC, right? – pretty much expects you to lay down that track from scratch every god damn time. I think Induction shows there’s a better way to handle this and bring it closer to the Sokoban gold standard: still painfuilly laborious, just not more so.

    One thing I sense is that we need, what might be turned, high quality but low frustration puzzles, which I think you can see from The Zen Lie. Perhaps not everything has to be a ball buster or mathematically pure. Hence, The Room gets called a “puzzle game”. :)

    I just remembered there are some puzzles in turn-based Cityglitch in which other characters move in tandem with you, but won’t if they hit a wall. I admit I’m not really enjoying this. And, yes, Mirror Isles, I had to dabble with that a couple of months ago, I think I gave up because of the same issue – that I just didn’t enjoy that type of puzzle. Oh oh, the Lara Croft DLC also has it! A mirror version of you faces different constraints. Oooh I’ve got some praises to sing for Lara Croft GO later on.

    I think I’d have to argue “the whole game is one puzzle” is not always cool. It’s the sense of a massive puzzle being a sequence of tightly coupled with little margin for error that bugs me. If the answer to a Sokoban game is to push 20 numbered blocks into a narrow corridor to read out the first 20 digits of pi and it only tells you what’s wrong once you’ve finished the sequence… I mean, that’s how I felt a bit after Vacuum. Like AHA you still need extra moves after spending 5 minutes refining your route to this point!


    Holy shit, Matt, if I thought Talos playback puzzling was laborious then imagine what I’ll feel like having to type in to text commands to move around :) Maybe I’ll take a look at them but I am hideously bad at following up on game links in the comments these days. I started Subnautica last night, all bets are off for the next few months.

    Real time – yes all of the examples are real-time. I think this is definitely part of the “problem”. Turn-based you could simply count the moves but, I wonder, if counting the squares is boring? I don’t know. Open question. Sometimes when games make me do that, I hate them. Other times, I think Imma genius.

    On the Cosmic Express front, it still needs you to understand the puzzle as a whole and you often “rip up the whole track” when trying to get to the solution. I think it’s less taxing than Sokoban because you don’t have to run through moves to establish configuration, but it’s the “same problem”.

    Er, I wonder if our potential for frustration lays out this way:

    Design of Solution Algorithm < Be The Algorithm in Discrete Time < Be The Algorithm in Real Time


  11. I’m currently making a snake-like game, and I just realized that a “snake” player object is akin to a “single square” player object and several time-delayed copies of itself :)

  12. I’m sure Joel will love my time-travelling puzzle platformer if I ever finish it.


    The earliest game I can think of with playback was 1983: Jumpman on the Atari 8-bit, which had ‘dodge your previous self’ N++-style in a level or three.

    I made a rather-unknown game where the enemies are playbacks of yourself from previous levels: https://silverspaceship.com/succor/


    My frustration with Sokoban-style gameplay was a significant motivator for my development of the Chromatron series. Instead of having to unwind your efforts and rebuild them, you can just move pieces around and the laser instantaneously seeks out the new path. Maybe it doesn’t strike others that way, but it was literally a reaction to that element of Sokoban.

    (But then I turned aorund and made Promesst, which is sort of a deeply-interlinked Sokoban in terms of those structural elements: everything I hate.)

  13. I agree with almost everything everyone had to say here, especially Andy’s “global” solution take, plus musings on editable playback mechanics. It could be as simple as rewinding part-way and overwriting and might a way of making it iterative.

    Anyway, I only popped in to mention that The Talos Principle gained a fast-forward function in an early update, although I don’t think it’s tutorialised, and it could stand to be faster. [F] I think is the default binding. It made a big difference for me.

  14. Guhhh, I am pretty sick this weekend. Been in bed for more hours that I care to count.


    Hm, I wonder if there would be any way of turning that realisation into something interesting!


    I did remember the Jumpman level “Follow The Leader” but I didn’t bring it up here because I was going to mention it in a later post for Other Reasons. Although there seems to be a remarkable difference between the co-op and antagonistic versions of playback puzzles. There aren’t too many antagonist puzzles, they are usually about dexterity and staying ahead. Rose and Time is the only one I know of. The co-op variety is more common.

    I just tried Succor quickly, got through to Collect and then just decided to stop. Interesting but such an extremely “delicate” feel with death assured by contact with just about everything :) Reminds me of those old Lunar Lander games. I’m not sure if it really feels like you’re “fighting yourself” versus computer-generated opponents, although if you want to beat the challenge times I guess you’d have to plan in advance. I am also reminded of George Buckingham’s Hell Is Other People.

    Interesting that Chromatron was a response to Sokoban, I’m definitely going to use this fact. (I’m going to mention Chromatron in passing later.)

    I’ll have to try Promesst at some point but only after I finish Corrypt which scared me so much with its global connectedness that I ran away and never came back.


    Thinking more about Andy going from “playback” to full editor, I’d expect an entirely different game to emerge. If you can splice together different segments of time, you’re really getting back to an algorithm creator (more of a Zachlike, or Cosmic Express). And if you conceit is that it’s actually time travel – well, I don’t think that’s going to feel like time travel at all. This is all just speculation though, things could be quite different in the hands of the right developer.

    On The Talos Prinicple: yes, I did remember rather late that the Talos Principle gained a fast forward button, but it was only available on PC if IIRC. Induction is still superior because it has rewind/fast forward a la Braid which makes for extremely precise navigation of the puzzle. I wrote of Talos: “The game does a mock rewind of events, making it even more galling that an actual rewind isn’t available; Croteam must surely have considered it.”

Leave a comment

Notify me of followup comments via e-mail.

No trackbacks yet.