Back

What makes or breaks agency in roguelikes

Over time, while gathering ideas for my own roguelike design docs, I began to notice a common thread running through many of the design principles: agency. It shaped how I started thinking about what makes roguelikes work. Hence, I decided to write some of those thoughts down here.

Mastery and meta-progression

The mastery loop is what draws me to roguelikes most. The feeling that you are the one improving, not just your character. You find this in competitive shooters and MOBAs too, but roguelikes emphasize it more directly.

I won’t step into the dictionary debate of roguelikes versus roguelites 1, but suffice to say, meta-progression 2 takes away from the mastery loop. If the game isn’t consistently beatable on the first run with skill alone, then agency diminishes. It should be the player’s skills and choices, not luck or grind, that are the path to success.

Roguelites with heavy meta-progression aren’t worse games, but they’re prioritizing something other than mastery. This post isn’t about those games. My focus is also only on modern rogueli(k|t)es, since that’s what I’m most familiar with (sorry, NetHack).

Why post-action luck hurts agency

Randomness is essential to roguelikes. Procedural generation, random challenges, and having to adapt to what you’re given are key to replayability and adaptive skill expression 3.

Balancing a roguelike between agency and randomness is challenging. Too much predictability kills tension. Too much randomness makes choices meaningless. Good roguelikes balance between the two. This is something one can only perfect through continuous iteration.

However, there is another related principle which can be thought of ahead of time and may aid in this balancing act. It’s the concept of pre-action versus post-action luck:

  • Pre-action luck: the randomness happens before your decision. You’re dealt a random hand, but you choose how to play it.
  • Post-action luck: the randomness happens after your decision. You commit to a move, and roll a die to see the result.

In the context of roguelikes, think about a chest you have to unlock with a key to get the item inside. In case of pre-action luck, you can see what random item is in there and choose whether it’s worth spending a key on. With post-action luck, it’s left purely to chance.

Overusing post-action luck reduces agency and shrinks the space for skillful decisions. This doesn’t mean you should entirely avoid post-action luck, as it can still provide a desirable sense of excitement in a game. But use it sparingly, and when you do, lean toward positive post-action luck over negative. For example, let players get excited when they stumble upon a powerful item, but don’t frustrate them for using a rare key on a mystery chest only to get junk.

My personal principle here is to make the game as random as possible, while also sacrificing as little agency as possible.

Skill expression

While there are many skill models like DKART, here’s the way that I like to think of it:

  • Instinct: Muscle memory (dodging, aiming, reacting) and micro-decisions. Mostly applicable to action roguelikes.
  • Knowledge: Understanding the game’s systems, enemies, items, and environment.
  • Strategy: Built on top of knowledge. Dynamic, adapative, and gained through experience. Involves both short- and long-term decisions. Think of risk management or optimizing your builds.

Skill expression arrives via complexity, which can in turn be split into emergent and intrinsic complexity (sometimes called depth & complexity):

  • Emergent complexity: arises from interactions between deceptively simple systems. Low floor, high ceiling. Think Go, Tetris, Devil Daggers.
  • Intrinsic complexity: baked into the systems themselves. Complex mechanics. High floor, high ceiling. Think Dwarf Fortress, Path of Exile.

Having low skill floors and high ceilings is a good design principle in many genres, but deserves special mention when it comes to roguelikes. Emergent systems let players uncover mastery organically, through experimentation and failure. This process is the long-term meta-gameplay of a roguelike. Piling on intrinsic complexity doesn’t create a meaningful skill journey. The goal is for players to discover new ideas and feel like they’re outsmarting the game, not like they’re just reading its manual.

Replayability

If agency is the feeling that your choices matter, then replayability is agency that survives repetition.

Focus on increasing decision space. Present the player with as many interesting choices as possible. The same strategy shouldn’t work twice in a row, and there should never be a fixed, optimal path. Relatedly, don’t let players optimize the fun out of your game 4 via stale, obvious choices.

How you handle your power scaling also matters. Decisions should build on top of each other, and early choices shouldn’t be drowned out by later ones. An item you picked up at the very start should still matter by the final boss. Slay the Spire gets this right, as early cards still shape your entire run, keeping the game fun from start to finish.

Tying into that: don’t treat the early game as a warm-up. The early game is absolutely important for building tension, but shouldn’t be designed in a way where it eventually becomes a trivial chore. Out of everything mentioned, this is the one point I see the most roguelikes fail at.

Solutions? For example, time constraints can incentivize optimizing early gameplay instead of sluggishly playing through it. When early items matter, then collecting as many as possible in preparation for the late-game will stay engaging. Encouraging early game scarce resource conservation and letting the player take on more risks are similar solutions.

Final thoughts

Some games break these rules and still succeed, but they tend to trade away agency in the process. And usually, they’re not the ones I return to.

Let me know what you think!

Footnotes

  1. Some differentiate roguelikes from roguelites based on whether there’s meta-progression. Others call a game a roguelite if it deviates from any of the main Rogue tenets, most often turn-based grid gameplay. Then there’s the Berlin roguelikes, roguelike-likes, … I hate words. ↩

  2. Meta-progression is when you can carry over some of your progress to the next run. Permanent upgrades, unlocks, and other elements that make the game easier on future replays. I don’t consider unlocks that add variety without impacting difficulty problematic. ↩

  3. After finish this post, I came across an interesting article on understanding randomness in terms of mastery, which I absolutely recommend reading. It describes in deep detail how randomness is a way for players to express mastery over a game’s systems, or mastery over noise. ↩

  4. A well-known idea in game design is that players will optimize the fun out of your game if they can. This can mean anything from hoarding consumables to eternally hiding in lockers in a stealth game. Hence, avoid such “optimized play” from being an option in the first place. Make the most fun strategies also the most viable ones. Reward experimentation, too: having hidden OP item combinations is part of what makes roguelikes fun. ↩