The Roguelike Canon (the actual ones)

The parenthetical in the title is a joke that’s also a real position, so let me get the tedious part over with.

In 2008 a group of developers met in Berlin at the International Roguelike Development Conference and tried to define the word, because it had already started drifting. What they produced — the Berlin Interpretation — is a list of high-value factors: random environment generation, permadeath, turn-based play, grid-based movement, non-modal design, complexity, resource management, hack-and-slash combat, and exploration through discovery. It’s a family resemblance test rather than a checklist, and its authors said so at the time.

Sixteen years later the word means “has runs in it”, and every argument about this is exhausting. I’m going to make the case for the old meaning once, quickly, and then list the games — because the definition points at a design idea that the modern usage has genuinely lost.

The idea the word used to carry

Non-modal is the one nobody talks about and it’s the load-bearing one. In a real roguelike, everything happens in the same system. There’s no combat mode. Fighting a monster, eating a ration, reading a scroll and walking down a corridor are all the same kind of action, resolved by the same rules, costing the same currency: a turn.

That’s what makes the emergent stories possible. When every object in the game can interact with every other object through one shared verb space, you get the situations these games are famous for — the ones where you kill something by luring it into a trap you didn’t set, using an item whose actual function you’re guessing at. A run-based action game with a meta-progression tree gives you variety. This gives you combinatorics, which is a different animal, and it’s the thing the loop is really arguing for.

The other half is that permadeath means the death is the end. No permanent upgrades. No “you keep the currency”. Your competence is the only thing that carries forward, which makes learning the game the entire progression system. Permadeath and the stories it buys is the whole reason to tolerate the rest.

Rogue (Michael Toy, Glenn Wichman, Ken Arnold, 1980)

Written on BSD Unix at Berkeley, using Arnold’s curses library to draw a dungeon out of letters. Toy and Wichman wanted a dungeon that would surprise them, which is a fine statement of the genre’s founding motive.

I’ll be honest that I came to this the long way round — Epyx put Rogue out on home computers in the mid-eighties, including a C64 version with actual graphics, and that’s where a lot of British teenagers met it. It’s a lesser thing than its descendants. It’s in the canon because it named the shape.

NetHack (The DevTeam, 1987 onwards)

Descends from Jay Fenlason’s Hack (1982) and has been in continuous development for roughly thirty-seven years, which makes it older than most of the people currently playing it.

NetHack is the genre’s maximalist argument and the source of its best-known law: the DevTeam thinks of everything. Try something absurd — dip a longsword in a fountain, write in the dust, sacrifice a corpse on the wrong altar — and there’s a coded response. That density is the point. NetHack’s world has more edges than any player can hold, so the game stays surprising after decades, and the wiki became load-bearing infrastructure rather than a cheat sheet.

Angband (Alex Cutler, Andy Astrand, 1990)

Out of Robert Koeneke’s Moria (1983), written on a VAX at the University of Oklahoma by someone who’d played Rogue and wanted it deeper. Angband takes Tolkien’s Middle-earth, digs a hundred levels down to Morgoth, and commits to depth as difficulty with a purity nothing else matches.

It’s also the genre’s great forking tree: Sil, Frogcomposband, and dozens of variants descend from it, because the codebase invited people to argue with it in C.

ADOM (Thomas Biskup, 1994)

Ancient Domains of Mystery does something the others don’t: it has a world. There’s an overland map, quests, factions, and a corruption mechanic that mutates you as you’re exposed to chaos, on a clock you can’t stop.

The corruption is the design lesson. It’s a resource that only moves one direction, so every hour you spend optimising is an hour spent decaying, and the game creates urgency without ever putting a timer on screen.

Dungeon Crawl Stone Soup (2006 onwards)

Forked from Linley Henzell’s Dungeon Crawl (1997) and shaped by a maintainer culture with an unusual policy: they delete things. Stone Soup has spent nearly two decades removing mechanics that create tedium without creating decisions — food clocks, most of the inventory management, whole item classes.

That editorial rigour makes it the best-designed game on this list. Every remaining system earns its turn cost. If you want to see what the crafting menu as busywork tax looks like when a team takes it seriously, this is the exhibit.

Brogue (Brian Walker, 2009)

The beautiful one. Walker built a roguelike where the ASCII is genuinely well art-directed — coloured, lit, with fluid dynamics for water and gas and fire that spread through the dungeon on their own logic.

Brogue’s design position is legibility. You can learn the whole game without a wiki, because every item’s behaviour is discoverable through play and the interface tells you what it knows. It’s the counter-argument to NetHack’s density, and both are right.

Dwarf Fortress (Tarn and Zach Adams, 2006)

Adventure mode qualifies, and the fortress mode next to it is the most ambitious simulation anyone has attempted with a text display. Tarn Adams wrote it as a life’s work and lived on Patreon donations for over a decade before the 2022 Steam release with tileset and music finally paid the brothers properly.

It’s here because it’s the extreme end of complexity as a high-value factor. The world generates centuries of history — civilisations, wars, individual named artefacts — before you arrive, and the limits of surprise in procedural generation is a conversation this game keeps winning by brute force.

Tales of Maj’Eyal (Nicolas Casalini, 2012)

The one that made the old form approachable without lying about it. ToME has a proper interface, a talent-tree build system with real depth, and an optional life system that lets newcomers survive their education.

Purists grumble about the extra lives. They’re wrong: the game’s hardest modes are pure permadeath, and the ramp brought in a generation who’d otherwise have bounced off Angband’s opening hour. Difficulty as a design choice rather than a badge — which is the argument.

Cogmind (Josh Ge, 2015 onwards)

You’re a robot made of parts, every part is scavenged from things you’ve destroyed, and damage removes capabilities rather than reducing a number. Lose your legs and you’re crawling. Lose your sensors and the map goes dark.

That’s the cleanest mechanical idea in modern roguelikes. Your character sheet is your inventory, so every fight is a negotiation about what you’re willing to have broken. Cogmind also has the best-looking terminal interface ever built, which shouldn’t matter and does.

Caves of Qud (Freehold Games, 2015–2024)

Nine years in early access, 1.0 in December 2024, and worth the wait. Qud is a science fantasy world with procedurally generated ruins, a fully simulated body-part system, and NPCs whose factions and histories generate at world-gen alongside the geography.

The design achievement is that Qud is readable despite the complexity. You can talk your way through most of it, mutate into something unrecognisable, or dig through a wall because the walls are made of a material with properties. It’s NetHack’s density with forty years of interface learning applied.

The honest caveat

The pedantry has a limit, and here it is: the games that borrowed the word made better entertainment. Hades solved narrative repetition in a way no traditional roguelike has managed. The deckbuilder canon is a genuinely new design space that grew out of this soil. Dead Cells’ roguevania blueprint works on its own terms.

The word drifted because the drift was productive. Roguelikes went mainstream and roguelites ate them, and complaining about it is a hobby. What’s worth defending is the design: turn-based, non-modal, combinatorial, with nothing carrying forward except what you learned. Ten games still do that properly. Most of them are free.

Where to play them

NetHack, Angband, Brogue CE, Stone Soup and ADOM are free downloads from their own sites; several also run in a browser or over public SSH servers where you can watch other people die in real time, which is the genre’s great spectator sport. Dwarf Fortress, Cogmind, Caves of Qud and ToME are on Steam. Rogue itself is easiest through a DOS or C64 emulator.

Start with Brogue. It’s the only one on this list that will teach you the genre’s actual pleasures inside two hours, and it’s a 4MB download.