Module 2 · Console Wars

The crystallization of genres: from mastery to progression

Between 1985 and 1993 mechanics stopped flying solo and settled into genres — stable bundles of "verb + win structure + content form". This is knowledge engineering: name the bundle so the next person doesn't rediscover it from scratch.

~16 min

The gist in 20 seconds

A genre is a stable bundle of three things: a core verb (jump / explore / fight / build), a win structure (an endless mastery loop vs a progression arc) and a content form (linear levels / an open world with gates / an arena / a sandbox). The arcade asked "can you handle the pattern?" — an endless loop with difficulty rising to eat your coins. The home console (a cartridge, no coins) asked "will you reach the end?" — and a macro-narrative grew on top of the game loop from m01: levels, bosses, a rhythm of "tension↔breather". And the craft that made progression work is teaching through play (Miyamoto's method): introduce a mechanic somewhere safe → repeat it with variation → combine. A genre is the compression of hard-won design into a reusable form.

The mechanism

Genres weren't "invented" — they crystallized: bundles of mechanics that consistently worked got a name and started being reused. Three parts of the bundle and two forces that precipitated them in this era.

What a genre is: a three-part bundle

Take any game of the era and break it into three layers:

The core verb — the player's main action (Mario: jump; Zelda: explore; SF2: strike on timing; SimCity: zone).
The win structure — a mastery loop (you play until you lose) or a progression arc (start → middle → finale).
The content form — how the world is sliced (linear levels / an open world locked behind items / a 1-on-1 arena / an endless simulation).

When a bundle repeats across several games and gets a name — that is a genre. "Platformer" = {jump, progression arc, linear levels}. "Metroidvania" = {jump+shoot, progression, an open world gated by items}.

The structural shift: mastery loop → progression arc

The arcade fought for coins: difficulty rose monotonically until it killed you — so you'd buy the next attempt more often. The home console asks for no coins, but the cartridge is expensive: the goal is not "milk you longer" but get you to the end enjoying yourself. Hence the arc: a climb → a boss → a breather → a new climb. Formally both curves chase the flow channel — keeping difficulty $D$ near skill $S$ :

| D - S | \leq ε \Rightarrow flow D ≫ S \Rightarrow anxiety D ≪ S \Rightarrow boredom

The player's skill $S$ grows over the course of the game. The arcade keeps $D$ above $S$ and pushes you into "anxiety" on purpose (a death = a coin). Progression leads $D$ along behind $S$ , adding breathers so the channel doesn't collapse into anxiety. Bosses are local peaks; the levels after a boss are a dip for recovery.

The arcade drives difficulty past skill (into anxiety — for the coins). The console saws: a boss peak → a breather dip, staying inside the rising flow channel.

The craft of progression: teaching through play

For the arc to work, the player has to learn along the way — without a manual. In Super Mario Bros, Miyamoto set the algorithm that became the gold standard: introduce in a safe context → repeat with variation (raise the stakes) → combine → test. World 1-1 is literally a curriculum laid out in tiles (about 210 tiles long):

Tiles	What is introduced	The lesson
0–20	flat ground	walking, the controls
~20–30	the first Goomba on flat ground	contact = death (safe to dodge)
40–60	the first pit	jumping
60–80	Goomba + pit	combining two skills
80–100	a mushroom from a ?-block	growth/power-up
100–140	everything together, denser	a test under load

Every element is isolated before it is combined, none is skipped, and a mistake teaches (the death is avoidable on the second try) rather than punishes. This is not "a level with a tutorial" — the level is the tutorial. Same technique in Zelda 1: the first dungeon teaches dungeon navigation before the world demands it.

🕹 Games to play — and what to notice

Five genre bundles that settled in this era, from pure teaching-through-play to endless simulation. For each: what the bundle {verb · structure · form} is and what to notice with your hands.

Super Mario Bros 1985 · platformer · teaching through play

{jump · progression arc · linear levels}. World 1-1 is the curriculum from the table above. One verb (the jump), and all the depth comes from the placement of obstacles, not from the number of buttons.

🎮 Play: go through World 1-1 and say out loud where each skill is introduced. Then find the moment where two skills are first required together (a Goomba by a pit). That is the seam between "variation" and "combination".

The Legend of Zelda 1986 · action-adventure · a world locked behind items

{explore · progression · an open world with gates}. You go where you like, but the ladder, the bombs and the raft unlock new chunks. Progression is not by level but by capability: a new item = a new piece of the map.

🎮 Play: in Zelda 1, run into an obstacle you can't pass (a cliff, water), remember it — and come back once you have the item. That is an "item gate"; in a continuous space it becomes a metroidvania.

Street Fighter II 1991 · fighting game · asymmetric balance

{strike on timing · mastery loop (a match) · a 1-on-1 arena}. 8 unique fighters with different speeds and ranges. Balance is not "everyone is equal" but non-transitive rock-paper-scissors: each one beats somebody and loses to somebody (see the 🎲 deep end on frame data).

🎮 Play: take Dhalsim (long limbs, slow) against Zangief (a grappler, up close) — feel how distance decides the match. That is the asymmetry that gives birth to the "who to pick" metagame.

Tetris 1984/89 · puzzle · a mastery loop, not an arc

{stack · an endless loop · a falling grid}. The counterexample to "everything became progression": Tetris stayed endless mastery. 7 pieces + a 10×20 grid — that is the whole game. Minimal mechanics, infinite depth.

🎮 Play: catch the moment when rising speed makes difficulty outrun skill — the flow channel collapses into "anxiety" and you die. Here that is the design: a mastery loop, not an arc with a finale.

SimCity 1989 · simulation · emergence out of rules

{zone · no win condition · a sandbox simulation}. The designer writes the rules (taxes, demand, pollution) and the city's behavior is emergent: high taxes → business leaves → housing demand falls. The player is a systems designer, not the executor of a script.

🎮 Play: crank the taxes and watch the city degrade on its own through a chain of rules nobody scripted. Find a feedback loop (industry→pollution→housing flees). That is emergence with your own eyes.

Deep end · design: frame data and why asymmetry runs deeper than equalityskippable

Fighting game balance is not "everyone is equal" but a non-transitive ecosystem. Two tools hold it together: frame data and the matchup matrix.

Frame data: every attack has three phases

An attack lives in frames (60 fps): startup (the wind-up, not hitting yet) · active (hitting) · recovery (the come-down, vulnerable). Whoever reaches active first lands the hit. If at the same distance your startup is 6 and your opponent's is 4, they get there sooner:

t_{hit} = s_{startup}, 4 < 6 \Rightarrow the opponent hits first

Advantage "on block" = how much sooner you recover than the opponent exits blockstun (the pause the blocked attack holds them in): their blockstun minus your recovery.

adv = b_{block} - r_{attack}

Positive means you are plus (your turn), negative means you get punished. The whole theory of "what is safe to poke with" grows out of these numbers.

Non-transitivity: why there is no "best" fighter

Healthy balance is a cycle with no dominant apex:

A ≻ B, B ≻ C, C ≻ A

The matchup matrix (A's win share against B):

vs →	zoner	rushdown	grappler
zoner	5–5	6–4	4–6
rushdown	4–6	5–5	6–4
grappler	6–4	4–6	5–5

The zoner walls out rushdown at range but drowns against the grappler, who forces the distance closed and catches you in a throw; rushdown in turn crushes the slow grappler — a closed cycle of zoner ≻ rushdown ≻ grappler ≻ zoner. "Perfect balance" (all 5–5) kills the metagame — picking a fighter stops meaning anything. A tilt plus non-transitivity = depth: "who do I take against him?".

Deep end · economics: how the business model bent the difficulty curveskippable

Difficulty isn't only design, it is also money. Three models of the era shaped three different curves.

The arcade: a quarter-driven economy

Revenue = coins per attempt. The operator's optimum is to keep the match short and the difficulty high: the faster you die, the more often you pay. So the "unfair" difficulty of arcades is direct monetization, not designer sadism.

Cartridges plus rentals: "Nintendo Hard"

A cartridge cost $40–60, a Blockbuster rental a couple of dollars. If a game can be finished in a weekend, the rational player rents it instead of buying it. The industry's answer was to pad the length with difficulty (which is "free": numbers in a table, not new content): limited lives, no saves, pattern memorization. That is how the Nintendo Hard phenomenon was born (Battletoads, Ghosts'n Goblins, early Mega Man) — rental economics dressed up as game design.

Shareware: a funnel instead of difficulty

id Software (Commander Keen, Doom): the first episode free, the rest by mail order (≈$30–40 for the full version). Here difficulty isn't needed for retention — what retains you is viral distribution and the desire to see what comes next. The business model moved the lever from "difficulty" to "marketing", and the curve straightened out.

The takeaway: before explaining a difficulty curve by "the designer's taste", ask how the game made money. More often than not the shape of the difficulty is the shape of the monetization.

Analogy

A genre is a trellis for a climbing plant: a ready-made structure the new game climbs without growing its own support. Literature settled "the detective story" and "the sonnet" the same way: a bundle of devices both the author and the reader already know. And the crystallization itself is like a supersaturated solution: mechanics floated freely in arcades, and the home console (a cartridge, no coins, a game finished in one sitting) became the seed crystal — and the forms precipitated out as named genres.

Why it matters

A genre is compressed design knowledge: by naming the bundle {verb · win structure · content form} you let the next developer reuse a hard-won structure instead of rediscovering it. That is exactly what design patterns are in code. And in reverse: once you can decompose any product into that triple, you have a tool for reasoning about the "genre" of anything — from UX onboarding to the structure of the course you are reading right now.

🔁 Beyond games — where this transfers

The lesson gives you three transferable moves: teaching through play (ordering from simple to complex), non-transitive balance (no dominant strategy → you need a population) and a named bundle as compressed knowledge.

ML / AI (your domain): Miyamoto's method is curriculum learning (Bengio, 2009): feed examples from simple to hard and training is faster and more stable — precisely "safe context → variation → combination". SF2's non-transitivity is populations in self-play: the AlphaStar league, fictitious self-play and PSRO exist exactly because there is no "best policy" (rock-paper-scissors), and you need a population of opponents rather than one champion. SimCity's emergence ⇄ multi-agent RL and agent-based modeling: you set the rules and rewards, and behavior arises on its own.

UX / product: teaching through play = progressive disclosure and onboarding without a manual; "genre conventions" = interface idioms the user already knows (don't reinvent the scrollbar). The content form ⇄ information architecture.

Software engineering: a genre ≡ a design pattern (GoF): a named reusable bundle with the same epistemics — it raises the median but can ossify the ceiling. A breakthrough is often breaking the bundle (the metroidvania = a platformer verb + RPG gates), the way a new abstraction breaks an old pattern.

The principle: name the reusable bundle so the next person climbs it instead of growing a support from scratch; teach in simple→complex order; and where no dominant strategy exists, keep a population rather than a single champion.

🔧 Run it and poke at it — on your home machine

What to play is above (🕹). Here — take the bundles apart like a designer:

🔧 Poke at it (deconstruction) ~40 min

Take 3 games of the era and write each one out as the triple {verb · win structure · content form}. Then map the difficulty curve of SMB World 1-1 by tile (an emulator plus a notepad): where the peaks are, where the breathers are, where skills are first combined. Compare it against the curriculum table from the lesson.

🧪 Test it (designer's eyes) ~15 min

In a fighting game, find the non-transitivity: pick a trio of fighters where A beats B, B beats C and C beats A (by feel is fine). In a platformer, find "unfair" difficulty (RNG / no warning) versus "fair" difficulty (learnable) — and decide which comes from design and which from economics (Nintendo Hard).

Checklist: decomposed 3 games into the bundle triple; mapped the 1-1 curve by tile; found a non-transitive trio of matchups.

Connections

foundation

The game loop — the progression arc grew on top of the basic loop from m01: the loop turns the same way, but is now wrapped in a start→finale macro-structure.

contrast

Cartridges and mappers — "hardware vs design": the NES is four times weaker than the Genesis, yet it was design (not megahertz) that produced the better games. This lesson is the second half of that thesis.

The controller and input schemes — SF2's asymmetry rests on motions (↓↘→); the next lesson is about the input channel that makes them possible.

Game feel — the "feel" of Mario's jump is the layer beneath the genre: the design of the verb that the bundle makes reusable.

Questions worth asking

If asymmetry runs deeper than equality, why isn't "maximally unfair" the best balance?

Because balance is about non-transitivity and viability, not inequality for its own sake. Every option has to beat somebody, lose to somebody and be learnable. "Maximally unfair" = one dominant strategy with no counterplay → the choice collapses to a single peak → the metagame is as dead as it would be under perfect equality. Depth comes from the cycle A≻B≻C≻A, not from the size of the tilt.

"SMB 1-1 teaches without text" — is that real design or a myth after the fact?

Mostly real and deliberate (Miyamoto confirmed the "teach through play" philosophy), but watch out for survivorship bias: we canonize the games that worked and don't remember the hundreds of games of the era that taught terribly. The principle (isolate → vary → combine) is sound and testable; the halo of "the geniuses calculated everything" is partly selection. Take the algorithm, not the mythology.

Tetris never became progression (still endless mastery). Doesn't that break the "arcade → progression" thesis?

No. The thesis is that progression appeared as an option, not that it displaced mastery. The era added arcs without deleting loops. "Endless mastery" itself settled into a full-fledged genre (puzzlers, arcade runners, roguelikes). Tetris is proof that both win structures crystallized in parallel; the choice between them is part of a genre's definition, not a stage of evolution.

If console games are about "reaching the end", where did Nintendo Hard's sadistic difficulty come from?

From rental economics. A game that can be finished over a rented weekend loses a sale. Difficulty is the cheapest way to stretch engagement: numbers in a table, not new content. Limited lives, no saves, memorization — Blockbuster economics dressed up as game design. Before praising or damning a difficulty curve, ask: how did the game make money?

A genre compresses knowledge — but doesn't it also breed clones and smother novelty?

It does; it is two-sided compression, like abstraction in code. A genre raises the median (anyone can assemble a solid platformer from the bundle) and simultaneously ossifies the ceiling (endless clones). Novelty is usually a deliberate break of the bundle: the metroidvania broke the platformer's "linear levels" by inserting RPG gates; the immersive sim broke the "scripted level" by inserting a systemic sandbox. The same trade-off as "a pattern speeds you up but goes stale": the value is being able to both apply the bundle and break it at the right moment.