Fighting Game

What it is. A one-on-one duel at a distance, decided by who controls the space between two characters and who reads the other first. Under the mind-game sits a fully legible frame system: every move has a startup, an active window, and a recovery you can be punished during.

Player fantasy / why it’s fun. “I read you.” The pull is the mental duel — whiff-punishing a button they threw too early, calling the throw and teching it, walking them into the corner one step at a time. Depth without hidden rolls: you lost because you were read, not out-rolled.

Pillars

1. Neutral is the game. The footsies phase — spacing, dashing in and out of range, baiting a button to whiff-punish it — is where matches are won. Combos are the reward for winning neutral, not the point of it. If your neutral is shallow, the whole duel is.
2. Commitment & punish. Every action commits you to recovery frames. Frame advantage is the currency: act and you’re vulnerable until you recover; read the opponent’s commitment and you punish it. The tension is when do I press?
3. The strike-throw-block RPS. Up close it collapses to a legible triangle — strike beats throw, throw beats block, block beats strike. That readable rock-paper-scissors under pressure is the mind-game surface; see system-counter-systems.

The loop stack

Essential systems

Content & difficulty model

• Content is the roster + the matchup lattice. Each character is a moveset with distinct ranges, speeds, and win conditions; depth comes from the cross product of matchups, not the count of characters. See system-unit-rosters.
• Difficulty is the opponent, not a knob. The AI ladder ramps by playing smarter neutral and tighter punishes — not by gaining hidden frames or damage. A “difficult” CPU that cheats frame data violates the genre’s honesty.
• Every move proven in isolation. Startup, active, and recovery are data; each interaction (does X beat Y on wakeup?) is a test before the cast combines. Derive frame outcomes, don’t vibe them.
• Determinism is the referee. Same inputs, same frame → same result, always. Route randomness through a deterministic RNG only for cosmetic flourish, never for a hit that connects. A “miss” the player couldn’t foresee is a betrayal.
• Combos gate on execution; wins gate on reads. Keep the required execution to convert a hit low so the mind-game — not the muscle memory — decides the match. This is the genre’s central design lever.

Signature-mechanic seeds

• Fighter but no stick (constraint) — Divekick-style: two buttons and no directional stick — one jumps and dives, one kicks; 100% of the depth is spacing and reads. Strips the execution barrier to zero and leaves pure neutral. The cleanest way to prove the mind-game is the game.
• Fighting game but simultaneous-turn, resolved after commit (mechanic-swap) — both players lock a move blind, then the frame exchange resolves; a duel of pure prediction with no execution at all. Bridges the read-heavy honesty of genre-tactics onto the strike-throw-block triangle.
• Fighter but the stage is the third character (theme + mechanic-swap) — a windswept courtyard whose hazards, wind, and shifting floor edit both fighters’ ranges each round; positioning gains a second axis. Pairs system-hazards-and-environment.
• Fighter but health is shared and shrinking (structure) — one bar between you; every hit is a tug-of-war over the same pool, and time pressure forces commitment. Reframes turtling as a losing strategy structurally.
• Fighter but you draft your normals between rounds (structure) — a deckbuilt moveset assembled from a shared pool mid-set; matchup knowledge becomes live adaptation. Bridges genre-deckbuilder.

Common pitfalls

• Execution gates the mind-game. The genre’s signature trap: a one-frame link or a quarter-circle motion stands between the player and the actual game (neutral and reads). If the fingers fail, the mind never gets to play. Lower the execution floor — see antipattern-input-lie and system-onboarding.
• Input lie. A dropped combo the player did input right, or a move that came out on a lenient buffer they didn’t mean — the display and the sim disagree. Fatal in a game about precise commitment. antipattern-input-lie.
• A solved metagame. One character or one setup dominates and the RPS collapses to a single winning option. Once neutral has a right answer, the mind-game is dead. antipattern-solved-metagame.
• Unreadable startup. A move with no telegraph, or a “mixup” the defender cannot react to and cannot guess, is a coin flip, not a read. antipattern-guess-the-designer / pattern-readability.
• Ceiling without a floor. All ceiling and no on-ramp is a wall, not a curve. A new player must land a punish in their first session. antipattern-difficulty-cliff.

Anchors

• anchor-street-fighter — the definitive reference: footsies, frame advantage, and the strike-throw-block triangle. Steal the neutral vocabulary and the legibility of its frame data; leave the six-button execution barrier where it doesn’t serve the read.

Verify

Prove it in FUN.md §4 · Action-adventure: same-input/same-frame determinism, each move’s startup/active/recovery asserted as data, the strike-throw-block interactions proven in isolation, and a golden round replaying from an input log to an identical end-state.

Composes with

• anchor-street-fighter — the footsies-and-frame-data model to steal wholesale.
• mechanic-combo-string — the reward for winning neutral; keep its execution floor low so reads, not fingers, decide.
• system-mastery-curve — the high ceiling that is the point of the genre.
• system-counter-systems — the readable RPS triangle under the mind-game.

See also

• examples/sokoban — the pure Puzzle<State,Move> logic/view split and clone-and-score bot; a frame exchange resolved from data is the same deterministic shape.
• sandboxes/physics-lab — hitbox/hurtbox overlap and knockback in isolation before the cast combines.