What it is. A platformer where the walls are ideas, not gaps. You give the player one verb with a deep rule, then spend the whole game asking questions that verb can answer. Movement is a means; the answer is the win.

Player fantasy / why it’s fun. The click of “oh — I could do THAT.” The room looked impossible; you were missing one implication of a rule you already knew. Mastery here is comprehension, not dexterity — you feel smart, not fast.

Pillars

1. One deep verb. Not a moveset — a single rule with more consequences than it first shows. anchor-braid‘s rewind, anchor-portal‘s two linked holes. The verb must be generative: every room is a new sentence in the same grammar.
2. Wordless teaching. The level is the tutorial. No text box explains the mechanic; a safe room forces the player to discover it, then the game builds. See system-onboarding and pattern-readability.
3. The idea fully explored. Each mechanic gets introduced, practiced, combined, and finally subverted before you retire it. Leave nothing on the table; a shallow verb outstays its welcome in three rooms.

The loop stack

The moment is cerebral, not twitchy. If the player knows the answer but keeps dying to the jump, you have leaked a reflex game into a thinking game — the core failure mode (see Pitfalls).

Essential systems

• system-onboarding — the introduce→practice→combine→subvert cadence lives here. Teach one implication per room; never two new things at once.
• system-encounter-design — a room is a question. Author the insight, then build the walls that hide it. No filler rooms.
• system-mastery-curve — difficulty is conceptual depth, not tighter timing. Ramp the idea, keep the execution generous.
• system-save-and-checkpoint — a wrong idea should cost seconds, not minutes. Instant reset; the puzzle is the enemy, not the retry.
• system-difficulty-and-dda — optional side-rooms for the hungry; the main path stays a clean line so nobody gets walled.
• system-collectibles — the classic “hard optional star per world” reward for exploring the verb’s edges (Braid, N++, Fez).

Content & difficulty model

• Unit of content is the idea, not the room. Budget mechanics, not levels. A world is one mechanic × its four beats. Ten shallow rooms lose to four that each teach a distinct implication.
• Ramp the concept; hold the execution. Every room should be a fresh thought. If two rooms teach the same insight, cut one.
• Generous inputs, strict logic. Coyote time, buffered jumps, forgiving hitboxes — none of that cheapens a thinking game, and their absence turns insight into a stopwatch. See pattern-anti-frustration and pattern-fairness-and-trust.
• Solvability is provable. Puzzle content wants a machine-checked winnable path per room — see the grid-puzzle proof discipline the verify target enforces. Design so a solver can exist: keep state discrete where you can.
• Pace the reveal. Alternate a hard room with a breather that lets the new idea settle. pattern-pacing-and-tension.

Signature-mechanic seeds

Twist the verb, not the platforming. Each is X but Y with a vector in parens:

1. Puzzle-platformer but the verb is chosen fresh each world (structure) — every biome hands you a different single verb (mechanic-clone, mechanic-teleport, mechanic-grow-shrink) and never reuses it; the game is an anthology of one-idea games. Risk: no compounding mastery — buy it back with a finale that stacks all verbs.
2. One-verb but the verb is controlled by rhythm (mechanic-swap) — the manipulation only fires on the beat, so timing re-enters as structure, not twitch. Solving becomes choreography. Pairs with recipe-rhythm-platformer and genre-rhythm.
3. Time as the verb (mechanic) — mechanic-rewind or mechanic-time-stop as the single rule; the room is a chronology puzzle. Braid’s whole thesis.
4. Space as the verb (mechanic) — mechanic-portal or mechanic-gravity-flip: the puzzle is where “up” and “adjacent” secretly point.
5. The avatar splits (mechanic) — mechanic-clone / mechanic-possess: you author a past self, then cooperate with the recording. One player, two bodies.

Common pitfalls

• Execution muddying the idea — the signature trap. When the answer is known but the hands can’t deliver, you’ve built genre-precision-platformer by accident. Diagnose via antipattern-input-lie: if inputs feel dropped or death feels unfair, the puzzle is no longer the challenge.
• Guess-the-designer — antipattern-guess-the-designer: the solution hinges on an unstated rule or an unfair leap. Every insight must be derivable from what the verb visibly does.
• The verb runs dry — antipattern-false-depth: a shallow rule dressed in twenty rooms. If you can’t find four distinct implications, the verb isn’t a genre, it’s a gimmick — cut the world short.
• Endless teaching — antipattern-endless-tutorial: hand-holding past the “combine” beat. Once the player owns the idea, stop explaining and start asking.
• Content desert — antipattern-content-desert: padding a thin verb to hit a level count. Ship four brilliant worlds, not ten tired ones.

Anchors

• anchor-braid — one verb (rewind), each world a new rewind rule, wordless teaching, and a finale that recontextualizes everything. The reference.
• anchor-portal — the purest introduce→practice→combine→subvert cadence in games; two holes, infinite consequences.
• Adjacent: genre-grid-puzzle (discrete cousin), genre-precision-platformer (the reflex sibling to avoid collapsing into), genre-metroidvania (verbs-as-keys at world scale).

Verify

Prove it against design/FUN.md#1-·-grid-puzzle — machine-checked winnability, an authored insight per room, and reset cheap enough that thinking, not retrying, is the loop. Design here; prove there.