What it is. Two tokens of the same tier collide and become one token of the next tier up. The board is finite; every merge frees a cell, every non-merge fills one. 2048 in a sentence.
Player fantasy / why it’s fun. Watching a number climb by your own hand. The tier-up pop is a tiny slot-machine payout you caused — legible cause, escalating reward, and the constant low dread that the board is filling faster than you can clear it.
The verb
Bring two like tiles together; they collapse into one tile one tier higher. That’s it — the depth is where and when, not how.
How it feels / why it’s fun
- Cause you own. No RNG in the fuse itself — the payout reads as skill, not luck. Contrast the drip of
system-reward-schedules; here the schedule is self-paced. - The pop. The merge is the whole hit. Land the juice on it — scale-punch, pitch-rising chime, tier color shift. See
pattern-juice-choreographyand the choreography cousinmechanic-stack. - Compounding stakes. Each tier is exponentially rarer, so a high tile is a visible trophy and a visible liability (it hogs a cell forever until it merges).
- Board pressure. Spawns arrive whether you want them or not. The tension is space, not time — the
pattern-pacing-and-tensioncomes from the shrinking margin. See alsopattern-risk-reward.
Tuning levers
| Lever | What it controls | Sane default |
|---|---|---|
| Merge count | tiles needed to fuse | 2 (2048) — 3 tightens the vise |
| Tier ratio | value jump per tier | ×2 (doubling reads instantly) |
| Spawn rate | new tokens per action | 1 per move |
| Spawn mix | which tiers can spawn | 90% tier-1, 10% tier-2 |
| Spawn control | does the player place? | none (slide) vs. aim (drop) |
| Board size | cells before choke | 4×4 (classic squeeze) |
| Win/loss | tier goal vs. board-full | reach tier-11 / no legal move |
Small board + forced spawn = pressure cooker. Big board + player-placed spawn = cozy sandbox. Pick the axis first; it is the genre.
Slots into
genre-grid-puzzle— the native home (2048, Threes).genre-match3— merge is match-3’s greedier sibling; see its own [[design/FUN.md#13-·-match-3]] proof.genre-incremental— merge boards are a cleansystem-resource-loopsengine; the tier IS the number-go-up. Seerecipe-merge-factory.genre-city-builder/genre-management-tycoon— merge-2 mobile builders (Gram Games’ Merge Dragons) stack a soft economy on top; watch forsystem-economyandantipattern-pay-to-skip.- Anchors:
anchor-katamari(fuse-by-absorption — same “small-becomes-large” dopamine at physical scale),anchor-tetris(the finite-board-fills sibling; merge is Tetris where lines clear by likeness, not rows).
Twist seams
- Merge but merging costs the space you need (constraint) — the fused tile is bigger than its inputs, so ascending shrinks your board instead of freeing it. Now every merge is a bet against your own room. Ties to
pattern-risk-rewardand the pressure logic ofanchor-tetris. - Merge but two DIFFERENT things fuse into a third (a mechanic-swap toward crafting) — abandon like-with-like; wood + fire → torch. This is a recipe graph, so lean on
system-craftingandsystem-tech-tree. The design cost: players must learn combos, so pair withsystem-onboardingor a discoverablepattern-emergence. - Merge but the tiers are a deck (mechanic-swap) — fuse two cards into a rarer card mid-run;
anchor-balatro/genre-deckbuilderterritory, where the merge shapes a build. Seesystem-build-diversity.
How it wires to Hayao
- Pure-logic first. The board is a
Puzzle<State, Move>— a grid of tiers, a move enumerator, a deterministic resolver. Every merge is a state transition; every level is machine-proven reachable. This is exactly the logic/view splitexamples/sokoban/demonstrates. - Spawns through the deterministic RNG only — seed it so a board is replayable and the solver can prove the goal tier is attainable. No
Math.randomin the fuse or the spawn. - The pop is cosmetic: scale-punch, particle, chime live in view nodes flagged
cosmeticso they never touch the world hash. Study a particle/tween isolation insandboxes/before choreographing. - Ordered iteration when resolving a whole row of merges — left-to-right, one fuse per tile per move (the 2048 rule), or the sim drifts non-deterministic.
Fails when…
- Spawns outrun clears. If the board fills before skill can matter, it’s not tension, it’s a coin flip —
antipattern-rng-frustration. Cap spawn tiers and pace them. - One optimal path. Corner-hoard-and-snake is 2048’s solved line; if your board rewards a single rote pattern you’ve built
antipattern-boring-optimal. Add spawn control, hazards, or an objective that punishes hoarding. - The number stops meaning anything. Tier-40 feels identical to tier-8 —
antipattern-false-depthandantipattern-stat-inflation. Re-skin tiers into distinct things, or gate a new mechanic every N tiers. - Grind for a paywall. Merge economies slide fast into
antipattern-grind-wallandantipattern-pay-to-skip. Keep the loop paced by the board, not the wallet. - The pop is mushy. No punch on the fuse and the whole verb goes flat — that’s
antipattern-unreadable-juiceread backwards: too little, not too much.
See also
mechanic-stack · genre-match3 · genre-grid-puzzle · genre-incremental · system-crafting · recipe-merge-factory · pattern-risk-reward · pattern-juice-choreography · anchor-tetris · anchor-katamari · process-the-twist