recipe

Use when

You want a merge toy that grows into a systemic automation game.

Merge Factory

What it is. A merge toy that eats itself. You combine two low-tier tiles into one higher tier by hand — then you build a machine that does the combining, and the game becomes about designing the machine, not swiping the tiles.

Player fantasy / why it’s fun. “I loved the merge — now I never have to do it again, and I love that more.” The pull is watching a chore you enjoyed become a humming line you designed. The first auto-merger is the dopamine spike; the first bottleneck is the hook.

The brief

2048 but you automate the merging — a structure twist (process-the-twist). Take the merge verb, keep it, then move it up a level of abstraction: the player stops merging and starts building the thing that merges. Factorio did this to crafting; do it to Threes.

The spine

Finite board space is the one scarcity, and the automation that clears cells by merging is the same automation that keeps flooding them — you must out-design your own throughput before the board jams.

PartThis game
ObjectiveKeep the line running and climb the tiers — reach tier-8 without the board jamming
SuperpowerAutomate the merge — place droppers, mergers, belts, sorters so the combining runs itself
ScarcityBoard space — a finite grid of cells. Every merge frees a cell, but every dropper and every belt-tick fills one; throughput is spent against area you don’t have
ObstacleYour own faster half: tier-N output outruns the tier-N+1 merger downstream, tiles pile, and unmerged stock crowds the cells the line needs to keep flowing
RenewalThe bottleneck moves as you solve it (tier-3-limited → tier-4-limited, iron→copper→power); each new tier re-poses “throughput vs. space” one abstraction higher

Resonance

Every element traces to the spine — the coherence proof (see process-the-spine).

ElementArrow back to the spine
Verb: mechanic-merge (two-of-a-kind → one-of-next-tier)The unit of throughput and the only thing that frees a cell — the agency the whole board fights over
Power creates the problemAutomating the merge is exactly what floods the board: the droppers and belts you place to clear cells are the same machines feeding new tiles in faster than the next tier can consume them (passes the gate)
Scarcity: finite board spaceTurns “more throughput” into a trade — you can’t buy speed without spending the area the line runs on
Renewal: the moving bottleneck (system-resource-loops)Re-poses “out-run your own fill-rate” against a fresh binding constraint each tier; the loop narrows somewhere new every beat
Manual first, then automate (mechanic-merge by hand)Teaches the merge that costs a cell before the machine hides it — you feel the space you’re spending before you scale it
Ascension feel (anchor-katamari)Climbing tiers is the objective made visceral; each tier is bigger stock competing for the same finite cells
Tech tree spends (system-tech-tree)Faster belts and higher tiers are more throughput — buying deeper into the very pressure the scarcity punishes
Economy sink (system-economy)Cashing out finished tiles is how you reclaim board space — the pressure-release valve priced on the spine, not free (antipattern-dissonance avoided)
No-desert pacing (genre-incremental)Every minute a close next unlock keeps the throughput-vs-space decision live; a wait with nothing to optimize would let the tension go slack
Feel: juiced fuse — squash, pop, tier-color bump (pattern-juice-choreography)Makes the one act that frees a cell feel like the relief it is, so the flood reads as pressure and the clear reads as reward ([[design/JUICE.md]])

No row is decoration; no row fights the spine. The gate holds: the automation that clears the board is the automation that floods it — you can only survive by out-designing your own throughput.

Anchors

Genre + systems pulled

Link them; don’t restate them. The recipe is the wiring — open those modules for the levers.

The twist applied

Vector: structure (the “operate on the structure that produces the play, not the play” move from process-the-twist). One verb, lifted one level:

X but YTwist vector
2048 but you automate the mergingstructure — play the machine, not the board
Merge game but merges have throughput (a belt clogs)mechanic-swap — swipes become flow
Idle merger but the layout is the puzzle, not the waitconstraint — spatial logistics, not a timer

The trap it dodges: most merge games are pure genre-match3 dressing with a prestige bolt-on (antipattern-second-system). Here automation is the game the merge was training you for — the two halves share one loop.

The 3 pillars

  1. Satisfying merge. The base verb has to feel great by hand before you automate it — juice the fuse (squash, pop, tier-color bump; see pattern-juice-choreography). If the manual merge is flat, automating it just hides a dead toy.
  2. Build the machine. The mid-game verb is placement: droppers, mergers, belts, sorters. The player designs a line and watches it run (pattern-emergence — you place rules, outcomes emerge). Admiring your own working factory is the reward.
  3. No-desert ramp. The tier curve pays back on schedule the whole way up. Every minute there’s a close next unlock; the moment the player is waiting with nothing to optimize, the game has failed (genre-incremental pacing law).

Scope & first playable

Cut to the smallest thing that proves the loop bends from hand to machine:

  1. Manual merge. A small grid; drag two tier-1 tiles together → one tier-2. Pure toy, fully juiced. This alone must be fun for ~60 seconds.
  2. First auto-merger. Earn/buy one machine that consumes a queue of tier-1s and emits tier-2s on a belt. The player’s hands come off the merge — the aha moment. Keep everything system-resource-loops: tiles are counts in world.state, each drop/place/merge is an input action so the whole run replays and hashes.
  3. A bottleneck. Tier-2 output outruns the single tier-3 merger downstream; tiles pile up. Now the player must add capacity or re-route — the first design problem, and the seam every later tier reuses. Move the bottleneck as they solve it (iron→copper→power, merge-style): solved tier-3, now tier-4-limited.

Everything past step 3 — more tiers, sorters, prestige, sinks — is content on a proven skeleton, not new skeleton. Resist antipattern-feature-soup: ship the three-beat loop first.

Handoff

A design isn’t done until it names its proofs (process-refine-and-handoff).

Composes with

See also

This module rendersthe repo's markdowndirectly — edit it there, it changes here.