Cube State Engine

A tiny, dependency-free engine that tracks a Rubik's cube state, move by move.

Powers the solve tracking and analytics core of NexusTimer

---

What it does

Give it moves, and it tells you what the cube looks like and what just got solved.

import { CubeEngine } from "cube-state-engine";

const cube = new CubeEngine("R U' F R2 D"); // start from a scramble
cube.applyMoves("R U R' U'", { record: true });

cube.isSolved();   // false
cube.state();      // { UPPER: [[...]], LEFT: [[...]], ... }
cube.getMoves();   // "R U R' U'"

It handles the whole standard notation set (faces, wide moves, slices, rotations), works on 3x3 and 2x2, and ships with higher-level tools built on top: state predicates, a move simplifier, and a full solve analyzer that breaks a solve into CFOP or Roux stages.

---

Installation

pnpm install cube-state-engine

Ships CJS + ESM + TypeScript types. Works in Node and in the browser.

---

Zero dependencies

Not one runtime dependency. Nothing to audit, nothing to break, nothing bloating your bundle. The whole engine is plain JavaScript.

---

Core: CubeEngine

import { CubeEngine } from "cube-state-engine";

const cube = new CubeEngine();                    // solved 3x3
const scrambled = new CubeEngine("R U R' U'");    // start from a scramble
const pocket = new CubeEngine("", { size: 2 });   // 2x2

Applying moves

The friendliest way to drive the cube is a notation string:

cube.applyMoves("R U2 R' D' Rw M' x y2", { record: true });

Supported notation:

Kind	Tokens
Faces	U D L R F B
Wide	Uw Dw Lw Rw Fw
Slices	M E S
Rotations	x y z

On record: applyMoves(seq) on its own does not record into history handy for setting up a position. Pass { record: true } when you want the moves logged. Unknown tokens are skipped silently.

Every move also has a direct method, all of which record by default:

cube.rotateU(true);   // U
cube.rotateR(false);  // R'
cube.rotateY(false);  // y'

Full method list

Method	Description	Returns
constructor(scramble?, { size? })	New cube, optionally scrambled (not recorded). size is 2 or 3 (default 3).	CubeEngine
applyMoves(sequence, { record? })	Apply a notation string. Not recorded unless record: true.	void
state()	Current state as six face matrices.	object
isSolved()	Whether every face is a single color.	boolean
getMoves(asString?)	Move history — string (default) or array.	string | string[]
reset()	Back to solved, history cleared.	void
rotateU/D/L/R/F/B(cw?)	Turn a face layer.	void
rotateUw/Dw/Lw/Rw/Fw(cw?)	Turn two layers (wide).	void
rotateM/E/S(cw?)	Turn a middle slice.	void
rotateX/Y/Z(cw?)	Rotate the whole cube.	void

On a 2x2, wide and slice moves are no-ops they have no inner layer to turn.

The state shape

state() returns six faces, each a size × size matrix of color letters (W O G R B Y):

{
  UPPER: [["W","W","W"], ["W","W","W"], ["W","W","W"]],
  LEFT:  [[...]], FRONT: [[...]], RIGHT: [[...]], BACK: [[...]], DOWN: [[...]]
}

Perfect for driving a 3D model map each sticker straight onto your mesh.

---

Predicates: "is this solved yet?"

Ready-made checks for the milestones people actually train. Every one accepts a CubeEngine, a state() object, or a flat sticker array pass whichever you already have.

import { isCrossComplete, isF2LComplete, isLastLayerOriented, matchesGoal } from "cube-state-engine";

isCrossComplete(cube);        // cross done (defaults to the DOWN color)
isF2LComplete(cube);          // cross + all four pairs
isLastLayerOriented(cube);    // OLL done on the U face
matchesGoal(cube, "oll+cp");  // OLL + corners permuted

matchesGoal is the flexible one great for building trainers:

matchesGoal(cube, "full");                        // fully solved
matchesGoal(cube, "cross");                       // cross only
matchesGoal(cube, { goal: "oll", face: "D" });    // OLL, but on a different face
matchesGoal(cube, (c) => c.isSolved());           // or your own predicate

Goal names: full / solved, oll, oll+cp, f2l, cross.

🧭 Every check compares stickers against the current center of their face, so it stays correct through x/y/z rotations, wide moves, and AUF exactly what real speedsolves are full of.

---

Move simplifier

The engine records U2 as two separate U turns. simplifyMoves joins them back up:

import { simplifyMoves } from "cube-state-engine";

simplifyMoves("U R R D R' B");      // "U R2 D R' B"
simplifyMoves(["y'", "y'", "L"]);   // ["y2", "L"]
simplifyMoves("F F'");              // "F F'"  ← unchanged, no cancellation

It only ever merges two identical quarter turns into their double. It will not cancel opposites or fold net rotations, so your move count stays honest. Accepts a string, a token array, or timed moves ([{ m, t }]) and returns the same shape.

---

Solve analyzer

Hand it a timed solve and it reconstructs the whole thing: the scramble (derived as the inverse of the solution), the method used, and when each stage finished.

import { analyzeSolution } from "cube-state-engine";

const result = analyzeSolution([
  { m: "U", t: 0 },
  { m: "R", t: 210 },
  // ... `t` is cumulative elapsed ms
]);

{
  method: "CFOP",
  solved: true,
  total: 14371,
  tps: 5.08,
  startFace: { face: "D", color: "W" },
  cross: { color: "W", at: 2340, duration: 2340, moveIndex: 14, move: "U'" },
  f2l: [
    { slot: "G-O", at: 3660, duration: 1320, moveIndex: 19, move: "F"  },
    { slot: "B-O", at: 5160, duration: 1500, moveIndex: 23, move: "B'" },
    { slot: "B-R", at: 8460, duration: 3300, moveIndex: 35, move: "B"  },
    { slot: "G-R", at: 9871, duration: 1411, moveIndex: 48, move: "F'" }
  ],
  oll: { at: 12540, duration: 2669, moveIndex: 63, move: "R'" },
  pll: { at: 14371, duration: 1831, moveIndex: 81, move: "D'" }
}

• Detects CFOP and Roux automatically. Roux solves come back with firstBlock, secondBlock, cmll and lse instead (each block noting the side color it was built on). When neither staging fits, method is "unknown" and the timing fields are null.
• startFace names the face the solve was started from, the same way for every method: the cross face on CFOP, the first block's face on Roux. face is one of U L F R B D (its position at the end of the solve, so rotations mid-solve don't change it) and color is that face's center color. null when no staging was detected.
• allCrosses reports when the cross completed on every face color, not just the one you used.
• unsupported lists any move tokens it couldn't parse, so bad timings never pass silently.

Also exported: invertSequence(tokens), which reverses and inverts a move list.

---

Advanced

For anyone building on the internals:

import { getMovePermutations, getCubeGeometry } from "cube-state-engine";

getMovePermutations(3);  // { U: { cw: [...], ccw: [...] }, ... }
getCubeGeometry(3);      // derived edges, corners, neighbors, opposites

The geometry is derived, not hardcoded a sticker is displaced by a face turn only if it physically belongs to that face, so grouping stickers by which turns move them reconstructs every edge and corner slot on its own.

---

License

MIT © Bryan Lundberg