Shikaku (“division rectangles”) hands you a grid dotted with numbers and asks you to slice the whole board into rectangles, one per number, where each rectangle’s area (its width times its height) exactly equals the number it contains. There is no fixed shape for a given clue: a 6 could be a 1×6 strip, a 2×3 block, or a 3×2 block, and any of those is correct as long as it covers exactly six cells and holds exactly that one number.
Every board is generated fresh from a seed baked into the page URL, so a link always reopens the exact same puzzle. Pick a difficulty above to start (from a gentle 5×5 grid up to a demanding 15×15 board carrying larger, harder-to-spot rectangles).
A clue’s number constrains its rectangle’s shape less than you’d expect on its own (a 6 could be a 1×6 strip or a 2×3 block), so what actually narrows a Shikaku board down is how clues sit relative to each other: two clues close together often force a boundary neither could dictate alone. Larger boards don’t just add more rectangles, they add more of those forcing pairs stacked across a bigger grid, so a solve leans more on cross-referencing clues far apart from each other than on any single clue’s number.

Every Shikaku board here ships only once countShikakuSolutions has proven its clue set pins the partition down to exactly one arrangement of rectangles: an inconclusive search, where the solver’s own node budget runs out before it can decide, is treated the same as “not unique” and never accepted. There is exactly one way to slice the board that satisfies every numbered clue, reachable by deduction alone.
When a clue draw fails to pin a partition down uniquely, the generator re-rolls just the clue cells on that same partition first (a cheap retry) before ever resplitting the board from scratch, since a fresh clue draw usually succeeds long before a fresh partition is needed. Every one of those retries is proof-gated the same way: an inconclusive count is never treated as evidence that a partition is safe to ship.