Setup and opening position
The nine cells show side counts and the twenty-four grid edges begin undecided. Each edge can be marked as part of the loop or excluded; the seeded presentation shuffles the decision order without changing the puzzle.
Classify every edge so the clues are exact and the selected edges form the puzzle's single solver-verified loop.
Your unfinished game is saved only in this browser.
Classify every edge so the clues are exact and the selected edges form the puzzle's single solver-verified loop.
Complete guide
Loop Puzzle is a complete browser implementation with its own versioned rules profile. Draw one closed loop around a 3×3 field while satisfying the number printed in every cell.
The nine cells show side counts and the twenty-four grid edges begin undecided. Each edge can be marked as part of the loop or excluded; the seeded presentation shuffles the decision order without changing the puzzle.
A cell's clue is the exact number of its four sides used by the loop. Every touched grid vertex must have degree two, and all selected edges must belong to one connected cycle rather than several smaller loops.
Classify every edge so the clues are exact and the selected edges form the puzzle's single solver-verified loop. Before closing the final gap, trace the entire route once; closing a short local circuit can isolate the remaining selected edges outside it.
Start with corner clues, where only two sides are available, then propagate degree-two decisions from vertices that already have two selected or two excluded incident edges. Matching every numeric clue is not sufficient if the selected edges split into disconnected cycles or leave a vertex with only one line.
Classify every edge so the clues are exact and the selected edges form the puzzle's single solver-verified loop.
A cell's clue is the exact number of its four sides used by the loop. Every touched grid vertex must have degree two, and all selected edges must belong to one connected cycle rather than several smaller loops.
Start with corner clues, where only two sides are available, then propagate degree-two decisions from vertices that already have two selected or two excluded incident edges.