How our puzzles are verified

1. Connected, complete regions

Each n×n board has exactly n nonempty regions. A flood-fill checks that every cell of a region is reachable from every other cell through shared edges. The checker also validates IDs, cell ranges, region labels and fixed cats. Corner contact does not count as connectivity.

2. Exactly one solution

The browser search chooses row positions while enforcing distinct columns, distinct regions and the no-touch rule. Before publication, an independent Python oracle enumerates all row/column permutations that satisfy adjacency, filters them by regions and givens, and requires exactly one result.

The current 240-board dataset passed that exhaustive check: 80 boards at each supported size. This proves uniqueness under the specified rules, not that every human finds the board easy or that a screenshot was interpreted correctly.

3. Explanations with checkable premises

A solution trace records fixed cats, a single candidate in a group, a candidate lock between two groups, or a cell touching every candidate of a group. Each record names its premises and the cells it removes or fills.

A separate validator reconstructs candidates from the beginning. It checks each rule’s preconditions and exact elimination set rather than merely comparing the ending cats with an answer. Every removed cell must preserve the unique solution. The complete trace must place all cats. Selected original walkthroughs expose those actual steps.

4. Duplicate checks

Region labels are normalized after all eight rotation/reflection transformations. The canonical region shape is hashed and must be new in the collection. Recoloring, renaming or rotating a board cannot count as a new puzzle.

Published IDs are permanent. The first daily schedule assigns 180 different existing puzzle IDs to consecutive UTC dates. It does not generate another board merely because your device clock changes.

What these checks do not prove

Automated rule checks do not replace browser testing, mobile performance measurements, screenshot sampling tests or human difficulty calibration. Difficulty names are currently derived from givens and deduction types. A unique-solution badge on a user-entered board is shown only after the search actually finishes proving uniqueness.

Report a suspect board or explanation with its ID at Contact GridMew. We do not claim search ranking, visitor totals or earnings from these validation results.