Last updated: September 14, 2026
This page explains how Sudoku Online checks Sudoku puzzle quality and product-facing puzzle claims. It describes the tests implemented by the current theme and intentionally separates automated checks from subjective human difficulty.
1. Unique-solution verification
Classic Sudoku puzzles are checked with a solver that counts solutions up to a limit of two. A puzzle is accepted as unique only when exactly one valid solution is found. Generated puzzles that fail this check are rejected rather than published as normal playable inventory.
2. Stored solution and clue consistency
The puzzle catalogue also validates the stored solved grid independently. Clues must agree with the stored solution, the completed grid must obey Sudoku row, column, and 3×3 box rules, and invalid or non-unique catalogue rows can be quarantined with a zero quality score instead of silently remaining in the normal playable pool.
3. Difficulty grading
Difficulty is estimated with a human-oriented grader rather than by clue count alone. The Classic grader tracks techniques such as naked singles, hidden singles, locked candidates, naked/hidden pairs, and X-Wing before considering unresolved search work. A generated puzzle can therefore be stored in the difficulty bucket produced by the grader instead of being forced into the level originally requested by the generator.
Difficulty remains an estimate, not a promise that every player will experience a puzzle the same way. Familiarity with a technique, note-taking habits, and solving speed can make the same board feel easier or harder to different players.
4. Killer Sudoku checks
Killer Sudoku keeps the normal Sudoku constraints and adds cage totals and cage membership. The Killer grading path evaluates the actual cage layout with a cage-aware solver. Cage seed/profile information is kept stable so the public puzzle and the grader are evaluating the same intended layout rather than a newly generated cage arrangement.
5. Catalogue quality audits
Administrative quality audits can re-check clue consistency, stored solutions, uniqueness, grader classification, and exact duplicate groups. Rows that fail core validity checks are quarantined instead of being deleted in a way that could destabilize catalogue indexes or references.
6. Gameplay and interface regression checks
Puzzle correctness is only one part of release quality. Shared gameplay controls, mobile layouts, keyboard/focus behavior, saved state, daily/competition flows, locale routing, SEO output, and release-package integrity are also covered by regression contracts in the theme. These checks reduce regressions but do not mean that every browser, assistive technology, network condition, or device combination is guaranteed to be defect-free.
7. What these tests do not prove
Automated validation cannot prove that a difficulty label feels perfect for every player, that every explanation is the clearest possible wording, or that a future browser/plugin update cannot introduce a new issue. We avoid presenting internal QA as a formal independent certification.
8. Reporting a puzzle or content issue
If you find a puzzle that appears invalid, a cage total that looks wrong, a difficulty label that seems clearly mismatched, or a guide example that does not follow logically, send the exact URL and details through the Contact page. Editorial reports are handled under the Content Correction Policy and publishing decisions follow the Editorial Policy.