A Samurai book punishes a bad file in a way no sample puzzle ever shows. Your buyer judges puzzle forty-one, two hours into a solve, when a digit they placed with confidence in the top-left grid's bottom-right box — a box that is also, cell for cell, the central grid's top-left box — quietly contradicts a row on the other side of the overlap, in a grid they have not looked at for forty minutes — or when an answer key regenerated from five separate grids prints a shared box twice with two different digits in it, so the key disagrees with itself on the same nine cells — or when the whole figure is printed so small that the overlap boxes, the only doors between the five grids, blur into the ordinary gridwork and the buyer solves five puzzles that were never separate and never joined. Samurai Sudoku, in the form this book sells, is five classic puzzles sharing four boxes: five 9 by 9 grids set into a 21 by 21 frame — one in the centre and one at each corner — where every grid obeys the classic law in full, digits 1 to 9 once in every row, column and outlined 3 by 3 box of that grid, and each corner grid shares exactly one 3 by 3 box with the central grid, those four shared boxes and nothing else joining the five grids together. The frame holds 369 active cells — five grids of 81, minus the 36 shared cells counted twice — and the parts of the 21 by 21 frame that belong to no grid are not cells at all and are never filled. Underneath it is a contract in five parts, each part exact. First, the row rule, in every grid. Second, the column rule, in every grid. Third, the box rule, in every grid. Fourth, the overlap rule: a shared cell holds one digit, the same in both grids that own it, because it is one cell, not two. Fifth, closing the contract, given integrity with uniqueness: the givens are fixed exactly as published, and exactly one completed figure satisfies every part at once, reachable by deduction alone, with no guessing. A puzzle that breaks any part is not slightly harder. It is false as stated. A good Samurai book is a verified stack of overlaps: every shared box checked from both of its grids, on the final file, puzzle by puzzle.
Four warnings belong at the front, because they are the mistakes this genre makes most often, and all four start with puzzles Samurai is not. Samurai is not five independent sudokus printed on one page — a book generated as five separate puzzles and merely arranged in a cross has five solutions multiplied together, givens in a shared box that were chosen twice and agree only by luck, and no right to the name: the overlaps are the puzzle, and a generator that does not solve across them has not generated a Samurai at all. Samurai is not one giant 21 by 21 sudoku — no row of the frame is a house, no column of the frame is a house, and the digits do not run 1 to 21; a checker that lays frame-wide rows across the empty corners is judging a law nobody promised, and it will condemn sound puzzles and bless broken ones with equal confidence. Samurai is not Gattai with any number of grids — the standard figure is exactly five grids and exactly four overlaps, each overlap exactly one full 3 by 3 box shared with the central grid alone; corner grids never touch each other, and a figure whose grids overlap by a row, a column or a single cell is a different construction wearing this one's name. And a shared box is not two boxes that must merely agree — it is one set of nine cells read by two grids at once, so it has one set of digits, one given pattern, and one place in the answer key; a workflow that stores it twice, once per grid, has built the disagreement in at the filing stage, and every export after that is a coin flip.
The book is every overlap, not the sample puzzle
A sample puzzle proves almost nothing about a Samurai book, because the law lives in the joining, and the joining is invisible until somebody works through it. Take the central grid's top-left box. In the central grid it is nine ordinary cells. In the top-left grid those same nine cells are the bottom-right box, the last box a solver finishing that grid will fill, and the first news the central grid receives. A digit placed there is placed in two grids in one pencil stroke: it closes a row in one grid and opens a column in the other. No single grid announces that pressure. It appears only when a solver carries information through the overlap — which is the whole solving experience this genre sells. They do not solve your sample. They solve your fortieth figure at midnight, three grids deep, trusting that the box they filled an hour ago still agrees with the grid they have only just reached — and when it does not, they do not think the puzzle is hard. They think the book is careless, and they are right, because the overlaps are not an edge case of the product. They are the product, repeated four times per puzzle, hundreds of times per book.
So the unit of quality in this genre is not the grid. It is the overlap. Every Samurai puzzle resolves into five grids and four shared boxes, and the book is exactly as sound as the worst shared box on its worst page. A book of forty puzzles carries two hundred grids and one hundred and sixty overlaps, and a single shared cell transcribed differently into its two grids — one digit mis-copied between a puzzle page and a key, one given dropped on one side of an overlap in the export — falsifies its box, both of its grids, and the buyer's trust in every figure after it. Sellers who audit grid by grid, glancing at five finished grids, miss this, because each grid can look perfectly classic-sound in isolation while the figure is nonsense: every row, column and box correct somewhere, and the shared cells quietly holding two values at once. The audit has to run at the level the law runs at. Overlap by overlap, grid by grid, book by book — there is no coarser level at which this genre can be honestly checked, and no finer one at which a buyer will ever check it for you before leaving a review.
The contract, part by part
The row rule. In each of the five grids, each of the nine rows holds the digits 1 to 9 exactly once. That is forty-five row instances across one figure, and every one must be audited in its own grid — a row that passes in the top-left grid says nothing about the different row, in the central grid, that shares three of its cells through the overlap. No shared box, however elegant its joining stands, excuses a doubled digit in a row, and no buyer forgives it: the classic frame, five times over, is the floor the whole book stands on.
The column rule. In each of the five grids, each of the nine columns holds the digits 1 to 9 exactly once — forty-five column instances, audited grid by grid like the rows. The column rule matters doubly at the overlaps, because a shared cell stands in one column of its corner grid and a different column of the central grid, and an error that is invisible down one of those columns can be glaring down the other. A checker that audits a shared cell against only one of its two columns has audited half a cell.
The box rule. In each of the five grids, each outlined 3 by 3 box holds the digits 1 to 9 exactly once. Forty-five box instances stand in the figure, but only forty-one distinct sets of nine cells, because four of the boxes are counted twice — each shared box is a box of its corner grid and a box of the central grid at the same time, with the same nine cells and the same nine digits. A mis-drawn box border anywhere is a broken house; a mis-drawn border on a shared box is two broken houses and a broken door between grids, which is why the box audit in this genre is an audit of the drawing as much as of the digits.
The overlap rule. Each corner grid shares exactly one 3 by 3 box with the central grid — top-left with top-left, top-right with top-right, bottom-left with bottom-left, bottom-right with bottom-right — and those four boxes, thirty-six cells in all, are the only cells that belong to two grids. Five consequences follow, and a setter who cannot recite them is not ready to generate a book. First, a shared cell holds exactly one digit: it is one cell, so the corner grid and the central grid do not each get a value for it — there is nothing to agree, because there is only one thing there. Second, the givens in a shared box are givens of both grids at once, printed once, and a given dropped from the figure weakens two grids in one stroke. Third, every deduction in a shared box fires in both grids: placing a digit there eliminates it from two rows, two columns and two boxes — six houses in two grids — which is why overlaps are where Samurai solves are won. Fourth, the corner grids never touch each other: the top-left grid shares nothing with the top-right, and all traffic between corner grids passes through the central grid's boxes, rows and columns. Fifth, the accounting closes: a completed grid's digits sum to 405, so five completed grids sum to 2025 counting shared cells twice, each shared box sums to 45, and the 369 physical cells of a sound figure therefore sum to exactly 2025 minus 180 — 1845 — an identity a setter can check against any finished figure in seconds, and one a transcription slip in a shared box will break from either side.
Given integrity and uniqueness. The givens are exactly the givens as printed — no digit added, dropped or moved between the puzzle page and the answer key, and no given in a shared box recorded for one grid and not the other — and exactly one completed figure satisfies all five parts, reachable by deduction alone. The proof must run with all five grids coupled through the four overlaps, and that figure must survive two controls that protect you from your own toolchain: once with the grids decoupled, each solved as a standalone sudoku ignoring the overlaps, which will report freedom the coupled law never allows, and once with the frame read as one 21 by 21 grid, which invents houses across the empty corners that the standard rule expressly does not demand. If the solution changes under the decoupled control, your main proof was only as strong as the coupling your solver actually performed, and the settings — not the puzzle — are what need fixing before the book ships.
Verify like a buyer: the eight-step audit
Run this audit on the final file — the exact PDF or download you will ship — never on a working draft, because the failures this genre ships live in export: a shared box stored twice and exported inconsistently, a figure shrunk until its overlaps blur, a key regenerated from an older set of grids. Eight steps, in order, for every puzzle in the book.
Step 1: Frame inventory. Map the figure before touching a digit: five grids, their positions in the 21 by 21 frame, 369 active cells, and the frame areas that belong to no grid, marked as void, never as cells. Count the givens per grid and in total. The answer key must carry the identical frame over the identical figure — a key drawn from a second export is a second book, unaudited, hiding at the back of the first.
Step 2: The overlap map. List the four shared boxes, cell by cell: which corner grid, which corner of the central grid, which nine physical cells. Write the map down; every later step works from it. If any pair of grids shares anything other than exactly one full box — or any corner grid touches another corner grid — the figure is not the standard construction, and it must be re-drawn or re-labelled before it is audited further.
Step 3: Shared-cell extraction. For every one of the thirty-six shared cells, record its digit from the completed figure once — once, from the physical cell, not once per grid. Then read that same cell as a member of its corner grid and as a member of the central grid. Any workflow, spreadsheet or key that holds two values for one shared cell has failed here, whatever either value is, because the cell itself can only hold one.
Step 4: Per-grid classic audit. For each of the five grids, in turn, check all nine rows, all nine columns and all nine boxes against the completed figure: digits 1 to 9, exactly once, in every house — forty-five row checks, forty-five column checks and forty-five box checks across the figure, with the four shared boxes honestly checked from both of their grids. A grid skipped because its neighbour passed is a grid unaudited, and overlaps make neighbours of grids that share no border on the page.
Step 5: The accounting recount. Add the completed figure two ways. First by grids: five grids, each summing to 405, total 2025 with shared cells counted twice. Then by physical cells: 369 cells, which must total 1845 exactly, because the four shared boxes contribute their 45 each one extra time in the grid count. Then re-add the shared boxes separately, in reverse order, box by box — sums computed a second way catch the quiet arithmetic slips that a single pass repeats. Any mismatch means an extraction error somewhere, even when step 4 passed, because two wrong readings can agree inside one grid and disagree between grids.
Step 6: Given integrity across the overlaps. Compare the shipped puzzle page against the master, given by given, with the overlap map open: every given in a shared box appears once on the page, belongs to both grids, and matches the key's digit for that physical cell. Count givens per grid from the page, not from the file that generated it — the page is what the buyer solves, and a given lost in export is lost to both grids that owned it.
Step 7: Uniqueness, with controls. Re-prove the figure has exactly one solution with the full law enforced — classic frame in all five grids, coupled through the four overlaps as mapped in step 2. Then run the impostor controls from the contract: the decoupled reading, which solves each grid standalone, and the frame reading, which treats the 21 by 21 as one grid. A puzzle that is unique only under the decoupled reading was verified against a law your rules page does not promise, and its second solution is waiting for a buyer with a pencil and a long evening.
Step 8: The print proof. Print one page at final size, in greyscale, on the cheapest paper your buyers plausibly own. The whole 21 by 21 figure must sit on the page with cells large enough to hold a pencilled digit and its candidates; the grid borders of all five grids must read at arm's length; the four shared boxes must be findable at a glance, because a buyer who cannot see where the grids join cannot carry a deduction through the join; and the void areas of the frame must read as void, not as cells awaiting digits. If the joining cannot be seen, the law cannot be followed, and the puzzle is broken in print however sound it is on screen.
Difficulty is overlap work, not given counts
Two Samurai books can sit side by side on a marketplace, one labelled gentle and one labelled fierce, with near-identical given counts — and opposite solving experiences. Buyers do not forgive a dishonest one — a book labelled gentle whose third figure walls them for an evening, or a book labelled fierce that folds grid by grid in an hour. In this genre the given count is the weakest lever on the shelf, because the law's power concentrates in the joining, not in the clues. Grade difficulty by the overlap work the solve actually demands, and name the levers in the listing.
The first lever is how much of each shared box is given. A shared box printed nearly full hands the buyer a working door between two grids: deductions walk through it in both directions from the first minutes, and both grids solve almost as one. A shared box printed nearly empty seals the door — each side must be worked up to the threshold from its own givens, and the figure solves as five puzzles first and one puzzle late, which is a different and much longer evening. The second lever is central dependence: a figure whose corner grids stall until the central grid feeds them through the overlaps forces the buyer to keep five grids' worth of candidates alive at once, while a figure whose corners largely solve alone and merely confirm themselves in the centre asks for stamina more than insight. The third lever is how deep the classic work runs inside each grid once the overlaps have spoken — a figure whose doors open early and then folds to singles in every grid is a gentle figure wearing a large drawing, while a figure that still demands advanced work in its last grid, hours in, is fierce however friendly its picture looks. The fourth lever is sheer scale honestly counted: 369 cells is five puzzles of labour even when every step is easy, and a listing that calls a full Samurai figure a quick solve is mislabelling time as difficulty. The weakest signal, as always, is the raw number of givens — worse here than in single-grid genres, because the same total can be spread evenly across five grids or piled into three and starve two. Name the levers instead: how full the shared boxes typically print, whether the corners depend on the centre, and the solving time your own testers honestly took. A buyer who knows those three things knows the book they are buying.
Print honesty: a figure a buyer can actually follow
A Samurai book sells a joining law printed as one large figure, and the buyer's whole experience of the law runs through the drawing of that figure. Print is not decoration here; it is the rule made visible, and it has failure modes a screen proof will never show you.
The buyer must be able to see, at a glance and in bad light, exactly where each of the five grids begins and ends, and exactly which four boxes are shared — because a shared box wrongly seen invents a door the law never made, a shared box missed deletes the only door that exists, and either error strands every deduction that needed to cross. Give the figure the page it needs: one puzzle per page at a size where a cell holds a full-size digit with room for pencil candidates around it, and refuse the temptation to shrink two figures to a page to save paper the buyer never asked you to save — a Samurai cell is a sudoku cell, and it needs a sudoku cell's room, multiplied across a 21 by 21 frame. Draw the box borders unmistakably heavier than the ordinary gridlines, heavier than feels elegant on screen, because cheap printing thins everything, and let the shared boxes carry the same border weight in both of their grids — a box drawn heavy on one side of the join and light on the other will be read as two boxes, and solved as two, wrongly. Keep the void areas of the frame clean: no gridlines, no shading that mimics a cell, nothing a tired buyer could mistake for a sixth region awaiting digits. Puzzle page and answer key must draw the identical frame over the identical figure; a key that re-flows its grids, splits the figure across pages the puzzle page kept whole, or prints the shared boxes in two places with two layouts is a second book, unaudited, hiding at the back of the first. And number your figures and grids consistently — a buyer mid-solve, and a seller mid-correction, both need to say top-left grid, central grid, shared box, and be pointing at the same nine cells.
Licensing, provenance, and whose puzzles these are
A Samurai book is easy to assemble dishonestly and hard to assemble well, and the difference is records. Variant sudoku has a living community of setters whose puzzles circulate in packs, portals and streams, with the five-grid standard wording attached. A figure lifted from a portal, re-drawn in your template and sold under your cover is not made yours by the re-drawing, and this community is small enough — and archives its work carefully enough — that provenance is not a courtesy. It is a survival trait.
Keep four records for every book. The rights record: written licence or your own authorship, in a form you could show a marketplace without embarrassment. The source record: where each figure came from, setter by setter, including your own. The audit record: the eight steps, dated, against the final file, with the overlap maps and the per-grid house checks kept so a disputed shared box can be re-read cell by cell from both of its grids. And the rules record: the exact Samurai reading the generator enforced — five 9 by 9 grids in a 21 by 21 frame, four overlaps of one full box each with the central grid only, shared cells holding one digit, classic rules inside every grid — because those settings are where this genre's books quietly diverge. When a marketplace asks where a book came from — and marketplaces increasingly ask — the seller with records answers in minutes. The seller without them discovers that "my generator made it" is a claim, not a record.
Corrections follow from the same records, and they are a selling point, not an embarrassment. If a buyer finds a shared cell that disagrees with itself, believe the physical cell before you believe your file: re-read the box from the shipped PDF, cell by cell, from both grids, and if the buyer is right, fix the master, re-run all eight steps on the corrected figure — not only on the corrected box, because an overlap error touches two grids and rarely travels alone — and re-deliver the corrected edition to every past buyer, not only to the one who wrote in. A download tool that versions your file and re-delivers the corrected edition to past buyers turns an errata notice from a hope into a fact, and it is why I deliver my own puzzle books through FileFreight — one flat monthly price for download delivery and license keys, the corrected file replaces the old one for every buyer at once, and nobody has to write in and ask. Pair the corrected file with a plain note naming the figure, the grid, the shared box, and what changed. Buyers forgive a corrected book. They do not forgive a silently different one, and they do not forgive being told the overlap was fine when their own pencil says otherwise.
The launch checklist
Before the listing goes live, the book must be able to say all seven of these, truthfully:
- Every figure has been inventoried — five grids, 369 active cells, void frame areas marked as void — and the answer key carries the identical frame over the identical figure.
- Every overlap has been mapped: four shared boxes, thirty-six shared cells, each shared with the central grid only, written down cell by cell.
- Every shared cell has been extracted once from the physical figure and read from both of its grids, with no cell holding two values anywhere in the workflow.
- Every grid has passed the full classic audit — forty-five row, forty-five column and forty-five box checks across the figure — and the accounting recount closes at 1845.
- Every figure has exactly one solution under the full coupled law, proven with the overlaps on, and the decoupled and frame-wide impostor controls have been run.
- One page has been print-proofed in greyscale at final size, at one puzzle per page, and the five grids and the four boxes that join them all survive cheap printing.
- The rights, source, audit and rules records exist, and the Samurai setting — five grids, four full-box overlaps, centre only — is written down.
Top comments (0)