A corral book punishes a bad file in a way a sample puzzle never shows. Your buyer judges puzzle forty-one, when the key's loop lets a printed 8 see nine cells, or a loop in the key crosses itself on a page that promised it never would, or a puzzle finishes cleanly with a second, different loop that satisfies every number just as well. Corral — the Nikoli logic puzzle also published as Bag and as Cave — looks like the calmest puzzle in the world: draw one fence around the numbers. It is a different contract, in five parts, each part exact. The standard puzzle prints a rectangular grid with numbers in some of its cells. The solver's job is to draw the loop using nothing but the rules. First, the loop rule: draw a single, closed loop along the grid lines, never crossing or touching itself. Second, the enclosure rule: the loop must enclose every numbered cell — no number may be left outside. Third, the visibility rule: each number states exactly how many cells inside the loop can be seen from that number's cell, looking horizontally and vertically, counting the number's own cell, and stopping at the loop, which acts as a wall. Fourth, the outside rule that follows from the first three: every cell outside the loop is simply outside — the loop decides, cell by cell, what is in the corral and what is not, and in the equivalent Cave shading form, all outside cells must stay connected to the edge of the grid and all inside cells must form one connected area. Fifth, tying it shut: exactly one solution, reachable by deduction alone, with no guessing. A puzzle that breaks any part is not slightly harder. It is false as stated. A good corral book is a verified set: every number on every page recounted against its own key's loop, every loop checked that it is single and closed, and every puzzle certified unique by an independent solve.
One warning belongs at the front, because it is the mistake this genre makes most often. The number counts what can be seen inside the loop, not the length of the loop, not the size of the enclosure, and not what can be seen in all directions added carelessly. The count is the sum of four straight sight-lines — up, down, left, right — each running from the number's cell to the loop wall, with the number's own cell counted once, not four times. A book that states the count any other way — as the area of the corral, as the perimeter, as a count that forgets the number's own cell — has made its own puzzles unsolvable as printed, because its keys were built to the true rule the copy denied. State the rule the Corral way, exactly, on the rules page and on the listing: one loop, every number inside it, each number seeing exactly that many cells in straight lines, itself included — and nothing vaguer.
The book is the puzzle set, not the sample grid
Buyers do not solve your listing image. They judge the book by whether each puzzle can be finished as stated — one loop, every number enclosed, every visibility count exact, the key at the back agreeing with the page at the front. That sounds obvious until you look at how corral files are usually made. The common production shortcut is to draw a pleasing loop at random, count what each enclosed number happens to see, print those counts, and ship. A random loop guarantees a key exists — the generator is looking at it — but it guarantees nothing else. It does not guarantee the loop is the only one those numbers allow; loops drawn first and counted afterwards routinely admit second and third completions, because the counts were read off the loop instead of forcing it. It does not guarantee the puzzle can be solved without guessing; a solver who cannot see the key needs a chain of deductions the random loop may never offer — a number whose sight-lines are all long gives the solver almost nothing to start from. And it does not guarantee the numbers are honest: a count copied one cell off during layout, from the neighbouring puzzle, quietly changes what the loop must do while the printed grid still looks perfectly plausible.
A puzzle book is a promise about process, not just about answers. The app can afford a guess because the app restarts, and the online player can afford an undo button. The book buyer who hits a forced guess on paper has no restart — the page is used, the pencil marks are down, and the failure reads as the book's fault, because it is. So the unit of quality is not the puzzle that looks right. It is the puzzle whose solve was walked, wall by wall, from its own printed page, by someone or something that was not allowed to peek at the key. If your production process cannot describe how each puzzle was proven to have exactly one completion, the book is a stack of unverified claims with page numbers.
Sell the set for what it is: a graded run of puzzles, each with a stated grid size, each number printed on the page, each key at the back drawing the one loop. Say on the listing which sizes are inside, say that every puzzle is solvable by deduction alone, and only say it if the verification below actually happened.
The contract, part by part
The loop rule fixes the shape of the whole answer. There is exactly one loop, it is closed, and it runs along grid lines without ever crossing or branching. This sounds like bookkeeping until you see what it forbids: two separate small loops, one around each cluster of numbers, can satisfy every visibility count on a badly made puzzle while breaking the one promise that makes Corral Corral. The single-loop condition is a solving tool, not decoration — late in a hard puzzle, the fact that the fence must come home as one loop is often the fact that closes the last deduction, and a key whose loop is secretly two loops has broken the solver's final instrument.
The enclosure rule is what the loop is for. Every numbered cell must lie inside, without exception, and every number therefore casts its sight-lines from inside the corral. A number left outside in the key is not a small error: that number's count was computed, if it was computed at all, against walls that were never meant to bound it, and the solver — who is correctly trying to enclose it — is solving a different puzzle from the one the key answers. Any edit to a loop segment, anywhere, can enclose or expel a number at the far end of the grid. There is no such thing as a small fix to a finished puzzle — there is only recounting.
The visibility rule is the puzzle itself. Each number is a surveyor standing in its own cell, looking in four straight lines until a wall stops the view. The count includes the surveyor's own cell once, plus every inside cell visible before the loop in each direction. This is the rule keys break most quietly, because a loop nudged one cell outward along a single row changes the counts of every number sharing that row's sight-line — sometimes three numbers at once — while looking, to a tired eye at layout size, exactly like the loop it replaced. It is also the rule that makes small numbers so powerful and so fragile: a printed 1 sees only itself, so all four of its neighbours must be outside, walled off on every side; a printed 2 has exactly one inside neighbour in exactly one direction. Get either wrong by one cell and the puzzle's opening move — often its only opening move — is a lie.
The outside rule rides on the loop whether your book states it or not, and honest books state it. The cells outside the loop are not a leftover; in the Cave form of the same puzzle they must all connect to the grid's edge, and the inside must be one connected area, because a loop that is truly single and closed guarantees both. This matters in verification: a supposed solution that strands a pocket of outside cells in the middle of the grid, walled off from every edge, cannot be bounded by one simple loop, so something in it is already wrong — a crossing, a branch, or a second loop hiding in the drawing. Solvers use this constantly, even when they cannot name it: a corridor of outside cells must always keep an escape route to the edge.
Exactly one solution ties the contract shut, and the no-guess promise rides on it. Two different loops can satisfy every printed number more often than intuition suggests, especially where numbers are sparse and a stretch of wall can bulge one cell outward without crossing any number's sight-line. When that happens, a buyer can solve the puzzle "wrong" — one loop, every number enclosed, every count exact — and still disagree with the key. That buyer did nothing wrong. The puzzle did. Uniqueness is a property you test by solving, not a property you assume from generation.
Verification you can actually run
None of this needs trust. It needs three mechanical passes, run on the final pages, not on the generator's intentions.
First, the recount pass. Take the key — the loop alone — and recount the page from it: confirm the loop is single, closed, and uncrossed; confirm every numbered cell lies inside it; then, for each number, walk its four sight-lines to the loop, count the cells including its own, and confirm the total equals the printed number. Any mismatch anywhere fails the puzzle. This pass is pure counting and catches every layout accident, every nudged wall, every stale key left over from an earlier draft of the puzzle. Run it on the loop as printed, not as designed: the file the buyer receives is the only version that counts.
Second, the independent solve. Solve the puzzle from its printed page using deduction only — a 1 walls itself in on all four sides; a number whose remaining possible sight-lines have narrowed to one arrangement forces those walls; a stretch of loop that must connect, because the fence is one loop, forces the segments that join it; an outside cell must keep a path to the edge, which forbids walls that would strand it — and keep solving until the loop is finished or no rule fires. A finished solve proves the puzzle is guess-free along at least one honest path. A stalled solve with walls unplaced proves the opposite, no matter how the puzzle was generated. Ideally a second solver, human or program, walks it independently; two different deduction paths reaching the same loop is the strongest practical evidence a small publisher can get.
Third, the uniqueness check. The independent solve doubles as a uniqueness test when it is run strictly: if deduction forces every wall, the loop it reaches is the only one deduction allows, and any alternative loop would have to contradict a forced step. Where a solver uses search, run it to completion and count solutions — the answer must be exactly one. Puzzles that fail are not repaired by adding a hint or redrawing one wall in the listing copy. They are regenerated or dropped, and the page count is whatever survives verification. A forty-puzzle book that is forty times true beats a hundred-puzzle book with six liars in it, every time, because the buyer meets the liars.
Run the same passes on the key pages themselves. Keys get transposed, mirrored, and pasted from the wrong puzzle during layout more often than any other error in puzzle publishing, and numbers get redrawn one cell off when a page is reflowed. The grid on the puzzle page, the loop in the key, and the grid in the production file must be the same puzzle, cell by cell and number by number — and the count of numbers inside the key's loop must equal the count of numbers on the page, because a key that encloses one number fewer has broken the enclosure rule before the solver draws a single wall.
Difficulty is not just grid size
Listings love to sort puzzles by grid size and stop there, as if a 12 by 12 grid were automatically harder than an 8 by 8. Size matters, but it is the bluntest of the levers, and a book graded by size alone will feel uneven to every buyer who works through it in order.
Number placement is where real difficulty lives. A small number near the middle of the grid is a wall-building machine: its short sight-lines force loop segments in four directions early, and those segments become the fixed points the rest of the solve leans on. A large number — one that sees most of its row and column — constrains the loop almost as tightly from the other side, because its walls must stand at the maximum distance or the count can never be reached. Sparse puzzles, with few numbers spread thin, are the genuinely hard ones at any size: whole regions of the grid are governed by no number at all, and only the single-loop and outside-connection rules discipline them, which is advanced solving by any honest account. An honest book uses small-number, well-anchored puzzles early, as teaching puzzles, not buried at the back as filler, and says plainly when a section turns sparse.
Sight-line pressure bends difficulty in a way size never captures. Numbers that share a row or column see each other, in effect: one number's wall is inside another number's view, so their counts must agree about the same stretch of grid, and a solver who satisfies one has half-solved the other. Clustered numbers therefore solve faster than their size suggests, while numbers placed so that no sight-line ever meets another create four independent problems on one page — much harder, and much more frustrating when the book implied otherwise. Neither is simply "harder" in the abstract — they are different pressures, and a graded book says which pressure each section leans on instead of pretending the page order is a difficulty curve because the grids grow.
Opening pressure differs from most puzzles, and listings should say so plainly. Corral prints no part of its answer: every puzzle starts as a bare grid of numbers, so the opening is entirely about the smallest numbers and the longest ones — how early the first wall is forced, and how long the solver works by elimination before it comes. A variant puzzle that prints a stretch of loop as a given is a different, gentler product, and it must be labelled as one, not mixed silently into a no-givens book. A book graded honestly says how its puzzles progress — by size, by number density, and by how interlocked the sight-lines are — in an order a buyer can actually progress through. The buyer who wanted a gentle evening book and got a wall of sparse grids does not leave a neutral review, and the expert who bought a book of dense, self-solving clusters leaves a worse one. Both were sold the wrong book by copy that only talked about size.
Print honesty: the wall must be drawable
In Corral, the printed grid is not decoration on the puzzle. It is the puzzle's working surface. Every decision the solver makes is placing or forbidding a wall segment on a grid line, so print quality is not cosmetic — a grid line too faint to draw over confidently is a move the solver cannot record, and a number misread is a changed puzzle.
Legibility starts with the grid itself. Ordinary grid lines must be light enough to read as background but present enough to follow; the solver's own loop, drawn in pencil along those lines, must be unmistakably heavier than the printed grid once drawn, or the half-finished page becomes an argument about which lines are walls and which are paper. Numbers must be printed in a form that survives greyscale and cannot be misread at working distance — a 1 that looks like a 7, or an 8 that fills in to look like a 9 on a cheap home print, is not a difficulty feature. Proof the actual printed size, not the zoomed screen, and read every number on the proof against the production file, digit by digit.
The solver's own marks join the system. A corral page is worked in three states — wall, definitely no wall, undecided — usually a drawn segment, a small cross or dot on a forbidden segment, and a blank. Leave grid lines clean enough to hold a pencil wall and a tentative mark without crowding the numbers in the cells they border, and print the key's loop in a line weight that cannot be confused with the grid at a glance. Proof one full puzzle, printed, worked in pencil, erased once, and worked again. That battered page is the product. The pristine PDF preview is not.
Say on the listing that the book is greyscale-safe only if you have proofed it as such, because in this genre that claim is a solving claim, not a design claim: there is no colour to lose in a corral book, which is its quiet advantage — but size, contrast, and number clarity are claims of exactly the same kind, and they are true only of the proofed page.
Licensing, provenance, and the errata rule
Puzzle books get bought by more than solo evening solvers, and the licence is part of the product. State, in plain language on the listing and inside the file, what each kind of buyer may do. Solo use: one buyer, their own puzzles, forever — say so. Family use: a household printing copies for its own table is the use most buyers assume; either grant it or say clearly that you do not, because silence here reads as permission followed by disappointment. Classroom use: a teacher printing a puzzle for thirty students is republication at small scale, and it is either a tier you sell, a use you grant, or a use you forbid — pick one and write it down. Club and care settings — puzzle clubs, libraries, memory-care groups, waiting rooms — live on reprinting the same puzzle for new hands; a licence that ignores them ignores some of this genre's best buyers. Republication — another seller, a newsletter, a worksheet site printing your puzzles for their audience — is the one use to reserve explicitly unless you are selling it, because a puzzle set is trivially copyable and your only practical protection is that the terms were stated before the sale.
Provenance matters in this genre now, and honest sellers say it. State whether the puzzles were made by hand, generated and then hand-verified, or generated and machine-verified — and what "verified" meant in your process: the three passes above, or something less. Buyers of puzzle books have learned, from a flood of unverified generated books, to ask whether anyone ever solved the puzzles in the file. The seller who can answer precisely — every puzzle deduction-solved from its printed page, every number recounted against its key's loop, uniqueness counted, not assumed — is answering the exact question the disappointed buyers of other books taught them to ask. Do not claim a hand-made book for a generated one, and do not claim verification you did not run; both lies are found out on puzzle forty-one, in public, in a review.
The errata rule is the promise that closes the sale. Name the failures that earn every buyer a corrected file, and mean the list: a number whose key count is off by even one cell; a key loop that crosses itself, branches, or is secretly two loops; a numbered cell left outside the key's loop; a puzzle that forces a guess; a puzzle with a second loop that satisfies every number. Any one of those is not a difficulty spike — it is a broken puzzle, and the fix goes to every buyer, not only the one who wrote in. State where corrections are announced and how a buyer gets the corrected file without buying again. A book with a real errata rule can survive one bad puzzle. A book whose seller argues with the buyer who found it cannot survive its reviews.
Delivery is the last, quiet part of the contract, and it is where a download store earns its fee or does not. The buyer should get the file immediately, from a link that does not expire before they have had a fair chance to use it, with a way back to their purchase when the first download is lost to a full downloads folder. This is the part FileFreight exists for: flat-rate delivery for digital sellers, with expiring links you control and license keys where a product needs them, at a price that does not grow with every sale. Whichever delivery you use, test it as a buyer: buy your own book, on a phone, on home wifi, and print the file you actually receive — not the production file — before the listing goes live.
The launch checklist
Before the listing goes live, walk the book once, in order, as its harshest buyer:
- Rules stated the Corral way — one closed loop along the grid lines, every number enclosed, each number seeing exactly that many inside cells in straight lines including its own — on the rules page and the listing, with no vaguer count language anywhere.
- Every puzzle recounted from its key: the loop single, closed and uncrossed; every number inside; every number's four sight-lines summing, with its own cell counted once, to the printed number.
- Every puzzle independently deduction-solved from its printed page, with no guess at any step.
- Uniqueness counted, not assumed: a completed search or a fully forced solve for every puzzle, with second-loop puzzles regenerated or dropped.
- Print proofed at final size: every number read back digit by digit against the production file, grid lines drawable, and one full puzzle worked in pencil, erased, and worked again.
- Licensing stated in plain language — solo, family, classroom, club and care, and republication — inside the file and on the listing, with provenance stated honestly.
- Errata rule printed in the book, the corrected-file promise named, and a test purchase made, downloaded, and printed from the actual delivered file before launch.
Top comments (0)