How we make our puzzles.

Start with a puzzle people know

Each case file begins with a classic puzzle that solvers already understand, and asks one question: how could the solved puzzle become evidence? A finished word search leaves letters over; they can spell a location. A finished logic grid says where everyone was; that can convict a thief. A finished loop traces a route; the landmarks it passes can break an alibi.

The story is there to make the deduction feel like an investigation. It never replaces the logic, and nothing in a story is a clue unless the case file says so.

Generate, do not hand-write

Cases are not written one at a time. Each game has a generator that builds a fresh case from a seed, a number that is shown to you as the case number. The same case number always rebuilds the same case, so you can note it down, reopen it later, print it, or hand it to a friend to solve the same mystery. Because the supply is made by the generator, there is no end to it.

The names, settings, objects and phrasing come from pools we write and review. The generator chooses from those pools, so two cases can share a setting and still be entirely different puzzles.

Prove it before it ships

A generated case is only a candidate. Before it reaches your screen it has to pass checks that we treat as rules rather than aims. If a check fails, the case is thrown away and the generator tries again.

1. Exactly one answer. The puzzle layer has a single solution, and the detective layer names a single culprit. A second solver, which tries every possibility by brute force, cross-checks this in our automated tests.
2. Solvable without guessing. A solver that uses only named, human deductions (the ones our guides describe) must complete the whole case. If it stalls, the case does not ship. Where a puzzle needs a fixed point before the logic can carry it, the generator adds a given, and only as many as the logic needs.
3. No dead clues. Remove any one clue, statement or evidence card from a shipped case and the answer must stop being provable. Every clue earns its place, and no single clue convicts on its own.
4. Nothing readable early. In games with a code or a decoy layer, the generator checks that the statements alone cannot name the culprit, so you cannot skip the puzzle.

The cross-checks have earned their keep. While building The Red Thread, the brute-force counter found two valid routes on a map the human-style solver believed was finished. The faulty deduction rule was corrected before launch. We keep those tests because programs make mistakes too.

Hints that teach the method

The solver that proves a case is the same one that powers its hints, so a hint is always a step you could have taken yourself. The first press names a deduction and explains the reasoning. A second press on the same spot points at the exact square, and the mark is still yours to make.

Check only flags marks that cannot be right; it never adds any. Reveal shows the whole solution and explains it, but a revealed case does not count towards your cases-cracked tally. A wrong accusation is stamped on the file and the case stays open.

Then a person plays it

Proofs make a case fair. They do not make it clear or enjoyable, so every game is played by a person before launch and again after any change. Karan tests as a new player would, on a phone, looking at the objective, the first move, the controls, the wording and the feel of each difficulty rank.

Play-testing changes things. The warehouse game gained its red-edged rules card after testers kept missing the rule that caches never touch. A correct accusation that felt flat led to the case-closed stamp and lamplight that now end several games. A decode step that read like a form became a jumbled heap of tags you sort yourself. Three games that opened new players on a middle rank now open on the easiest.

The strategy games are judged differently

The Tactical Desk games are matches you can win or lose, so a uniqueness proof makes no sense for them. Their standard is a fair opponent: the computer may never read hidden information, your plans or anything a human in its seat could not see, and each game's page documents what its opponent can see. Difficulty tiers are honest strength tiers, never information leaks. Balance comes from play-testing both sides until each is worth playing.

For Operation Kingpin, where every piece starts face down, we built a reduced board first and tested the opponent before committing to the game: the hard tier beat a careless bot about nine games in ten, a careful bot beat the easy tier about four games in five, and automated tests confirm its choices depend only on what is publicly visible.

Cozy by rule

Every case is a mystery about something missing, swapped or hidden: a diamond, a brooch, a prize, a file. There are no victims, no weapons and no horror. Suspects are cleared, not harmed; culprits are arrested, not hurt. The easier ranks are written for younger solvers playing with an adult, and the harder ranks add deduction, never grimness. This is a house rule, not a mood.

Original work

Classic puzzle forms belong to everyone, and we say plainly which one each game is built on. Everything we add is our own: the rules variations, the detective layers, the names, the stories, the artwork and the instructions. Where a game was inspired by an existing format at the level of its idea, we design our own mechanics and write our own copy, and we check names before launch.

When something slips through

With generated cases, proofs and tests, mistakes are rare, but wording, controls and the odd edge case can still go wrong. If a case does not hold up, if an instruction is unclear, or if something does not work on your device, please tell us. Reports are read by the people who build the games, and corrections become part of maintaining the library.