A logic puzzle hands you a small closed space, the complete set of rules that govern it, and no way to lose. Nothing is hidden, nothing is random, and no part of the answer depends on your hands. The only obstacle is that you cannot see the solution yet — and what separates one of these games from the next is the kind of seeing it asks for. There are roughly four.
Sokoban — 倉庫番, "warehouse keeper", published by Thinking Rabbit in 1982 — gives you crates, goal squares and one asymmetry: you can push a crate but you can never pull it. Shove one flat against a wall and it will not leave that wall again. The thinking is order of operations — which crate first, and which side of it you need to be standing on by the time you get there.
Push every crate onto a goal — arrow keys, or tap the pad
The catch: a room can become unsolvable several moves before it looks unsolvable, so the genre asks you to hold the whole future in your head at once. Reset is instant and carries no shame, but the moment you notice is always later than the moment you erred.
In Hempuli's Baba Is You (2019) the rules are furniture. WALL IS STOP sits in
the level as three pushable word blocks; shove STOP out of the sentence and
walls become scenery you can walk through. You do not solve the room so much as
rewrite the paragraph governing it — make yourself the wall, make the flag
WIN, make water PUSH.
The catch: the space of legal sentences is vast and almost all of it is nonsense, so being stuck feels less like narrowing a search than guessing at grammar. It is close to unhintable, too: any clue specific enough to help gives the whole trick away.
Draw a line from the dot to the exit. Everything else — separate the black squares from the white, collect exactly these, make that shape fit — is a rule nobody told you. The Witness (2016) teaches by sequence: a panel solvable only one way, then the same rule bent slightly, then bent again, without a word of written instruction anywhere on the island.
The catch: when the teaching is wordless, a rule you have misread feels identical to a puzzle that is unfair, and the game has deliberately left itself no vocabulary to correct you with. The sequencing has to be immaculate or the player is simply lost with no way to ask.
Zachtronics' Opus Magnum (2017) accepts any machine that produces the required output, which puts solving at the start of the puzzle rather than the end. Your working answer is then scored on cost, cycles and area, and each score is plotted as a histogram against everyone else's. You do not beat these. You compete with the version of yourself from an hour ago.
The catch: there is no completion state. The histogram always has a tail to the left of you, so nothing in the game ever tells you to stop — exhilarating if you like that, and unbearable if you need a thing to be finished.
All four run one loop — study, hypothesise, test, revise — with the same tools: unlimited undo, instant reset, no clock, no dice, no reflexes. A failed attempt is not a loss, it is a measurement, and that is only true because the space is deterministic. Certainty is the point. In a good logic puzzle you know you are right before you move, which is precisely the line between these and physics puzzles, where a plan stays a hunch until you release it. The temperament turns up outside the puzzle family in turn-based tactics — Into the Breach shows you every enemy's next move and dares you to find the answer — and next door in word puzzles, where a cryptogram is deduction with letters for its material.
What they share is also the risk they share. Stuck is the only failure state and it is total: no partial credit, no luck to blame, no way to grind past it, and a game that will wait forever without offering a rope. Hints deflate the point rather than solve it, because the moment a puzzle clicks cannot be handed over, only found. Which leaves the designer one job: shorten the distance between confused and clicked without ever walking it for the player.
Type: Logic puzzle (deterministic, complete information, no fail state)
Core loop: Read the whole space, find the constraint that is really a
direction, test it, undo, commit
Perspective: Top-down grid, or first-person for constraint puzzles; static
Win / lose: Win by solving. There is nothing to lose — only to reset
Session: One puzzle in a spare moment, or one room all evening
Progression: One new rule at a time, then every consequence of that rule
Pick one of the four and stay inside it. A set of these is an argument — one idea, then every consequence of it — and a game that pushes crates on Monday and rewrites its own grammar on Tuesday has no argument to make.