You never enter an answer in a physics puzzle. You arrange one — an angle, a cut, a strut, a stack — and then hand it to a simulation that resolves it in front of you. Between the moment you commit and the moment the level settles there is no input that can rescue a bad plan. That handover is the genre.
The Incredible Machine (1993) ran all three beats decades before touchscreens: assemble a Rube Goldberg chain out of ramps, balloons, pulleys and the occasional cat, press start, watch it fail somewhere instructive. It is not the ancestor of everything here — the aim-and-fling wing owes more to artillery games — but it is where arrange, release and read first became a genre's entire verb set.
A failed attempt carries more data than a successful one: it tells you which direction you were wrong in and roughly by how much. That only works while failure stays cheap, which is why instant retry is the genre's real contract — one tap from wreckage to a fresh attempt, no penalty screen, no loading, no ammunition to re-earn. Make failure expensive and players stop experimenting and start being careful, which is the worst way to play one of these.
It is also why so many of them score elegance separately from completion. Stars for spare birds, medals for unused rope, a materials budget you came in under: finishing a level and mastering it become two puzzles on one layout, which doubles the content without doubling the levels.
Solutions arrive here as a hunch you have to test. You can see the shape of an answer, but you do not know it holds until you press go — which is exactly what separates this from a logic puzzle, where the answer is certain in your head before you push a single crate. Provisional against proven: both are satisfying, and they are not the same satisfaction.
To plan at all you have to run a rough simulation in your own head — that looks heavy, that will swing, that will roll — and be right often enough for the effort to be worth it. Games earn that with materials that behave: wood splinters, glass shatters, stone stands, and the same crate falls the same way in level two and level ninety. Every surprise should come out of the arrangement and never out of the rules.
The catch: the simulation that makes the genre possible can also dissolve it. Demand near-pixel precision from a system with noise in it and two near-identical attempts diverge, at which point reasoning degrades into rerolling — you stop adjusting and start hammering retry until the numbers fall your way. A physics puzzle you cannot predict is not a puzzle. The fix is tolerance: build levels where an answer that is roughly right is right.
The simulation is not the hard part, and it has not been the hard part since 2007, when Erin Catto released Box2D under the zlib licence — free for commercial use, no royalty, no attribution required in a shipped binary. Rigid bodies, joints, friction, restitution, stacks that settle instead of jittering: a research problem became a download.
What sits downstream of it is enormous. Box2D is what Angry Birds was built on, and it is the 2D physics inside Unity, GameMaker and LÖVE, so a great many developers ship it without ever naming it. Chipmunk2D is the other common choice. Either way the physics is a dependency, not a project.
The consequence is worth stating plainly: nobody will be impressed by your simulation, because they have already played games running the identical one. The design work lives entirely in the object set, the level and the tolerance you allow — never in the physics. And when the simulation stops being the problem and becomes the joke, you have crossed into physics party games, which run on the same libraries and point them at your friends.
Type: Physics puzzle (the simulation resolves it; you only set it up)
Core loop: Arrange or aim, release, read the failure, change one variable
Perspective: 2D side-on, usually — a force has to be visible to be planned for
Win / lose: Win on a goal state: delivered, toppled, arrived. Failing costs a
retry and nothing else
Session: Half a minute to a few minutes per level, in chapters built on
one new object each
Progression: New objects and forces gate later levels; stars rate elegance
The line to argue about is Win / lose. Decide how wide the goal state is
before you build a single level, because a simulation will cheerfully accept a
target one pixel across and your players will not.