The same sentence that governs a rotated glyph defines what a game is: the rules are the allowable transformations, and the game is whatever survives them. In politics the framework was a metaphor. In games — especially puzzle games — it stops being one.
“Identity emerges from invariance” says a game's identity is whatever its legal moves leave unchanged. The board, the pieces, the pixels are the set; the rules are the group acting on it; and “what the game is about” is simply the invariant of that action. Same pieces, different allowable moves — and you have a different game. Chess and a chess variant share every piece and square; what separates them is only the chosen set of transformations.
That makes games the cleanest possible instance of the explainer's load-bearing word. Politics has to argue about which changes count as “still us.” A game just lists them: these moves are legal, those are not. Everything the game means — its strategy, its identity, its fairness — is downstream of that one choice, exactly as the Erlangen program says geometry is downstream of its chosen group.
Four mechanisms — three that fix a game's identity, and one that deliberately moves it — each with a place where the math complicates the story a fanbase tells about itself.
The demonstrations on the main page aren't decorations here — several are precise models of how a game holds, or sheds, identity.
Speedrun categories are this, exactly. glitchless → any% → wrong-warp is enlarging the transformation group and shedding invariants step by step. any% keeps only the coarsest invariant — “the credits rolled” — the projective geometry of the game; 100% keeps far more, the rigid geometry. Each category is a different geometry of the same cartridge.
The sequel dilemma as a conservation law. The more a series fixes (large stabilizer), the fewer forms it can take (small orbit). Recognizability and reinvention are formally in tension; you cannot maximize both. The symmetry group of a board says the same thing — a symmetric map has a small orbit of genuinely distinct positions.
Mechanical skill vs. map knowledge. Aim and game-sense are intrinsic invariants — they transfer from one shooter to the next, stable like the cross-ratio. “I memorized this map's spawns” is a relative invariant, pinned to a build that evaporates on patch day.
Remaster vs. remake vs. reboot: the boundary of what counts as “the same game” is contested, not given — the same assets can read as a faithful update or a betrayal. The 15-puzzle's parity invariant is the literal version: the tiles look identical, but an invariant silently decides which arrangements are even reachable.
A symmetric map or ruleset conserves something real — fairness. Resource economies (mana, gold, tempo) are conserved quantities a designer balances around. Caveat: true symmetries are rarer than they look — chess's first-move advantage is a small, genuine symmetry-break — so treat this as the most aspirational analog, not a literal law.
A Rubik's Cube literally is a group of order ~4.3×10¹⁹: the solved state is the identity element, twists are generators, and “solving” is finding a word in the generators back to identity. Here the explainer's mathematics needs zero translation — the puzzle is the group.
Everything above is descriptive. The very same machinery underwrites a beloved faithful sequel and a cynical reskin; a brilliant glitch-hunt and a cheated leaderboard; a tight competitive ruleset and a stagnant annual one. Invariance explains how a game holds an identity together — it does not certify that holding it is good design, or that the invariants a fanbase defends are the ones worth keeping.
That leaves two pathologies, one on each side. Over-fixing: treating contingent features as sacred, multiplying purity tests, losing the ability to evolve — the brittle orbit of size one (the franchise that ships the same game forever). Under-fixing: no durable core at all, a reboot or live-service churn that keeps nothing recognizable. A living game, like a good representation, stores the invariants that carry meaning and lets the rest vary.
The cleanest test case — the place where “identity = invariant under allowable moves” is not a metaphor at all. Nim is decided entirely by the XOR of the pile sizes: that nim-value is the invariant that says who wins, and the whole Sprague–Grundy theory is the statement that every impartial game has one. The 15-puzzle hides a parity invariant that makes half of all arrangements unreachable. Lights Out is solvable iff the start state lies in the column space of the toggle matrix over GF(2). Rubik's Cube is a group of order ~4.3×10¹⁹ with God's number 20. In each, the explainer's sentence is the literal mathematics: the game is whatever its move-group leaves invariant, and the invariant decides the outcome.
A “fair” game is one invariant under swapping players or sides — mirror matchups, symmetric maps. A tier list is a measured symmetry-break. The meta is the fixed point of strategy-space under competitive pressure (a Nash equilibrium); a balance patch is the designers re-choosing the allowable group to move that fixed point somewhere healthier.
Communities discover the largest group the engine actually admits — every clip, skip, and wrong-warp — far past the designers' intent, then partition it into categories, each a chosen subgroup. A leaderboard is an orbit under an agreed transformation set; banning a glitch is removing an element from the group. The “real” game is renegotiated by who plays it.
Fortnite, Minecraft, Roblox enlarge the allowable transformations continuously — seasons, UGC, new modes. A thinner fixed core, traded for vast reach: Erlangen at the scale of a platform. The live question is the same as politics asked of generations — what survives a season wipe, an engine migration, a UGC flood?
The seed varies every run; the ruleset is the stabilizer. The game's identity is precisely whatever is invariant across every randomized run — the clearest live demonstration of orbit vs. stabilizer.
Players literally rewrite the transformation group. A total-conversion mod is a new geometry on the same engine — sometimes invariant enough to still feel like the base game, sometimes a different game entirely.
How much can you transform the 1997 game — visuals, combat, structure — and still have it be the same game? A recognition struggle over where the franchise draws its invariant.
Pick a franchise or a competitive game. What is its stabilizer — the elements the community will not let change? Which transformations is it currently allowing (a new engine, a new mode), and which is it resisting? The answers usually locate it precisely on the over-fixing ↔ under-fixing axis.
This page is analytic, not advocacy. The framework applies to every genre and community equally, and naming a dynamic is not a verdict on a game, a studio, or a fanbase. Treat it as a lens for seeing the move — never as a ruling on which design is right.