Pick up a cardboard triangle, close your eyes, and let a friend rotate or flip it. Open your eyes: it looks exactly the same. The complete list of moves that leave a shape looking untouched is its symmetry group — and that list is not just a pile of tricks, it has arithmetic. Do one move, then another, and the result is always somewhere on the same list. Here you can drive the moves yourself, watch them trace a Cayley graph (a subway map of the group), and fill in the multiplication table that records "this move then that move equals which move." The triangle's group is \(D_3\); the square's is \(D_4\). Same idea, richer table.
Start with the loosest possible idea of symmetry: a symmetry of a shape is any rigid motion that maps the shape onto itself — you could not tell, from the outline alone, that anything happened. For a regular triangle those motions are exactly three rotations (by 0°, 120°, 240°) and three reflections (across the three axes through a vertex and the opposite edge's midpoint). That's six motions. For a square there are four rotations and four reflections — eight. In general a regular \(n\)-gon has \(2n\) symmetries, and this collection is called the dihedral group \(D_n\).
Why "group," and not just "set"? Because these motions combine, and the combination has structure. Do one symmetry, then another: the net effect is still a symmetry of the shape — you never land on something new (this is closure, and it's exactly why six presses of the triangle's buttons can only ever produce six arrangements, never a seventh). There is a do-nothing motion, the identity \(e\). Every motion can be undone by another motion on the list (its inverse): rotations undo by rotating back, and every reflection undoes itself. And composition is associative. Those four facts — closure, identity, inverses, associativity — are the entire definition of a group. A group is not a mysterious abstraction bolted onto symmetry; it is simply the bookkeeping that symmetry already obeys.
The multiplication table (or Cayley table) writes this bookkeeping down: in the cell for row \(a\), column \(b\), you record the single motion equal to "do \(b\), then do \(a\)." Two features leap out. Every row and every column is a permutation of the whole element list — no motion appears twice in a line (this is the "Latin square" property, and it is Cayley's theorem in miniature: each group element acts as a shuffle of the others). And the table is not symmetric across its diagonal, because \(D_n\) is non-abelian for \(n\ge 3\): "flip then rotate" generally differs from "rotate then flip." The precise way they differ is captured by the relation \(srs = r^{-1}\) — conjugating a rotation by a reflection reverses it. Everything in the table follows from just \(r^n=e\), \(s^2=e\), and that one twisting rule, which is why we say \(D_n=\langle r,s \mid r^n=s^2=e,\; srs=r^{-1}\rangle\).
The Cayley graph is the same information drawn as a map. Put one dot per group element, and from each dot draw a coloured arrow for "multiply by a generator": a blue arrow for \(r\) (these chain into a ring — the rotation cycle) and a gold edge for \(s\) (these are their own reverse, so they're drawn as rungs linking the rotation ring to its mirror copy). Walking the graph is composing elements; a closed loop back to the start is a relation like \(r^3=e\). This is the deep payoff: an abstract algebraic object becomes a concrete, finite, navigable network you can literally trace with a finger. And it generalises far beyond cardboard — the same machinery classifies crystals, catalogues the ways a molecule can vibrate, powers error-correcting codes and public-key cryptography, and underlies the Standard Model's particle families. Learn to see the group hiding inside a triangle, and you have learned to see the hidden arithmetic of sameness everywhere.