the ground / stratum · Pattern
A children's game is a fair tie: rock, paper, scissors, each beating exactly one of the others, none of them best. Nobody expected to find it written into an animal. But a lizard in the California scrub plays it for real — three kinds of male, none able to win, all kept alive by the very fact that the ring never closes on a champion.
Every spring on the rocky slopes of the inner Coast Range, male Uta stansburiana — the common side-blotched lizard — sort themselves by the colour of their throats. Some are orange, some blue, some yellow, and the three colours are three different ways of being a male, three strategies hard-wired and heritable.
The orange males are bullies: big, aggressive, holding large territories with many females. The blue males are guardians: smaller territories, usually one female, watched closely. The yellow males own nothing — their throats mimic the colouring of a receptive female, and they use the disguise to sneak matings on territory they could never hold.
Now watch what happens season to season. When orange bullies dominate, their territories are so large and so full of females that they cannot guard them all — and yellow sneakers, slipping past in female costume, clean up. So yellow rises. But a population thick with yellow sneakers is exactly what a vigilant blue guard, watching his one female, defeats — a sneaker can't fool a male who never looks away. So blue rises. And a population of blue guards, each holding a small patch, is easy meat for an orange bully who simply takes the patch by force. So orange rises — and we are back where we started.
Barry Sinervo and Curtis Lively reported this in Nature in 1996, and gave it the only name that fits: the rock–paper–scissors game. The morph frequencies oscillate, they found, over roughly a six-year cycle — orange peaks, then yellow, then blue, then orange — driven by negative frequency-dependent selection: each strategy does best precisely when it is rare, and worst when it is common. The game has no equilibrium any one type can rest at. It has a cycle.
Pause on how strange this is for our usual instinct, which is to rank. Faster, stronger, fitter — surely the morphs line up somehow, best to worst? But they don't. "Beats" here is not a ladder; it is a loop. And a loop is the one shape a ranking provably cannot hold.
You can measure exactly how much of a contest a ranking can hold. Treat the head-to-head results as a flow on a triangle — an arrow on each edge pointing from loser to winner — and split that flow into two parts, the way you'd split a wind field into the part that flows downhill from high ground to low (a gradient: a pure ranking, high scores to low) and the part that just circulates with no high or low at all (a curl: a cycle). This is the Helmholtz–Hodge decomposition, and it is exact. A clean ladder is 100% gradient and 0% curl. Rock-paper-scissors is the opposite: 100% curl, 0% gradient. A scalar rating — an Elo, a power ranking, a single number that says who's better — is only the gradient part. Against a pure cycle it has nothing to grip. It reports 0% of what is there.
The triangle below is that calculation, live. The arrows are the lizards' three results; the bars beneath are the Hodge split, recomputed in your browser. The same instrument turned on the math behind AI leaderboards is a whole stratum of its own — No King of the Hill — where the cyclic fraction quietly eats the rankings nobody admits are cyclic. Here it eats the idea that one lizard is the best lizard.
A flow on the triangle, split into the part a ranking can hold (gradient) and the part it can't (curl). The two biological cycles are pure curl — a ranking captures 0%. Switch to a genuine ladder and the split flips: pure gradient, the ranking is exact. The decomposition is computed from the win/loss directions the papers record — no invented margins.
It is tempting to read the loop as a kind of dysfunction — a market that won't clear, an election that won't settle. But turn it around. Because no morph can win, no morph can take over. The moment one gets common, the strategy that beats it is handed a feast and grows. The cycle is not the problem; the cycle is the mechanism of coexistence. It is precisely what keeps three ways of being a lizard in the world at once, when a clean fitness ladder would have collapsed them to one.
You can watch this in the mathematics of evolution itself. Write down the replicator dynamics — the standard equation for how the fraction of each type changes when types that do better than average grow — for a symmetric rock-paper-scissors. The result is exact and beautiful: the only resting state is the dead centre, all three at one-third, and it is a neutral centre. Every starting mix circles it forever on a closed orbit. A quantity — the product of the three frequencies — is conserved exactly, the way energy is conserved on a frictionless pendulum, so the orbit never spirals in and never spirals out. Nothing ever wins. Nothing ever goes extinct.
Drag a starting point into the triangle and let it run.
Click anywhere inside the triangle to drop a population and trace its future. At zero tilt (true zero-sum rock-paper-scissors) the orbits are closed — the conserved product holds them — and the long-run average of any orbit is the dead centre, the ✦. Tilt the game so winning pays a little more or less than losing costs, and the centre becomes a spiral: tilt > 0 spirals inward to stable coexistence, tilt < 0 spirals outward until two morphs die. The sign decides the fate.
That last detail — the spiral — is the honest fine print. Pure zero-sum RPS sits on a knife edge: closed orbits, but only exactly when the wins and losses balance. The real world rarely balances perfectly. Sometimes the orbit spirals gently inward and the three settle into a stable mix; sometimes it spirals outward, swinging wider and wider until a morph hits zero and the cycle breaks, leaving the two-player game its survivors can't escape. Which way it goes is set by a single sign. The lizards have run their cycle for as long as anyone has watched — but the math is clear that a rock-paper-scissors world is not guaranteed to last. It can shake itself apart.
The lizards are vivid, but they're hard to run a controlled experiment on. So a second group built rock-paper-scissors out of something you can put in a flask: Escherichia coli.
Benjamin Kerr and colleagues, in Nature in 2002, engineered three strains. One produces colicin, a toxin that kills competitors — but making the poison (and the antidote that keeps it from killing itself) costs energy. A second is resistant: immune to the toxin, and spared the cost of producing it. A third is sensitive: it makes neither toxin nor resistance, so it pays nothing and divides fastest. Read the costs and the ring falls out on its own. The producer kills the sensitive. The resistant beats the producer, because immunity is cheaper than manufacture. And the sensitive beats the resistant, because growing fast beats paying for armour you don't need. C beats S beats R beats C — the same loop, drawn in metabolism.
Then Kerr's group did the thing the lizards wouldn't let them do: they changed the geometry of the fight. Grow the three strains in a well-mixed flask, where every cell effectively meets every other, and diversity collapses — one strain takes the whole tube. Grow them on a static agar plate, where a cell only ever meets its immediate neighbours, and all three persist indefinitely, carving the plate into shifting local territories. The cycle survives only when the interaction stays local. Mix it, and the loop that should protect everyone kills almost everyone instead. (A spatial model later pinned the threshold precisely: above a critical mobility, the patterns blow up and biodiversity is lost — Reichenbach, Mobilia & Frey, Nature 2007.)
The last instrument is that experiment, as a lattice. Each cell is one of the three strains; each step, a cell can be invaded by a neighbour that beats it. The mixing slider is the flask-versus-plate knob: at the left, fights are purely local and the grid organises into rolling spiral waves where all three coexist forever; slide right and you stir the pot, and watch the rolling waves give way to a single colour swallowing the plate.
Left of the slider — the agar plate: every fight is between touching neighbours, the grid rolls into spiral waves, and all three strains coexist. Push the slider right — the flask: each step, fights increasingly reach across the whole grid, the spirals dissolve, and one strain takes everything while the other two flatline. Same three rules, opposite fate, set by nothing but how far a fight can reach. A lattice model in the spirit of Kerr et al. (2002) and Reichenbach et al. (2007); the qualitative result — local sustains, mixing destroys — is theirs, reproduced here, not fitted to their numbers.
Three places now hold this same shape on the ground, and it is worth seeing them as one. In Always Bet Second, the cycle hides inside a fair coin — name any three-flip sequence and another beats it, round and round. In The Only Fair Vote, it surfaces in democracy itself — honest majorities can prefer A to B to C to A, and Arrow proved no fair rule can untangle the knot. In No King of the Hill, it breaks the leaderboards that rank machines. Coins, ballots, ratings — all human contraptions. This page is the one where it turns out nature was already playing it, on a lizard's throat and in a drop of gut bacteria, and where the loop everyone else treated as a paradox to be solved reveals its other face: the thing that keeps the world from collapsing to a single winner.
The two edges that bind this place: every factual claim is checked and every model is named a model. The dominance directions above are exactly as the two Nature papers record them; the six-year period is Sinervo & Lively's, attributed to them (later fieldwork has reported shorter cycles). The replicator orbits and the spatial lattice are models — the canonical symmetric game — chosen because they reproduce the published qualitative behaviour; they are not fitted to field data, and no win-probability magnitudes have been invented for the lizards or the microbes. The Hodge decomposition uses only the recorded win/loss directions. Everything numeric is recomputed in front of you and re-derived offline in /research/rock-paper-lizard/.