360 students in 60 groups of six, 300 rounds each, random pairing within the group, cash payoffs scaled by a parameter from 1.1 to 100. One group was excluded, leaving 354 players in the frequency analysis.

Aggregate frequencies sat near equilibrium (Rock 0.36, Paper 0.33, Scissors 0.32). Underneath, players repeated after wins and shifted after losses, and when they shifted they went in a consistent direction. The population as a whole cycled counterclockwise through the three throws at a small but persistent rate (about 0.02 to 0.03 per round, all significantly above zero). A six-parameter conditional-response model reproduced the cycling. An independent-decision model could not. Optimized conditional strategies beat the Nash player by roughly 10% at high stakes.

University students, repeated opponents inside a six-person group, lab incentives. The authors describe the result as bounded rationality, with win-stay lose-shift as a special case of their model, and are careful not to promote it to a universal rule.

31 undergraduates (26 women), 225 trials in three blocks against a computer playing exactly one third each, no money at stake.

The outcome-by-strategy interaction was significant (p < 0.001). After a loss, players switched rather than stayed. After a draw, they switched specifically to the throw that would have beaten their last one (an "upgrade"). After a win, nothing reached significance; staying was only numerically favored. Rock was thrown 35.6% of the time, which was not significantly above one third (p = 0.20). The shifts carried over. An upgrade tended to be followed by another upgrade.

Small, mostly female, unpaid, playing a machine. The authors call the effects small and attribute part of that to the lack of stakes. Dyson's reading is that the irrational choices are driven by an emotional reaction to a negative outcome, the Rock Paper Scissors version of poker tilt.

2,636,417 best-of-three matches in Roshambull, a 2007 Facebook app, 23 May to 14 August 2007, about 340,000 users. Before each match a "scouting sheet" showed the opponent's first-throw distribution, overall distribution, last five matches and record. The analysis centers on 7,758 experienced players with at least 100 clean matches.

82% of experienced players had first-throw distributions consistent with uniform, but 47% responded significantly to opponent history (more rock after an opponent's scissors-heavy history, less rock after paper-heavy). Experienced players won 34.7% and lost 22.2% of first throws, a net of about 1.5 cents per throw in the app's currency. A best-responder to the same predictable patterns would have netted about 14.3 cents. Level-k estimates put 74% of throws at level 0, 18.5% at level 1 and 7.7% at level 2.

An interface that hands you a scouting sheet, prestige points rather than money, and a self-selected 2007 Facebook population. The most durable finding is the gap between the edge available and the edge taken.

Paired a sighted player with a blindfolded one and asked whether the sighted player involuntarily copies the hand they can see forming, raising the draw rate.

Cook and colleagues reported draws at chance in blind-blind pairs and significantly elevated in blind-sighted pairs, and billed it as the first evidence of automatic imitation in a strategic context. Aczel and colleagues reviewed the analyses and ran the experiment with double the sample. Their data did not support the original conclusion.

WRPSA files this as disputed rather than settled, which is the correct filing. It also notes that in a refereed match, responding to a forming hand is a late throw, not a tell.

15 healthy volunteers, NIRS over prefrontal cortex, playing a computer task either to win or to lose on purpose.

The play-to-lose condition produced larger, laterally dominant prefrontal oxygenation changes than play-to-win. Inhibiting the obvious response costs more frontal effort than producing it.

Fifteen people and a modified task. It says nothing about tells, and WRPSA says so in the same paragraph.

Male Uta stansburiana come in three throat colors with three mating strategies. Orange males are aggressive and hold large territories with many females. Blue males hold small territories and guard their mates closely. Yellow males have no territory and mimic females to sneak matings. Orange beats blue by force. Blue beats yellow because a guarded female is hard to sneak. Yellow beats orange because an orange male cannot guard a harem that large.

Each morph can invade when rare and is invaded when common. Over six years of field observation the morph frequencies oscillated as the model predicted, with the fitness of each morph tracking year-to-year frequency changes.

An evolutionary analogy to the game's structure. The lizards are not throwing anything.

A colicin-producing strain kills a colicin-sensitive strain. The sensitive strain outgrows a colicin-resistant strain, because resistance has a metabolic cost. The resistant strain outgrows the producer, because producing the toxin costs more than resisting it.

On a static agar plate, where cells interact only with neighbors, all three strains coexisted. In a well-mixed flask, or on a plate where cells were redistributed at each transfer, diversity collapsed quickly. Local interaction and dispersal preserve a non-transitive community. Global mixing destroys it.

A result about ecology that reads as a result about organizational design. Communities with a dominance loop survive when interaction is local.

Seven chimpanzees at Kyoto University's Primate Research Institute learned a touchscreen task. Could they learn a circular relationship (A beats B, B beats C, C beats A) rather than the linear rankings animals usually learn? 38 children aged 35 to 71 months were tested with real hand gestures for comparison.

Five of seven chimpanzees learned the full loop after a mean of 307 sessions. The first pair (paper beats rock) took 1.7 sessions and the second (rock beats scissors) 3.1. The third pair, scissors beats paper, which closes the circle, took 14.3. The chimpanzees' mixed-pair performance matched that of four-year-old children. The children's accuracy rose with age and exceeded chance from about 50 months.

The chimpanzees never played against anyone. They chose the winning shape from two on a screen for food. The finding is about learning a loop, not about strategy.