Rock Paper Scissors Research Library
What published research has learned about Rock Paper Scissors, from huge online datasets to lizards and bacteria, plus live totals from games played on wrpsa.com. Every study is linked.
Published and last substantively reviewed August 11, 2026 · Owner: WRPSA
Rock Paper Scissors looks simple until a scientist gets hold of it. Here is what published research has actually found about how people play, and a few places where nature plays the same game. Every summary links to the original paper and comes with its limits, because a lab result is not a guaranteed win against your cousin.
What the research says
- Against an opponent who picks truly at random, no throw beats any other. Every strategy starts from that baseline; the game theory guide shows why.
- Real people are not random. In repeated lab play and in a huge online dataset, players reacted to their last result and to their opponent's history, and experienced players took advantage of it.
- The popular claim that players copy a hand they can see is disputed: a larger follow-up study did not back it up.
- Lizards and bacteria play their own version: three strategies that each beat one and lose to another can keep all three alive. That is biology, not a tip for your next match.
- One small brain-imaging study measured more frontal-brain activity when people tried to lose on purpose than when they tried to win. Interesting, but far too small to say anything about tells.
Social cycling and conditional responses
- What it tested
- Whether repeated RPS play under random pair matching follows fully independent mixed-strategy behavior or shows population cycling and conditional responses.
- Sample or system
- 360 Zhejiang University students in 60 six-person groups, scheduled for 300 rounds each. One group was excluded from the reported action-frequency analysis, leaving 354 players there.
- Finding
- The experiment found population-level cycling and action-to-action dependence consistent with conditional response behavior, even though aggregate gesture frequencies were close to the one-third equilibrium.
- Limitations
- A laboratory student sample, payoff design, repeated opponents, and group protocol do not establish one universal rule for every casual or tournament opponent.
What one million online games showed
- What it tested
- How players use visible opponent history in a large online strategic setting and whether experience changes exploitation of that information.
- Sample or system
- The published analysis reports 334,661 players with at least one completed match in an online RPS application, and 2,636,417 matches.
- Finding
- Players departed from independent randomization, used opponent history, and more experienced players exploited observable behavior more effectively.
- Limitations
- Application rules, interface cues, self-selection, and online behavior may not generalize directly to simultaneous in-person tournament play.
Automatic imitation and its replication
- What it tested
- Whether seeing an opponent’s developing gesture increases draws through automatic imitation.
- Sample or system
- A controlled sighted-versus-blindfolded RPS experiment; consult the paper for condition-level counts and analysis.
- Finding
- The original paper reported more draws when players could see an opponent’s gesture.
- Limitations
- A larger published replication and critique did not support the original conclusion. WRPSA treats the automatic-imitation claim as disputed rather than settled.
- Follow-up study
- Aczel, Bago & Foldes (2012), replication and critique
The side-blotched lizard evolution cycle
- What it tested
- Whether three side-blotched lizard mating strategies form a non-transitive cycle whose frequencies oscillate over time.
- Sample or system
- Field observations and parameter estimates covering several years of morph-frequency changes.
- Finding
- The model predicted oscillations, and the three male morph frequencies were observed to oscillate in the field.
- Limitations
- This is an evolutionary analogy to RPS’s non-transitive structure, not evidence about human gesture choice or tournament strategy.
Non-transitive E. coli biodiversity
- What it tested
- Whether spatially localized interaction and dispersal preserve coexistence in a non-transitive three-population Escherichia coli community.
- Sample or system
- Controlled microbial communities compared across localized and larger-scale interaction conditions.
- Finding
- All three populations coexisted under localized interaction, while diversity was rapidly lost when interaction and dispersal occurred over larger spatial scales.
- Limitations
- The RPS framing describes ecological dominance relations. It is not a study of people playing the hand game.
Prefrontal activity in modified RPS tasks
- What it tested
- Prefrontal oxygenation during modified “play to win” and inhibitory “play to lose” RPS tasks.
- Sample or system
- 15 healthy volunteers measured with near-infrared spectroscopy.
- Finding
- The lose task produced larger, laterally dominant prefrontal oxygenation changes than the win task.
- Limitations
- A small sample and modified computer-response task cannot establish broad claims about all brain regions, ordinary play, or readable physical tells.
For a careful explanation of what laboratory and behavioral findings can and cannot say about individual play, continue to Rock Paper Scissors psychology.
Live from wrpsa.com
What gets played here
These totals count games played on wrpsa.com. A match counts as played when two players finished at least one round with both throws in. Abandoned games, cancellations, disputes, and byes stay out of that total.
- Matches played
- 3,242
- Rounds played
- 10,571
- Player vs player
- 76
- Player vs the WRPSA computer
- 3,166
Counted from Jul 29, 2026 through Oct 16, 2026. Last updated UTC.
For researchers: how the count works
Every number on this page can be checked. These are working totals that include games not yet reviewed for research, and they describe play on wrpsa.com only, not the wider world.
| Measure | Count | Meaning |
|---|---|---|
| Matches recorded | 5,194 | Every match started on wrpsa.com, finished or not. |
| Player appearances | 3,318 | Each time a person, not the computer, took part in a played match. |
| Finished without a round-by-round record | 10 | Counted separately and left out of played matches. |
| Replaced by a correction | 0 | Kept on file and left out of played-match and research totals. |
| Not yet reviewed for research | 5,194 | Left out of research totals until reviewed. |
| Approved for a public dataset | 0 | No player-level dataset has been released under these counting rules. |
| Way to play worked out from older records | 2,492 | Matches saved before we recorded how each one was played. The way to play is worked out from the match details. |
Played matches by type
Working totals by way of playing, game mode, and match type. They include games that have not been reviewed for research, and they describe play on wrpsa.com only.
| Type | Matches | Rounds |
|---|---|---|
| Classic | 1,218 | 2,148 |
| Local (two phones) | 4 | 8 |
| Online | 2,003 | 6,595 |
| Video | 17 | 52 |
| Type | Matches | Rounds |
|---|---|---|
| Practice | 4 | 8 |
| Quick Match | 4 | 17 |
| Against the computer | 3,166 | 8,567 |
| Against a friend | 68 | 211 |
| Type | Matches | Rounds |
|---|---|---|
| Casual | 72 | 228 |
| Unrated | 3,170 | 8,575 |
Counting rules rps-match-research-v2. Generated UTC. Counted from Jul 29, 2026 through Oct 16, 2026. These figures describe play on wrpsa.com only and are not compared with any outside dataset.
Citable releases
Reviewed dataset releases
Each release freezes the counting rules, totals, dates, source coverage, citation, and the exact JSON and CSV files. A second administrator checks the checksums before it goes out. Corrections arrive as new releases linked to the old one; published files are never quietly swapped.
No reviewed release has been published yet. The live totals above are working numbers, not a frozen release.
How to cite and correct this page
Cite the paper itself, not this summary, when making a research claim. If a link, sample, summary, or limitation here is wrong or incomplete, send the page address and your source through the contact page. The full standard is in the research and citation policy.