Repeated games
144 games from Akata et al. (2025). Dictator and Ultimatum are well-known additions outside the paper.
Mean points per round
| Model | Points | % A | Opponent | Joint |
|---|
Details and records
The Play button opens a browser game against a scripted bot. Your choices stay in this browser and do not enter the benchmark results.
The 144 games are the authors’ numbered payoff tables, grouped here by their source names. Game descriptions interpret the source families and payoffs as behavioral challenges; variant descriptions summarize the displayed payoffs.
The paper’s main analysis covers win–win, Prisoner’s Dilemma, biased, unfair, cyclic, and second-best games; the source tables also include eight Tragic variants.
The Evolution of Trust is another Prisoner’s Dilemma example, with different payoffs and opponents.
An unambiguous allowed choice on the first or last nonempty line is accepted, including Markdown emphasis or inline code. Exact one-letter compliance is recorded separately. A score requires all scheduled episodes to finish. Combined scores give each variant equal weight and require complete results for every variant.
% A is the percentage of the model’s choices that use A in the payoff table, over the same rounds as its points. In Prisoner’s Dilemma, A is cooperation and B is defection; in other games, A is just a choice label. The models’ randomized letters are mapped back to the table’s A/B labels.
DataCodeSettingsSource tablesLicenseRun planOriginal pilotDictator and Ultimatum run plan
| Round | Model move | Opponent move | Model points | Opponent points |
|---|