This dataset contains behavioral and rater data collected across two studies. The first component comprises behavioral data from a repeated Split-or-Take-All Prisoner's Dilemma experiment conducted at the Economic Science Institute (ESI), Chapman University. Anonymously paired participants chose to either cooperate ("Split") or defect ("Take All") across two rounds, with payoffs determined by the combination of both players' choices: mutual Split yielded 5 USD to each player, mutual Take All yielded 0 USD to each, and mismatched choices awarded 10 USD to the defector and 0 USDto the cooperator. The dataset includes participant choices, partner choices, and round-by-round outcomes. The second component comprises data from an online rater prediction study conducted via Prolific, in which observers predicted participants' Round 1 and Round 2 choices across conditions varying access to visual and behavioral information. This component includes rater predictions, prediction accuracy measures, and demographic variables. These data support research on cooperative behavior, reciprocity, social prediction, and strategic decision-making in repeated social dilemmas. Researchers interested in prisoner's dilemma dynamics, conditional cooperation, impression formation, or behavioral prediction may find this dataset useful. Associated publications are listed under Related Identifiers. Thin-slice video clips and photographs of participants recorded immediately prior to their decisions are archived separately (see Related Identifiers). A restricted-access record linking visual stimuli to participant behavioral data is available upon approved request. For questions about the data, contact Eric Schniter (schniter@chapman.edu), Economic Science Institute, Chapman University.