This paper tests two established reciprocity models—the intention factor model and the reference value model—using twelve mini-ultimatum games. The results show that both models have limitations in predicting actual behavior. The reference value model, which measures kindness based on the distance of an offer from a reference value, fails to accurately capture the extent of kindness, as increasing this distance does not necessarily lead to higher rejection rates. Additionally, varying the game while keeping the reference value and distance constant leads to different rejection rates, indicating that the model is not robust against outliers. The intention factor model, which assesses kindness through a pairwise comparison with extreme outcomes, performs better in simple games with two alternatives. However, in richer games with more alternatives, the model shows little consistency between predictions and decisions. The pairwise comparison of alternatives does not adequately describe behavior, suggesting that a combination of both models may provide a better description of reciprocal behavior. The study highlights the importance of intentions in reciprocal behavior and finds that the intention factor model's predictions are more consistent in simple games. However, in richer games, the model's predictions are less reliable. The results suggest that neither model alone can fully explain reciprocal behavior, and a combination of both models may be necessary to capture the complexity of human reciprocity. The paper concludes that both models have strengths and weaknesses, and further research is needed to develop more accurate models of reciprocity.This paper tests two established reciprocity models—the intention factor model and the reference value model—using twelve mini-ultimatum games. The results show that both models have limitations in predicting actual behavior. The reference value model, which measures kindness based on the distance of an offer from a reference value, fails to accurately capture the extent of kindness, as increasing this distance does not necessarily lead to higher rejection rates. Additionally, varying the game while keeping the reference value and distance constant leads to different rejection rates, indicating that the model is not robust against outliers. The intention factor model, which assesses kindness through a pairwise comparison with extreme outcomes, performs better in simple games with two alternatives. However, in richer games with more alternatives, the model shows little consistency between predictions and decisions. The pairwise comparison of alternatives does not adequately describe behavior, suggesting that a combination of both models may provide a better description of reciprocal behavior. The study highlights the importance of intentions in reciprocal behavior and finds that the intention factor model's predictions are more consistent in simple games. However, in richer games, the model's predictions are less reliable. The results suggest that neither model alone can fully explain reciprocal behavior, and a combination of both models may be necessary to capture the complexity of human reciprocity. The paper concludes that both models have strengths and weaknesses, and further research is needed to develop more accurate models of reciprocity.