This repository contains five folders in total. The analysis code to go with this data can be found here: https://github.com/summerfieldlab/Garrett_Glitz_etal. Two of the folders (onset and outcome) contain preprocessed fMRI data needed to run the multivariate (non-crossvalidated) representational similarity analysis functions that have been shared in https://github.com/summerfieldlab/Garrett_Glitz_etal. In addition to that, the encoding model in https://github.com/summerfieldlab/Garrett_Glitz_etal requires the two files shared in the folder encoding. Lastly, the folder behavioural_data replicates the contents of https://github.com/summerfieldlab/Garrett_Glitz_etal/tree/main/data/behavioural_data but have been reuploaded here for completeness. Once you have downloaded the data, please change the file paths provided in the analysis scripts to match those on your computer. The folder *door_onset* contains data from the time of door onset. This data takes the form of 7 structs, each of which contains data from one brain region (state prediction error functional ROI; bilateral hippocampus, bilateral entorhinal cortex, bilateral amygdala, bilateral parahippocampus; bilateral early visual cortex V1, bilateral motor cortex M1). The data is organised in structs and contains a trial x voxel matrix for each subject, session, context (gem) and transition probability (p heist state | door presented) quartile. The folder *outcome_onset* contains the same kind of data file but now from the time of outcome presentation. The organisation of the files and the brain regions covered are identical apart from the fact that the structs now contain a trial x voxel matrix for each subject, session, context, outcome state and door colour the outcome was presented on. The folder *encoding_model* contains two files, BOLD.mat and pValues.mat, the former contains a struct that is organised as follows subject x session x context x data, where data is a trial x voxel activity matrix. The second file pValues.mat contains the model-derived transition probabilities (p heist state|door presented) and is organised in the same fashion. The folder *behavioural_data* contains two files called gem _ dat which are .csv files containing all the relevant recorded and computed variables from the behavioural pilot and fMRI experiment that are needed for the computational model and the behavioural analyses (see scripts on https://github.com/summerfieldlab/Garrett_Glitz_etal). The two files called valence _ means are required to replicate the analysis in Figure 6. The files tstat_simulation.csv and estimate_permutation.csv are t values and beta estimates generated by running the null distribution simulations for the behavioural experiment (code on github). The folder *SPM_maps* contains the group level maps for the univariate state prediction error analysis (both main effect of state prediction error and SPE * valence interaction), as well as four searchlight group level maps, which show where in the brain the diagonal is more similar than the off-diagonal for both dependent and independent condition at time of door onset (searchlights 1 and 2), where this effect is greater in the dependent than the independent condition (searchlight 3), and the same interaction at the time of outcome presentation (searchlight 4).