Illustration of the study design.

a illustration of three successive example trials within a block of trials in the grouped (right, blue) and disjoint (left, orange) training curriculum. Numbers represent nodes on the map, coloured boxes represent transitions between adjacent nodes learned. Notably, the successive trials in grouped training are overlapping (pseudo-random walk), whereas the ones in disjoint training are not b Top. Illustration of the mixed study design, note that background colours for design graphic correspond to plot marker colours in subsequent figures (Figure 2 and Figure 3). Bottom. Examples of the two stimulus categories, each was randomly assigned to one of the two within-subjects maps for each participant c top illustration of the between-subjects manipulation, filled squares correspond to blocks of training trials bottom illustration of the second between-subjects factor, map shape.

Task sequence and training accuracy.

a multi-day task schedule, text in boxes symbolises the sequence of events (from left to right), test events depicted with a grey background, training with a white background. Day 1, 2 and 3 refer to the day of the experiment on which the task in question occurred. b Trial sequence for a select-action training trial (for a select-state trial example, see Supplementary Results 1) including illustrations of feedback for a correct and incorrect choice. Washing machine is the starting node, the two shapes (here: yellow square and green triangle) indicate the two positions the adjacent node (here: newspaper) could be in relative to the starting node. Shapes at the top indicate which key participants need to press to select either of the options. c training accuracy (y-axis) averaged across initial training and refresher (left); diamond/hexagon markers indicate average accuracy for the different conditions, black verticals lines are standard error of the mean bars and grey horizontal lines are accuracies for individual participants, red horizontal line is chance accuracy; *** p<.0001, d trial sequence of a two-step navigation trial, as participants make a choice, the central node (current location) changes accordingly and the number of steps remaining decreases; e grid map layout with examples of the spatial relations between two-step starts and goals, example straight 2 step pair is indicated in purple and diagonal pair is indicated in green f illustration of the layout of a proximity test trial g as well as the types of proximity comparisons ; central image examples shown in black and options for proximity judgements shown in red; h illustration of the task layout of the reconstruction task employed for the grid groups.

Participants are significantly better at multi-step planning and inference after disjoint compared to grouped training.

a average two-step navigation accuracy across both types of two-step trials (y-axis), markers represent average accuracies, vertical black lines are standard error of the mean (SEM) bars, horizontal grey lines are accuracies for individual participants, red horizontal lines represent accuracy at chance; N.B. the same applies to the graphs in b, d, f, g and h b participants’ accuracy on diagonal two-step trials; c illustration of the two types of two-step trials (top) as well as the direction of difference for both types (higher accuracy after disjoint than grouped training, bottom) d average two-step navigation accuracy for straight two-step trials e illustration of the significance of difference in accuracy on the different proximity judgement types, *** p<.0001, ** p<.01, * p<.05, ~ n.s; black frame on grid indicates example central node, red frame on grid indicates two example comparisons for proximity. These icons are repeated in f and g f accuracy on proximity judgement task in grid groups; grid icons above the average markers indicate the type of proximity trial g accuracy on proximity judgement task in torus groups; grid icons above the average markers indicate the type of proximity trial h map reconstruction accuracy at the end of testing in the grid groups, y-axis shows Spearman correlation between actual grid maps and reconstructed maps. Red line indicates accuracy when performing at chance.

Example of a select-stimulus trial

sequence in which the incorrect option was selected. Had the correct option been selected, the washing machine would have appeared with a green frame around it and the kettle would have disappeared at the time of feedback.

Full table of results for mixed effects model results reported in the main text + training accuracy split into initial training and refresher

All models in this table were specified as follows, with Y being the outcome variable of interest, cohort our between-subjects manipulation (blocked (1) vs semi-blocked (0)), condition the within-subjects manipulation (grouped 0 vs disjoint 1) and shape being torus (1) or grid (0). Y~ cohort * condition + shape + (1|participant)

Main text results split by torus and grid cohorts

Y~ cohort * condition + (1|participant)

Additional analyses reported in the main text

The analysis conducted to determine whether there was an interaction between type of two-step navigation trial and our spatial manipulation was specified as accuracy ~ cohort * condition*trial type + shape + (1|participant) The analysis querying whether participants’ overall test accuracy increased from the first to the second test block was specified as accuracy ~ cohort * condition*test block + shape + (1|participant) The results from these analyses are reported in the table below.

Pairwise comparisons (grouped vs disjoint) for the four experimental groups for the main text results

All pairwise comparisons

All stimuli used