(A) Trial Structures in the Experimental Task.

(i) Initial Encoding: Participants memorized object-location pairs by creating a story or forming a “mental photo” of each scene, focusing on the exact location of the object within the scene. (ii) Learning Phase: Participants selected one of three possible object locations and received feedback: a happy face for correct responses, a sad face for incorrect ones, and a sleeping face for missed response. The correct object-location pairing was then displayed again. (iii) Retrieval Phase: Conducted inside the MR scanner, participants chose the object’s location in the scene from three options without receiving feedback. (B) Experimental Procedure. Testing took place across three days. On Day 0, participants learned 60 object-location associations (remote items). On Day 1 (short delay), they learned 30 new object-location associations (recent items) and retrieved 30 remote and 30 recent items. On Day 14 (long delay), participants learned another 30 new associations and retrieved 30 remote and 30 recent items. Throughout all sessions, participants also completed socio- demographic and psychometric questionnaires, which were distributed across sessions. Note: RT – reaction time; s – second, fMRI – functional magnetic resonance imaging.

Sample characteristics by age group

(A) Overview of Learning Performance. Individual learning trajectories across up to four encoding-retrieval cycles for children and young adults on Day 0, Day 1 and Day 14. Each coloured dot represents a participant’s accuracy (percentage of correct responses) at a given cycle. Transparent connecting lines illustrate within-person changes in accuracy across cycles. Across all sessions, children needed on average between two to four learning- retrieval cycles to reach the criterion of 83% correct responses, while young adults typically reached it within two cycles. (B) Final Learning Performance. Final learning accuracy is calculated as the percentage of correct responses during the last learning cycle for both children and young adults. For each group and session, distributions are visualized using half-eye plots (smoothed density estimates), overlaid with boxplots indicating the median and interquartile range. The shape and spread of density plot reflect individual data variability. Grey dashed line indicates the criteria of 83% correctly learned items.

Retention rates for initially correctly learned items.

Memory accuracy is operationalized as the percentage of correct responses in the retrieval task conducted during the MRI scanning sessions for items that were initially correctly learned, indicating strong initial memories. Memory accuracy for recently consolidated items did not differ between sessions in young adults and children and was collapsed across sessions. Overall, young adults show higher and more stable memory accuracy than children, with memory declining over time for both groups, particularly for long delay. All tests used Sidak correction for multiple comparisons. *p < .05; **p < .01; ***p < .001(significant difference); non-significant differences were not specifically highlighted. The boxplot summarizes the distribution of accuracy scores across sessions and delay conditions. In each boxplot, the central line indicates the median, the box represents the interquartile range (25th - 75th percentile), and the whiskers extend to the range of values within 1.5 times the variability. The red dashed line at 34% indicates the threshold for chance-level performance.

Mean Signal Differences Between Correct Remote and Recent Memories.

The figure presents mean signal difference for remote > recent contrast across sessions and groups during the object presentation time window in (A) Anterior and Posterior Hippocampus; (B) Anterior and Posterior Parahippocampal Gyrus; (C) Cerebellum; (D) Medial Prefrontal Cortex; (E) Ventrolateral Prefrontal Cortex; (F) Precuneus; (G) Retrosplenial Cortex; (H) Lateral Occipital Cortex. Note: Bars indicate the group mean for each session (solid lines for Day 1, dashed lines for Day 14), plotted separately for children and young adults. Error bars represent ±1 standard error of the mean. The colour indicated the age groups: purple for children and khaki yellow for young adults. Across all panels, mean of individual subject data are shown with transparent points. The connecting faint lines reflect within-subject differences across sessions. Orange asterisks denote significant difference of remote > recent contrast from zero. An upward orange arrow indicates that this difference is greater than zero, while a downward arrow indicates that this is less than zero. *p < .05; **p < .01; ***p < .001(significant difference); non- significant differences were not specifically highlighted. Significant main and interaction effects are highlighted by the corresponding asterisks. All main and interaction p-values were FDR-adjusted for multiple comparisons.

Multivariate short- and long-delay brain profiles of neural upregulation (remote versus recent neural activation differences) are associated with variations in memory accuracy.

A) Short Delay Brain Profile. Latent variables weights or saliences for each ROI build up one latent variable that expresses a composite short-delay brain profile across both age groups . B) Long Delay Brain Profile. Latent variables weights or saliences for each ROI build up one latent variable that expresses a composite long-delay brain profile across both age groups. The bar plot shows the bootstrap ration (BSR) values for the latent variable, reflecting the stability of the relationship between brain activation and memory performance. Stability of salience elements is defined by Z-scores (depicted as red line: a value larger/smaller than ± _1.96 is treated as reliably robust at (a < .05). C) Short Delay Brain Scores by Group. D) Long Delay Brain Scores by Group. Each box represents the distribution of brain scores within a group, with central lines indicating the median and boxes showing the interquartile range. Whiskers represent the full range of non-outlier values. Note: PHGa – anterior parahippocampal gyrus; PHGp – posterior parahippocampal gyrus; HCa – anterior hippocampus ; HCp – posterior hippocampus; PC– precuneus; vlPFC – ventrolateral prefrontal cortex; mPFC – medial prefrontal cortex; RSC – retrosplenial cortex; LOC – lateral occipital cortex; CE – cerebellum; r – Spearman’s rank order correlation index.

Scene-Specific Reinstatement (A) Index Computation (Scene).

A representational similarity index was computed by calculating the averag similarity between activation patterns in the fixation and scene time windows, separately for recent scenes, remote scenes on Day 1, and remote scenes on Day 14. (B) Scene-Specific Reinstatement. A corrected scene-specific reinstatement index was computed by assessing the average similarity within-trial similarity between the fixatio and scene time windows and subtracting the average between-trial (set-based) similarity across all other trials. This controls for baseline similarity unrelated to specific scene content. S – scene time window; F – fixation time window; r – Pearson’s correlation index; Δ z – difference between two Fisher transformed r values. * - Activatio patterns.

Corrected scene-specific neural reinstatement.

Scene-specific neural reinstatement defined as the difference between Fisher-transformed scene-specific and set- specific representational similarity. Scene-specific neural reinstatement index by group (children vs. adults) and session (Day 0 – recent, Day 1 – remote short delay, Day 14 – remote long delay). Bars represent the mean reinstatement index for each session within each group, with error bars indicating standard error of the mean. Transparent dors show individual participant data points, jittered horizontally for visibility. The x-axis is grouped by group and displayes (A) Hippocampus Anterior; (B) Hippocampus Posterior; (C) Parahippocampal Gyrus Anterior; (D) Parahippocampal Gyrus Posterior; (E) Cerebellum; (F) Medial Prefrontal Cortex; (G) Ventrolateral Prefrontal Cortex; (H) Precuneus; (I) Retrosplenial Cortex; (J) Lateral Occipital Cortex. *p < .05; **p < .01; ***p < .001(significant difference). Error bars indicate standard error.

Statistical overview of LME-model based Sidak corrected post hoc comparisons for scene-specific reinstatement analysis (based on LME-model described in Table S10).

Representational Similarity Analysis.

(A) Index Computation (Gist). A representational similarity index was computed by assessing the average similarity for fixation time window for within-category and between-category scenes separately for recent, remote (Day 1), and remote (Day 14) scenes based on both within-run and cross-run comparisons. The diagonal (similarity of fixation time window with itself) was excluded from the analysis. (B) Gist-like Representation. A gist-like representation index was computed by assessing the average similarity in fixation time window for the same- category pairs and subtracting from it the any-other-category pairs. S – scene time window; F – fixation time window; r – Pearson’s correlation index. Δ z – difference between two Fisher transformed r values.

Gist-like Representations Bar plots show mean gist-like representation index (difference in Fisher’s z-transformed (Δ z) similarity: within- category –– between-category) in each group (Children, Young Adults) and session (Day 0, Day 1, Day 14), computed from combined within- and cross-run comparisons.

Error bars indicate ±1 standard error of the mean. A representation value above zero (denoted by red asterisks) reflect greater neural pattern similarity during fixation time window between item from the same category than across categories. Bar positions are grouped by age group (x-axis). Session-specific estimates (Day 0, 1, 14) are differentiated by line of bar border. (A) Medial Prefrontal Cortex; (B) Ventrolateral Prefrontal Cortex; (C) Lateral Occipital Cortex (D) Precuneus; *p < .05; **p < .01; ***p < .001 (significant difference; FDR corrected for multiple comparisons); non-significant difference was not specifically highlighted.

Multivariate short- and long-delay brain profiles of scene-specific reinstatement are associated with variations in memory accuracy.

A) Recent Delay Brain Profile. Latent variables weights or saliences for each ROI build up one latent variable that expresses a composite immediate-delay scene-specific reinstatement brain profile. B) Short Delay Brain Profile. Latent variables weights or saliences for each ROI build up one latent variable that expresses a composite short-delay scene-specific reinstatement brain profile. C) Long Delay Brain Profile. Latent variables weights or saliences for each ROI build up one latent variable that expresses a composite long-delay scene- specific reinstatement brain profile. Stability of salience elements is defined by Z-scores (depicted as red line: a value larger/smaller than ± _1.96 is treated as reliably robust at (a < .05). The bar plot shows the bootstrap ration (BSR) values for the latent variable, reflecting the stability of the relationship between brain scene-specific neural reinstatement and memory performance. D) Recent Delay Brain Scores. E) Short Delay Brain Scores. F) Long Delay Brain Scores. Each box represents the distribution of brain scores within a group, with central lines indicating the median and boxes showing the interquartile range. Whiskers represent the full range of non-outlier values. Note: PHGa – anterior parahippocampal gyrus; PHGp – posterior parahippocampal gyrus; HCa – anterior hippocampus ; HCp – posterior hippocampus; PC– precuneus; vlPFC – ventrolateral prefrontal cortex; mPFC – medial prefrontal cortex; RSC – retrosplenial cortex; LOC – lateral occipital cortex; CE – cerebellum; r – Spearman’s rank order correlation index.

Multivariate short- and long-delay brain profiles of gist-like representations are associated with variations in memory accuracy.

A) Short Delay Brain Profile. Latent variables weights or saliences for each ROI build up one latent variable that expresses a composite short-delay gist-like representations brain profile across age groups. B) Long Delay Brain Profile. Latent variables weights or saliences for each ROI build up one latent variable that expresses a composite long-delay gist-like representations brain profile in child group. The bar plot shows the bootstrap ration (BSR) values for the latent variable, reflecting the stability of the relationship between brain gist- like neural representations and memory performance. Stability of salience elements is defined by Z-scores (depicted as red line: a value larger/smaller than ± 1.96 is treated as reliably robust at (a < .05). Note: vlPFC – ventrolateral prefrontal cortex; mPFC – medial prefrontal cortex; LOC – lateral occipital cortex; r – Spearman’s rank order correlation index.