Review Papers Claim by Claim

Concise appraisal

The authors present a preregistered fMRI study using a novel card planning task that dissociates instantaneous point value from instrumental set value and report that striatum positively tracks instrumental value while dmPFC and bilateral insula negatively track instantaneous value; vmPFC was not robustly active at choice time Paper

Detailed paper review and critique

Summary of design and main claims

• Task: fully instructed 20-card deck card collection task; 10 decisions per game; instantaneous reward on each pick plus end-of-game set bonuses (20 points per complete suit or digit set; up to 60 per game) designed to orthogonalize instantaneous and instrumental value Task
• Behavioral model: three participant parameters (beta for instantaneous point weight, sigma suit set weight, delta digit set weight), softmax choice; model explained ~75% of choices within-sample and outperformed five competing models Computational
• Neuroimaging results: instrumental (set) value positively correlated with striatal BOLD (putamen/caudate); instantaneous point value negatively correlated with dmPFC and bilateral anterior insula; vmPFC did not show robust positive encoding at choice time fMRI

Strengths

1. Careful task design that makes instrumental value explicit and orthogonalizes it from instantaneous reward, reducing confound with learning or prediction errors Task
2. Combination of computational modeling and model-based fMRI to link individual strategy weights to neural signals and performance metrics Model
3. Preregistration, public code and behavioral data on OSF improving transparency and enabling replication attempts Open

Limitations, potential confounds, and blindspots

• Sample size: final fMRI sample N=25 after 4 exclusions is modest for whole-brain fMRI contrasts and correlational individual differences; small N increases vulnerability to outliers and reduces power for some effects (authors note some correlations were weak and Bonferroni corrections removed several associations) Sample
• Absence of vmPFC encoding: authors interpret null vmPFC effect as that region perhaps coming online at outcome or consolidating value later; however, nulls can reflect timing, analysis choices (e.g., orthogonalization, collinearity of regressors), ROI sensitivity or spatial smoothing choices. Alternative analyses (e.g., trial-by-trial cross-validated ROI regressions, deconvolution, or model-based representational similarity analyses) could test whether vmPFC encodes combined value across different temporal windows vmPFC
• Task abstraction and ecological validity: the card-set task is artificial and non-spatial; hippocampal contributions to planning may be minimized for abstract non-relational tasks, limiting generalization to real-world planning that relies on episodic memory, mapping, or hierarchical goals Hippocampus
• Potential model misspecification: the set-value heuristics (increment by 1 or 2 when potential sets remain) are reasonable but coarse; alternative formulations (probabilistic expected set gain given remaining deck composition, explicit forward simulation, successor representation) might capture planning depth and could change model-based regressors used for fMRI. Testing richer forward-planning models could strengthen claims about model-based signals in striatum Model
• Multiple comparisons and ROI selection: while FWE-corrected clusters are reported, some key individual differences and ROI correlations were weak and sensitive to outliers; pre-registered vs exploratory contrasts should be transparent in supplementary materials to avoid post hoc rationalization. The authors note preregistration but some ROI correlations appear exploratory Multiple

Interpretation and relationship to prior literature

The principal novel claim is that striatum encodes instrumental future-oriented value during planning rather than only immediate reward; this aligns with prior model-based RL fMRI work showing striatal involvement in model-based predictions, while challenging simple dual-system accounts that equate striatum strictly with immediate reward processing Striatum

Concurrently, dmPFC and anterior insula showing negative correlations with instantaneous value is interpretable as signaling conflict or demand for control when immediate rewards are low and instrumentally valuable choices must be selected. That accords with dACC/dmPFC roles in expected value of control literature dmPFC

Concrete suggestions to strengthen and falsify claims

1. Model enrichment: implement and compare alternative computational models that compute instrumental value as explicit expected bonus given remaining deck state (probabilistic forward simulation) or successor representation approximations; then run model comparison and use the winning model for model-based fMRI regressors to test whether striatal encoding remains specific to instrumental predictions.
2. Temporal dynamics: perform finite impulse response or deconvolution analyses to search for vmPFC encoding at later time points (outcome/consolidation) or use time-resolved MVPA to test whether vmPFC contains mixed signals that cancel in univariate contrasts.
3. Power and robustness: replicate with a larger fMRI cohort (N 45 60) and pre-registered ROI and individual differences tests to clarify correlations and reduce influence of outliers; report robust regression and leave-one-out analyses for reported brain-behavior correlations.
4. Cross-task generality: use a more ecologically rich planning task with episodic or spatial structure to test hippocampal involvement and generality of striatal instrumental signals across task domains.

Practical reproducibility notes

Behavioral data and code are available on OSF enabling reanalysis of the computational model and behavioral fits; fMRI raw data require request to corresponding author but preprocessing and SPM12 pipeline details are provided aiding reproducibility. OSF links are provided in the paper Data

Conclusions and confidence

Overall the paper provides a careful behavioral and model-based fMRI study that meaningfully dissociates instantaneous from instrumental value during planning in a controlled task. The core neural dissociation (striatal positive for instrumental value; dmPFC/insula negative for instantaneous value) is supported by the reported analyses and by preregistration and open behavioral data, but (a) modest fMRI sample size and (b) model specification choices limit how broadly the results generalize; replication with richer forward-planning models and larger samples would materially strengthen confidence.

Confidence in the central neural dissociation given the current data: moderate (evidence consistent but not definitive). Key disconfirming evidence would be replication failures with larger samples, or demonstration that richer forward-planning models shift the explanatory neural correlates away from striatum to other regions.

Useful resources and next steps

• Behavioral data and code: https://osf.io/95ucg/ (authors state available) OSF
• Recommended reanalyses: implement probabilistic expected-set-value model and test model-based regressors; run robust regression and permutation tests for brain behavior correlations; perform ROI MVPA for vmPFC to detect mixed coding.

Author reviews