{"id":"50b397bc-5cc3-4d07-bcc0-c3c50f4ccfc6","arxiv_id":"2411.16558","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Applying structural balance theory to signed fMRI networks, this study reports more balanced triads and lower balance energy during N-back working memory than at rest.","lead":"Researchers compared brain scans during a working memory task with resting scans and found more balanced triangles of positive and negative connections during the task. The result is presented as evidence that working memory pushes the brain into a more stable network state, though the analysis does not rule out task-generic arousal or correlation changes.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Working-memory specificity is untested: task vs rest contrasts alone cannot rule out generic task engagement or global-signal confounds; 0-back vs 2-back comparison is needed.","rationale":"I read the paper in good faith. The trio of results (positive links up, balanced triads up, energy down) is internally consistent, but the central claim's specificity to working memory rests entirely on the sign-flip interpretation. The paper pools 0-back and 2-back, so the standard HCP control condition is unused. Because any task that raises global correlation would produce the same triad pattern, and because balance energy is definitionally tied to the triad fraction, the only decisive evidence would be a load contrast. This matches the reader's weakest assumption; I agree with the CONDITIONAL verdict. The concrete test above would settle it.","tokens_in":10121,"tokens_out":6042,"duration_ms":55914,"concrete_test":"Re-analyze the same HCP WM task data separately for 0-back and 2-back epochs: build per-condition signed connectivity matrices using the same Glasser 360 parcellation and Pearson correlation on concatenated blocks, then compute balanced/imbalanced triad counts and balance energy for each condition. Run paired Wilcoxon signed-rank tests comparing 2-back vs 0-back and each vs rest. If balanced triads do not significantly increase from 0-back to 2-back (or if 0-back also exceeds rest), the working-memory-specific claim fails; if 2-back > 0-back, the claim is supported. Also report correlations of link-sign changes with head motion (FD) and global signal to rule out artifacts.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that N-back working memory specifically increases balanced triads and lowers balance energy relative to rest—depends on the assumption that the rest-to-task sign changes in Pearson-correlation networks reflect memory-specific neural coordination rather than generic task engagement. The HCP WM task (Task section) includes both 0-back and 2-back blocks, yet the Functional Connectivity section concatenates all task time series into one array (`task_data`) and computes a single connectivity matrix per subject, pooling conditions. All reported contrasts are rest vs. this pooled task state. A global increase in positive correlations—from arousal, attention, motion, or hemodynamic shifts—would mechanically turn negative edges positive, increase T3 triads, decrease T1/T2, and lower energy; indeed, balance energy is a deterministic function of the balanced-triad fraction (E = 1 - 2B/T), so the energy result adds no independent evidence. Without a 0-back vs 2-back contrast or a non-memory control task, the observed triad shift cannot be attributed to working memory load.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper applies structural balance theory (SBT) to signed functional connectivity networks derived from HCP fMRI data, comparing resting state to the N-back working memory task in 138 healthy right-handed male participants. Connectivity is computed with Pearson correlation across Glasser 360 parcels; task time series are concatenated across runs and all task conditions into one array. The authors report a significant increase in positive links and balanced triads (T3), a decrease in negative links and imbalanced triads (T1, T2), a decrease in negative tendency-to-make-hub, and a decrease in balance energy during the task, and they identify temporal, parietal, and prefrontal regions as the main drivers of negative-to-positive link changes. The paper interprets these results as evidence that the brain shifts toward a more stable functional state during working memory.","tokens_in":10338,"tokens_out":4745,"duration_ms":45577,"significance":"If the reported effects survive appropriate controls, this would be a novel application of structural balance theory to working memory and a useful addition to the network-level literature on cognitive states. The paper has clear strengths: it uses a publicly available dataset, applies paired nonparametric tests, states the SBT formulas explicitly, and grounds its regional discussion in prior working-memory findings. However, the central interpretation as working-memory-specific stability is currently under-supported: the task/rest contrast pools 0-back and 2-back, no null model or global-correlation control is reported, and one of the headline findings (balance-energy decrease) is mathematically equivalent to another (balanced-triad increase) rather than an independent confirmation.","major_comments":[{"comment":"The task connectivity matrix is computed by concatenating all task runs and both 0-back and 2-back conditions into a single time series (`task_data`), so every reported contrast is rest versus this pooled task state. This design cannot distinguish working-memory-specific effects from generic task engagement due to arousal, attention, motion, or hemodynamic shifts. This is load-bearing for the abstract's claim that working memory specifically forces the network into a more stable state. Please add a within-task contrast (0-back vs. 2-back) computed with condition-specific concatenation, or a non-memory control task, and report whether the triad and balance measures track memory load; at minimum, explicitly discuss this confound and its bearing on the conclusions.","section":"Functional Connectivity and Results"},{"comment":"As defined in the manuscript, balance energy is a deterministic function of the balanced-triad fraction: E = (N_imbalanced - N_balanced) / N_triads = 1 - 2*N_balanced / N_triads. Therefore the reported 'decrease in balance energy' (p = 6.19e-5) and 'increase in balanced triads' (p = 6.19e-5) are the same measurement expressed in two ways, not independent evidence. The causal wording in the abstract ('The increase of balanced triads forced the network to a more stable state with a lower balance energy level') is not supported by the correlational rest/task design. Please reframe energy as a derived quantity and soften or remove the causal language.","section":"Balanced Energy"},{"comment":"No null model or global-signal control is reported. Because signed links are derived from Pearson correlations and the paper reports a task-related increase in positive links, the triad shifts could in principle be trivially explained by a global increase in positive correlations during task performance (for example, from arousal or motion), with no specific bearing on working memory or network stability. Please report the mean correlation (or distribution of correlations) in rest vs. task, and include null models such as sign-permutation or edge-shuffle that preserve the number of positive/negative edges or the degree sequence, to show that the balanced-triad excess is not a byproduct of the global correlation shift.","section":"Statistical Analysis and Functional Connectivity"},{"comment":"The outlier-removal step is described in a single sentence ('we employed the Interquartile Range (IQR) method to eliminate outliers for extracting the components of SBT') without specifying what variable was filtered, at what level (links, subjects, or triads), and whether the procedure was applied separately to rest and task conditions. Differential outlier handling across conditions could bias the paired comparisons. Please specify the procedure in detail and demonstrate robustness of the main results to the inclusion or exclusion of this step.","section":"Statistical Analysis"}],"minor_comments":[{"comment":"The Results state that positive links increased 'during the working memory task in all conditions, both 0-back and 2-back,' but the Functional Connectivity section concatenates all conditions into one array and computes a single connectivity matrix; this is inconsistent and should be clarified, or per-condition analyses should be provided.","section":"Results"},{"comment":"The text says 'We applied the nonparametric Wilcoxon signed-rank test... We used p < 0.05 (Fisher permutation) as a significant threshold,' which mixes two different testing frameworks; please state exactly which test was used and how the p-values were computed.","section":"Statistical Analysis"},{"comment":"The reference to the HCP preprocessing pipeline is listed as 'Perprocess' and appears to be a typo; please correct the entry and provide full citation details.","section":"References"},{"comment":"Figure 2 shows the four triad types, but the caption is only 'Types of Balanced and Imbalanced Triads'; please add a legend or labels identifying T0, T1, T2, and T3 directly in the figure.","section":"Figures"},{"comment":"The words 'forced' (abstract) and 'induce a demand' (Discussion) imply causation; given the rest/task contrast, please replace these with correlational phrasing such as 'was associated with'.","section":"Abstract and Discussion"},{"comment":"The manuscript states that data and code are 'available upon request'; depositing the analysis code in a public repository would strengthen reproducibility, especially given that the preprocessing and IQR steps are not fully specified.","section":"Data Availability"}],"recommendation":"major_revision","confidential_remarks":"The paper is a straightforward application of existing structural-balance measures to HCP data. The core empirical direction is plausible, but the interpretation as a working-memory-specific stabilization requires a 0-back vs. 2-back contrast and a null-model check; the balance-energy result should be framed as a derived consequence of the triad-count result rather than as independent evidence. I have no concerns about citation integrity or authorship. With those additions, the paper could make a modest but valid contribution to the SBT literature."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things. First, this is the first structural balance theory (SBT) analysis of working memory, and the basic observation—balanced triads increase, imbalanced triads decrease, negative hub tendency drops, all with decent p-values in 138 HCP subjects—is internally consistent. Second, the headline claim that working memory specifically drives the brain to a more stable state is not actually tested. The task data concatenates all blocks, so the only contrast is rest vs. pooled task. That leaves a generic task-engagement confound wide open: any global shift toward positive correlations, from arousal, attention, or motion, would mechanically produce more T3 triads, fewer T1/T2, and lower balance energy. The 0-back vs. 2-back contrast is sitting right there in the HCP WM task and they never run it.\n\nWhat the paper does well: it uses a well-preprocessed public dataset, applies paired nonparametric tests, reports exact p-values, and the region-level findings (temporal, parietal, prefrontal) line up with prior WM work. That is genuine credit. The authors also clearly build on their own earlier SBT papers, which is appropriate, not a flaw.\n\nThe soft spots, in proportion. The energy result is redundant: balance energy is a deterministic function of the balanced-triad fraction (E = 1 - 2B/T), so 'lower energy' adds no independent evidence. No null model is provided, so we cannot tell whether the triad shift exceeds what a degree-preserving or correlation-reshuffling null would produce. There is no multiple-comparison correction across the many link-level tests, and the IQR outlier rule is a free parameter. The sign construction—presumably sign of Pearson r—is never explicitly stated, and code is only 'available upon request.' The causal 'forced the network' language in the abstract overstates what a rest-vs-task comparison can show.\n\nNone of these flaws are fatal to the core descriptive finding, but they do mean the working-memory-specific stability interpretation is not yet established. A serious referee could fix this with a 0-back vs. 2-back contrast, a null model, correction for multiple comparisons, and a code release.\n\nWho this is for: cognitive neuroscientists working on signed network analysis or WM network dynamics. It is a legitimate new application, not a new theory, and it deserves a real peer review rather than a desk reject. I would read the revision, but I would not cite it as evidence until the specificity question is answered.","headline":"A plausible first SBT application to working memory, but the memory-specific stability claim is untested because 0-back and 2-back are pooled; still worth a serious referee.","tokens_in":10815,"tokens_out":1709,"would_cite":false,"duration_ms":17357,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Using structural balance theory on fMRI connectivity, this paper claims that the brain's functional network becomes more stable during working memory than at rest, measured by more balanced triads and lower balance energy.","keywords":["structural balance theory","working memory","fMRI","functional connectivity","N-back task","signed network","brain network stability","triadic balance"],"falsifier":"Run the same analysis on an active control task with matched sensory input and motor responses but no memory load, and compare balanced-triad counts and balance energy to the N-back values; if the control shows the same or larger effect, the shift is not memory-specific. A second check would be to recompute correlations after regressing out head motion and physiological noise and see whether the rest-versus-task difference survives.","tokens_in":9930,"feed_emoji":"🧠","tokens_out":7757,"duration_ms":70410,"temperature":0.7,"pith_summary":"The paper asks whether the brain as a whole shifts toward a more stable configuration when it performs a working memory task. Treating fMRI correlations between brain regions as positive or negative links, it counts triads—three-region motifs—and classifies them as balanced or imbalanced using structural balance theory. Across 138 healthy adults, balanced triads were significantly more numerous during an N-back working memory task than at rest, imbalanced triads were fewer, and the balance energy was lower. The authors read this as evidence that working memory processing pushes the functional network into a more stable, less frustrated state, with the largest sign changes occurring in temporal, parietal, and prefrontal cortices.","feed_headline":"Working memory tilts brain networks toward stable triads","feed_subtitle":"fMRI scans show more balanced, fewer imbalanced three-region motifs during N-back than at rest.","key_machinery":"The signed functional connectivity matrix: for each of 360 cortical regions from a multimodal parcellation, pairwise Pearson correlations of BOLD time series are turned into positive or negative links. On this signed network, structural balance theory classifies every three-node motif by the product of link signs: balanced triads have positive product (all-positive or one-positive-two-negative), imbalanced triads have negative product. The four types T0, T1, T2, and T3 are tabulated separately. Balance energy is defined as $$E = -\\frac{1}{\\binom{N}{3}} \\sum_{i