Stress-sensitive inference of task controllability
Open accessPublished version of a preprint
Abstract
Estimating the controllability of the environment enables agents to better predict upcoming events and decide when to engage controlled action selection. How does the human brain estimate controllability? Trial-by-trial analysis of choices, decision times and neural activity in an explore-and-predict task demonstrate that humans solve this problem by comparing the predictions of an 'actor' model with those of a reduced 'spectator' model of their environment. Neural blood oxygen level-dependent responses within striatal and medial prefrontal areas tracked the instantaneous difference in the prediction errors generated by these two statistical learning models. Blood oxygen level-dependent activity in the posterior cingulate, temporoparietal and prefrontal cortices covaried with changes in estimated controllability. Exposure to inescapable stressors biased controllability estimates downward and increased reliance on the spectator model in an anxiety-dependent fashion. Taken together, these findings provide a mechanistic account of controllability inference and its distortion by stress exposure.
AI snapshot
Written from the paper’s full text.
TL;DR
Ligneul et al. show that humans estimate task controllability by comparing an “actor” model (state–action–state transitions) with a reduced “spectator” model (state–state transitions), with the arbitrator tracking an approximation of transfer entropy. Uncontrollable stress biases this inference toward the spectator model, especially in more anxious participants.
Key findings
- In an explore-and-predict task, Bayesian model comparison favored the SAS′–SS′–Ω controllability model over an actor-only SAS′ model and alternative controllability schemes; the arbitrator ω predicted subjective controllability better than SAS′ alone (72.1% vs 66.5%, N = 50, z(49) = 3.90, d = 0.64, P = 9.7 × 10⁻⁵).
- Actor and spectator prediction errors both slowed decision times and independently explained reaction-time variance; higher estimated controllability (ω) slowed decisions, and this slowing correlated with predictive accuracy (N = 50, ρ = 0.67, P = 1.1 × 10⁻⁷).
- fMRI showed mPFC and nucleus accumbens encoded negatively the difference δSS′ − δSAS′; controllability prediction errors δΩ tracked PCC, right dorsal anterior insula, right TPJ and mPFC; subjective controllability was decodable in precuneus, right TPJ, SMA, left premotor cortex and left dlPFC, with mPFC and right TPJ overlapping across analyses.
- In the stress experiment, uncontrollable versus controllable shocks increased the threshold for inferring controllability (t(52) = 4.56, P = 2.7 × 10⁻⁴, d = 1.1) and reduced the arbitrator ω (z = 2.60, P = 0.009, d = 0.77); state anxiety moderated ω after uncontrollable stressors (r = −0.46, P = 0.015) but not after controllable stressors (r = 0.30, P = 0.12; group difference z = 2.87, P = 0.002).
Methods
- Three experiments used a novel explore-and-predict task with three states and three actions, hidden probabilistic transition rules that were either action-dependent (controllable) or previous-state-only (uncontrollable), covert reversals, and counterfactual prediction pairs: behavioural n = 50, fMRI n = 32, stress n = 54 final (62 recruited, with exclusions).
- The stress experiment used a between-subjects yoked design: participants received controllable or uncontrollable mild electric shocks before the explore-and-predict task, with shock sequences matched across groups.
- Computational modelling compared SAS′ actor and SS′ spectator transition-probability models, with Ω approximating transfer entropy and ω arbitrating between models; Bayesian model comparison used the VBA toolbox. fMRI analyses used SPM12 GLMs and searchlight SVM decoding.
Limitations
- Stated: the explore-and-predict task included no reinforcers, so interactions with reward/punishment processing were not tested; the authors suggest introducing reinforcers in future work. The causal contribution of mPFC and right TPJ remains untested and would require brain stimulation or more invasive methods. The reporting summary states that no blinding procedure was used.
- Stated: no explicit sample-size determination was performed before the behavioural experiment.
- Evident: the fMRI findings are correlational and cannot establish necessity or sufficiency of the identified regions; the stress effects may involve additional regions such as mPFC and dorsal raphe nucleus that were not directly manipulated. The samples were young adults, and the stress experiment included 52 women, limiting generalizability.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
0 of 13 cited sources checked · 0 supported · 0 flagged
- Findingown data§ Stress-sensitive inference of task controllability
BOLD responses in striatal and medial prefrontal areas track the instantaneous difference in prediction errors generated by actor and spectator statistical learning models.
“Neural blood oxygen level-dependent responses within striatal and medial prefrontal areas tracked the instantaneous difference in the prediction errors generated by these two statistical learning models.”
Quote in the paperSubstantial claim - Findingown data§ Dissociation of actor and spectator models.
The mPFC and nucleus accumbens encoded negatively the signal δSS′−δSAS′ required for updating controllability.
“Testing directly the effect of δ SS′δ SAS′ (mathematically equivalent to p SAS′p SS′ ) using a conventional parametric analysis at the whole-brain level showed that the mPFC and the nucleus accumbens encoded negatively this signal required for the update of controllability.”
Quote in the paperSubstantial claim - Findingown data§ Neural correlates of dynamic controllability.
Whole-brain analyses revealed a significant negative relationship between δΩ and neurovascular responses in PCC, right dAI, right TPJ and mPFC.
“Whole-brain analyses revealed a significant negative relationship between δ Ω and neurovascular responses in the posterior cingulate (PCC), the right dorsal anterior insula (dAI), the right TPJ and the mPFC (Fig. 4d and Supplementary Table 4).”
Quote in the paperSubstantial claim - Findingown data§ Stress-sensitive inference of task controllability
BOLD activity in posterior cingulate, temporoparietal and prefrontal cortices covaries with changes in estimated controllability.
“Blood oxygen level-dependent activity in the posterior cingulate, temporoparietal and prefrontal cortices covaried with changes in estimated controllability.”
Quote in the paperSubstantial claim - Findingown data§ Controllability drives learning and predictive decisions.
SAS′-SS′-Ω schemes outperformed the conventional model-based learning algorithm (SAS′ alone) in all experiments.
“Bayesian model comparisons demonstrated that SAS′-SS′-Ω schemes outperformed the conventional model-based learning algorithm (SAS′ alone) in all experiments (Fig. 3a and Supplementary Fig. 2a).”
Quote in the paperSubstantial claim
- Findingown data§ Neural correlates of dynamic controllability.
The threshold parameter increased significantly in the uncontrollable group compared with the controllable group.
“In particular, the threshold parameter increased significantly in the uncontrollable group compared with the controllable group (Fig. 5c, left).”
Quote in the paperSubstantial claim - Findingown data§ Stress-sensitive inference of task controllability
Humans solve controllability estimation by comparing predictions of an actor model with those of a reduced spectator model of their environment.
“Trial-by-trial analysis of choices, decision times and neural activity in an explore-and-predict task demonstrate that humans solve this problem by comparing the predictions of an 'actor' model with those of a reduced 'spectator' model of their environment.”
Quote in the paperSubstantial claim - Findingown data§ Discussion
State anxiety moderated the effect of induction type on the arbitrator variable ω.
“State anxiety moderated the effect of induction type on the arbitrator variable (ω) reflecting controllability estimation.”
Quote in the paperSubstantial claim - Findingown data§ Controllability drives learning and predictive decisions.
The arbitrator ω captured quantitative changes in subjective controllability.
“As expected, the arbitrator ω captured quantitative changes in subjective controllability, indexed by the proneness of participants to predict that different actions would lead to different states in counterfactual prediction trials (Fig. 3b).”
Quote in the paperSubstantial claim - Findingown data§ Controllability drives learning and predictive decisions.
The SAS′-SS′-Ω scheme with arbitration accounted better for dynamics of subjective controllability changes around reversals than SAS′ alone.
“Critically, the SAS′-SS′-Ω scheme that included an arbitration mechanism accounted better for the dynamics of subjective controllability changes around reversals than did the SAS′ model alone (Fig. 3c, correct prediction of subjective controllability: 72.1% versus 66.5%, N = 50, z(49) = 3.90, d = 0.64, P = 9.7 × 10 -5 , 95% confidence interval (CI) 0.004-0.028).”
Quote in the paperSubstantial claim - Findingown data§ Timing
Decodability extended to the dlPFC and ACC in participants who displayed slower reaction time (RT) in controllable contexts.
“Decodability extended to the dlPFC and ACC in participants who displayed slower reaction time (RT) in controllable contexts (red).”
Quote in the paperNot a substantial claim - Findingown data§ Timing
The SS′-SAS′-Ω scheme captured the dynamics of subjective controllability around reversals better than the SAS′ model.
“The SS′-SAS′-Ω scheme captured the dynamics of subjective controllability around reversals better than the SAS′ model.”
Quote in the paperSubstantial claim - Findingown data§ Timing
The arbitrator variable ω has a linear relationship with subjective controllability.
“Normalized distribution of the arbitrator variable ω (grey bar) and its linear relationship with subjective controllability (Subj.contr.; pink line).”
Quote in the paperSubstantial claim - Findingown data§ Timing
Simulations under controllable rule C2 showed that TE requires exploration to be used as a proxy for controllability.
“Simulations under controllable rule C2 showing that Te requires exploration to be used as a proxy for controllability.”
Quote in the paperSubstantial claim - Findingown data§ Timing
Parametric analysis of BOLD responses showed that the mPFC and the nucleus accumbens encoded negatively the difference term δ SS′δ SAS′ used to update controllability.
“Parametric analysis of BOLD responses showed that the mPFC (b) and the nucleus accumbens (c) encoded negatively the difference term δ SS′δ SAS′ used to update controllability.”
Quote in the paperSubstantial claim - Findingown data§ Timing
A searchlight analysis based on the six exploratory trials preceding a prediction pair revealed that the mPFC, the posterior dmPFC, the right TPJ and the precuneus were sensitive to upcoming reports of controllability.
“A searchlight analysis based on the six exploratory trials preceding a prediction pair revealed that the mPFC, the posterior dmPFC, the right TPJ and the precuneus were sensitive to upcoming reports of controllability (pink).”
Quote in the paperSubstantial claim - Findingown data§ Discussion
More anxious individuals displayed lower controllability estimates following uncontrollable stressors in our data.
“who also displayed lower controllability estimates following uncontrollable stressors in our data 19 .”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [19] Alvarez, R. P. et al. Increased anterior insula activity in anxious individuals is linked to diminished perceived control. Transl. Psychiatry 5, e591–e591 (2015). source
Not checked in the free run.
- Findingown data§ Timing
A conjunction analysis revealed brain regions whose activity encoded both δ SS′ and δ SAS′ positively.
“A conjunction analysis revealed brain regions whose activity encoded both δ SS′ and δ SAS′ positively.”
- Also stated in: § Timing · plain
“A conjunction analysis revealed brain regions whose activity encoded both δ SS′ and δ SAS′ positively.”
Quote in the paperSubstantial claim - Also stated in: § Timing · plain
- Findingown data§ Timing
Bayesian group model comparison pooled over three experiments showed the advantage of the SS′-SAS′-Ω model over the actor model (SAS′) alone and two models estimating controllability based on uncertainty (SS′-SAS′-H) or Jensen-Shannon divergence (SS′-SAS′-JS) of SAS′ transitions.
“Bayesian group model comparison pooled over three experiments showed the advantage of the model estimating controllability using an approximation of Te (SS′-SAS′-Ω) over the actor model (SAS′) alone and two models estimating controllability based on the uncertainty (SS′-SAS′-H) or the Jensen-Shannon divergence (SS′-SAS′-H) of SAS′ transitions (Methods and Supplementary Figs.1 and 2).”
Quote in the paperSubstantial claim - Findingown data§ Timing
The mPFC and nucleus accumbens encoded negatively the difference term δ SS′−δ SAS′ used to update controllability.
“Parametric analysis of BOLD responses showed that the mPFC (b) and the nucleus accumbens (c) encoded negatively the difference term δ SS′δ SAS′ used to update controllability.”
Quote in the paperSubstantial claim - Findingown data§ Timing
In the complete absence of exploration, conditional entropies H(S′|S) and H(S′|S,A) are both null because the agent maintains itself indefinitely in a single preferred state.
“In the complete absence of exploration, the conditional entropies H(S′|S) and H(S′|S,A) are both null because the agent maintains itself indefinitely in a single preferred state (Supplementary Fig.1c).”
Quote in the paperNot a substantial claim - Findingown data§ Timing
The right TPJ and mPFC were the only regions highlighted by each of the analyses of decodability of subjective controllability, controllability prediction errors, and difference term δ SS′−δ SAS′.
“The right TPJ and the mPFC were the only regions highlighted by each of these analyses (threshold of each map: P < 0.005 uncorrected).”
Quote in the paperSubstantial claim - Findingown data§ Timing
All areas surviving correction for multiple comparisons for signed second-order prediction errors δΩ showed a negative effect.
“All areas surviving correction for multiple comparisons showed a negative effect, implying greater activity when actions appeared less causal than expected.”
Quote in the paperSubstantial claim - Findingown data§ Timing
A searchlight analysis based on the six exploratory trials preceding a prediction pair revealed that the mPFC, posterior dmPFC, right TPJ, and precuneus were sensitive to upcoming reports of controllability.
“A searchlight analysis based on the six exploratory trials preceding a prediction pair revealed that the mPFC, the posterior dmPFC, the right TPJ and the precuneus were sensitive to upcoming reports of controllability (pink).”
Quote in the paperSubstantial claim - Findingown data§ Timing
The effect of ω on reaction times positively correlated with predictive accuracy across participants (N=50, ρ=0.67, P=1.1e-7, CI 0.45-0.81).
“Positive correlation between the effect of ω on reaction times and predictive accuracy across participants (N = 50, ρ = 0.67, P = 1.1 × 10 -7 , CI 0.45-0.81).”
Quote in the paperNot a substantial claim - Findingown data§ Timing
Both the mPFC and the nucleus accumbens showed no clear-cut modulation by δ SS′.
“Both regions encoded δ SAS′ positively, but showed no clear-cut modulation by δ SS′ .”
Quote in the paperNot a substantial claim - Findingreasoning§ Statistical procedures.
Empirical chance levels were higher than the theoretical level of 1/3 in all experiments (behavioural: 38.0 ± 2.5%; fMRI: 39.4 ± 1.6%; stress: 38.6 ± 2.6%).
“The resulting empirical chance levels were indeed higher than the theoretical level of 1/3 in all experiments (behavioural: 38.0 ± 2.5%; fMRI: 39.4 ± 1.6%; stress: 38.6 ± 2.6%), reflecting the fact that participants knew that three transition rules out of four did not allow state repetitions.”
Quote in the paperSubstantial claim - Findingown data§ Timing
At a more lenient voxel-wise threshold (P < 0.005 uncorrected), the right TPJ also survived correction for multiple comparisons.
“At a more lenient voxel-wise threshold (P < 0.005 uncorrected), the right TPJ also survived correction for multiple comparisons (Supplementary Table3and Fig.4e).”
- Also stated in: § Timing · plain
“At a more lenient voxel-wise threshold (P < 0.005 uncorrected), the right TPJ also survived correction for multiple comparisons (Supplementary Table3and Fig.4e).”
Quote in the paperNot a substantial claim - Also stated in: § Timing · plain
- Findingown data§ Timing
Decodability extended to the dlPFC and ACC in participants who displayed slower reaction time in controllable contexts.
“Decodability extended to the dlPFC and ACC in participants who displayed slower reaction time (RT) in controllable contexts (red).”
Quote in the paperNot a substantial claim - Findingown data§ Timing
Working memory capacity indexed by a two-back task positively correlated with predictive accuracy in the explore-and-predict task for controllable contexts (n=46, ρ=0.52, P=1.9e-4, 95% CI 0.18-0.67).
“Positive correlation between working memory (WM) capacity indexed by a two-back task (Supplementary Methods) and predictive accuracy in the explore-and-predict task for controllable (blue: n = 46, ρ = 0.52, P = 1.9 × 10 -4 , 95% CI 0.18-0.67)”
Quote in the paperNot a substantial claim - Findingown data§ Timing
Simulations showed dissociation of controllability, as indexed by TE, and predictability, as indexed by mutual information shared between successive state-action pairs.
“Simulations showing the dissociation of controllability, as indexed by Te, and predictability, as indexed by the mutual information (MI) shared between successive state-action pairs (random exploration policy).”
Quote in the paperSubstantial claim - Findingown data§ Timing
Predictive accuracy recovered rapidly for all reversal types (first pair after reversal: n=50, t(49)=6.64, P=2.4e-8, d=0.94, 95% CI 0.11-0.20).
“Rapid recovery of predictive accuracy for all reversal types (first pair of predictions after reversal: n = 50, t(49) = 6.64,P = 2.4 × 10 -8 , d = 0.94, 95% CI 0.11-0.20).”
Quote in the paperSubstantial claim - Findingown data§ Timing
Higher state anxiety was not associated with greater reliance on the spectator model after exposure to controllable stressors (N = 27, r = 0.30, P = 0.12, CI -0.16 to 0.61).
“but not after controllable stressors (bright pink, N = 27, r = 0.30, P = 0.12, CI -0.16 to 0.61)”
Quote in the paperSubstantial claim- Rests on an assumption: The arbitrator variable ω indexes reliance on the spectator model.
- Findingown data§ Timing
Higher state anxiety was associated with greater reliance on the spectator model after exposure to uncontrollable stressors (N = 27, r = -0.46, P = 0.015, 95% CI -0.70 to -0.08).
“Higher state anxiety was associated with greater reliance on the spectator model after exposure to uncontrollable stressors (dark pink, N = 27, r = -0.46,P = 0.015, 95% CI -0.70 to -0.08)”
Quote in the paperSubstantial claim- Rests on an assumption: The arbitrator variable ω indexes reliance on the spectator model.
- Findingown data§ Timing
The right TPJ and the mPFC were the only regions highlighted by each of the analyses of decodability of subjective controllability, controllability prediction errors and difference term δ SS′δ SAS′.
“The right TPJ and the mPFC were the only regions highlighted by each of these analyses (threshold of each map: P < 0.005 uncorrected).”
Quote in the paperSubstantial claim - Findingown data§ Timing
State anxiety moderated the effect of induction type on the arbitrator variable (ω) reflecting controllability estimation.
“State anxiety moderated the effect of induction type on the arbitrator variable (ω) reflecting controllability estimation.”
Quote in the paperSubstantial claim- Rests on an assumption: The arbitrator variable ω indexes reliance on the spectator model.
- Findingown data§ Timing
The value above which Ω was treated as evidence for a controllable environment increased significantly following uncontrollable stressors (N = 27 per group, t(52) = 4.56, P = 2.7 × 10 -4 , d = 1.1, 95% CI 0.12-0.39).
“The value above which Ω was treated as evidence for a controllable environment increased significantly following uncontrollable stressors (N = 27 per group, t(52) = 4.56, P = 2.7 × 10 -4 , d = 1.1, 95% CI 0.12-0.39)”
Quote in the paperSubstantial claim - Findingown data§ Timing
There was an interaction between induction group and Pe type on the slowing of decision times induced by δ SS′ and δ SAS′ (F (1,52) = 5.97, P = 0.018, η 2 = 0.07, 95% CI 0.005-0.20).
“interaction group by Pe type: F (1,52) = 5.97, P = 0.018, η 2 = 0.07, 95% CI 0.005-0.20”
Quote in the paperSubstantial claim - Findingown data§ Timing
ω was reduced at the group level following uncontrollable stressors (N = 27 per group, z = 2.60, P = 0.009, d = 0.77, 95% CI 0.08-0.29).
“as indexed by the reduction of ω at the group level (N = 27 per group, z = 2.60, P = 0.009, d = 0.77, 95% CI 0.08-0.29)”
Quote in the paperSubstantial claim - Findingown data§ Timing
Correlation coefficients were significantly different between uncontrollable and controllable group (z = 2.87, P = 0.002).
“Correlation coefficients were significantly different between uncontrollable and controllable group (z = 2.87, P = 0.002).”
Quote in the paperSubstantial claim - Interpretationown data§ Discussion
The ability to estimate controllability involves comparison of actor and spectator models that are dissociable computationally, behaviourally and neurally.
“Our results demonstrate that this ability involves the comparison of actor and spectator models of the ongoing task, which are dissociable computationally, behaviourally and neurally.”
Quote in the paperSubstantial claim - Study design and conductreasoning§ Stress induction task.
In the uncontrollable condition, participants were not able to acquire action-outcome contingencies to prevent the shocks.
“However, in the uncontrollable condition, participants were not able to acquire these action-outcome contingencies to prevent the shocks, whose sequences were merely replayed from the yoked participants performing the controllable condition.”
Quote in the paperSubstantial claim - Study design and conductnone given§ Replication
The conclusions drawn based on fMRI data were robust to qualitative variations in statistical methods.
“The conclusions draw based on fMRI data were robust to qualitative variations in statistical methods.”
Quote in the paperNot a substantial claim - Interpretationown data§ Timing
Greater activity in areas encoding signed second-order prediction errors δ Ω occurred when actions appeared less causal than expected.
“implying greater activity when actions appeared less causal than expected.”
Quote in the paperSubstantial claim - Interpretationown data§ Timing
Greater activity in areas encoding δΩ occurs when actions appear less causal than expected.
“All areas surviving correction for multiple comparisons showed a negative effect, implying greater activity when actions appeared less causal than expected.”
Quote in the paperSubstantial claim - Study design and conductnone given§ Discussion
The explore-and-predict task allowed isolation of the core computations supporting the inference of task controllability by excluding reinforcers and matching uncertainty across contexts.
“In sum, the explore-and-predict task allowed us to isolate the core computations supporting the inference of task controllability by excluding reinforcers and matching uncertainty across contexts.”
Quote in the paperSubstantial claim - Study design and conductcitation§ Sample size
For the fMRI experiment, 32 participants is sufficient to achieve brain-wide correlation of 2nd-level SPM maps superior or equal to 0.75 for working memory, social inference, motor, and emotion recognition tasks.
“Such sample size is sufficient to achieve brain-wide correlation of 2-nd level SPM maps superior or equal to 0.75 for working memory tasks, social inference tasks, motor tasks and emotion recognition tasks (Turner et al., 2018).”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Turner et al., 2018
Not checked in the free run.
- Study design and conductreasoning§ Stress induction task.
A between-subjects yoked control procedure matched the amount and order of aversive outcome stimuli between the controllable and uncontrollable conditions.
“Critically, we employed a between-subjects yoked control procedure to match the amount and order of aversive outcome stimuli between the controllable and uncontrollable conditions.”
Quote in the paperSubstantial claim - Interpretationcitation§ Discussion
Following uncontrollable stressors, increased reliance on the spectator relative to the actor model probably involves several other brain areas, including the mPFC and the dorsal raphe nucleus.
“Yet, this increased reliance on the spectator relative to the actor model following uncontrollable stressors probably involves several other brain areas, including the mPFC and the dorsal raphe nucleus, both of which are highly sensitive to stressor controllability [14][15][16] .”
Quote in the paperNot a substantial claim0 of 3 sources checked- cites [14] Amat, J. et al. Medial prefrontal cortex determines how stressor controllability affects behavior and dorsal raphe nucleus. Nat. Neurosci. 8, 365–371 (2005). source
Not checked in the free run.
- cites [15] Bland, S. T. et al. Stressor controllability modulates stress-induced dopamine and serotonin efflux and morphine-induced serotonin efflux in the medial prefrontal cortex. Neuropsychopharmacology 28, 1589–1596 (2003). source
Not checked in the free run.
- cites [16] Challis, C. et al. Raphe GABAergic neurons mediate the acquisition of avoidance after social defeat. J. Neurosci. 33, 13978–13988 (2013). source
Not checked in the free run.
- Interpretationown data§ Discussion
The mPFC and the right TPJ are the two most promising candidates for the neural implementation of controllability inference.
“The mPFC and the right TPJ emerged as the two most promising candidates for the neural implementation of controllability inference, and it would therefore be interesting to confirm their causal contribution using brain stimulation methods in the future.”
Quote in the paperSubstantial claim - Interpretationown data§ Statistical procedures.
Participants knew that three transition rules out of four did not allow state repetitions.
“reflecting the fact that participants knew that three transition rules out of four did not allow state repetitions.”
Quote in the paperSubstantial claim - Premisecitation§ Discussion
A lower perception of control mediates the exacerbation of dAI responses to physical threats in more anxious individuals.
“Supporting this idea, a study showed that a lower perception of control mediates the exacerbation of dAI responses to physical threats in more anxious individuals, who also displayed lower controllability estimates following uncontrollable stressors in our data 19 .”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [19] Alvarez, R. P. et al. Increased anterior insula activity in anxious individuals is linked to diminished perceived control. Transl. Psychiatry 5, e591–e591 (2015). source
Not checked in the free run.
- Interpretationnone given§ Discussion
The human brain can compute an approximation of TE.
“By showing that the human brain can compute an approximation of TE, our study may help to bridge the gap between neuroscience and artificial intelligence research, where TE plays an important role in solving unsupervised learning problems 55,56 .”
Quote in the paperNot a substantial claim - Interpretationown data§ Timing
Uncontrollable stressors resulted in increased reliance on the spectator model when making predictions.
“resulting in increased reliance on the spectator model when making predictions”
Quote in the paperSubstantial claim- Rests on an assumption: The arbitrator variable ω indexes reliance on the spectator model.
- May combine two claims. 14
- The quote only partly states this claim. 3
- Classified as probably not a substantial claim. 23
May combine two claims.
Findingown data§ Stress-sensitive inference of task controllabilityExposure to inescapable stressors biases controllability estimates downward and increases reliance on the spectator model in an anxiety-dependent fashion.
“Exposure to inescapable stressors biased controllability estimates downward and increased reliance on the spectator model in an anxiety-dependent fashion.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown data§ Dissociation of actor and spectator models.Both actor and spectator prediction errors slowed responding and independently explained variance in decision times.
“We found that both types of prediction errors slowed responding (Fig. 3d) and independently explained variance in decision times (Supplementary Fig. 3).”
Quote in the paperSubstantial claimThe quote only partly states this claim.
Findingown data§ TimingAll areas surviving correction for multiple comparisons showed a negative effect for signed second-order prediction errors δ Ω.
“All areas surviving correction for multiple comparisons showed a negative effect”
Quote in the paperSubstantial claimMay combine two claims.
Findingown data§ TimingIn the behavioural experiment, actor prediction error (δ SAS′), spectator prediction error (δ SS′), and ω were significant predictors of log-transformed reaction times (N=50; δ SAS′: t(49)=6.17, P=1.1e-6, d=0.79, CI 0.6-0.12; δ SS′: t(49)=4.39, P=0.001, d=0.49, CI 0.3-0.11; ω: z=4.10, P=4.1e-5, d=0.74, CI 0.08-0.18).
“Coefficients of the regression predicting log-transformed reaction times in the behavioural experiment using actor and spectator prediction errors as well as ω (N = 50; δ SAS′ : t(49) = 6.17,P = 1.1 × 10 -6 , d = 0.79, CI 0.6-0.12;δ SS′ : t(49) = 4.39, P = 0.001, d = 0.49, CI 0.3-0.11;ω: z = 4.10, P = 4.1 × 10 -5 , d = 0.74, CI 0.08-0.18).”
Quote in the paperSubstantial claimMay combine two claims.
Findingown data§ TimingAccuracy in prediction trials was above chance for both conditions in each of the experiments (behavioural: n=50, t(49)=11.31, P=2.9e-15, d=1.60, 95% CI 0.20-0.29; fMRI: n=32, t(31)=12.67, P=8.5e-14, d=2.24, 95% CI 0.24-0.33; stress: n=54, t(49)=11.64, P=3.1e-16, d=1, 95% CI 0.20-0.28).
“Accuracy in the prediction trials was above chance for both conditions in each of the experiments (behavioural: n = 50, t(49) = 11.31,P = 2.9 × 10 -15 , d = 1.60, 95% CI 0.20-0.29;fMRI: n = 32, t(31) = 12.67, P = 8.5 × 10 -14 , d = 2.24, 95% CI 0.24-0.33;stress: n = 54, t(49) = 11.64,P = 3.1 × 10 -16 , d = 1, 95% CI 0.20-0.28;Supplementary Table 1).”
Quote in the paperSubstantial claimMay combine two claims.
Findingown data§ TimingBoth the mPFC and the nucleus accumbens encoded δ SAS′ positively.
“Both regions encoded δ SAS′ positively, but showed no clear-cut modulation by δ SS′ .”
Quote in the paperSubstantial claimMay combine two claims.
Findingown data§ TimingBoth mPFC and nucleus accumbens encoded δ SAS′ positively, but showed no clear-cut modulation by δ SS′.
“Both regions encoded δ SAS′ positively, but showed no clear-cut modulation by δ SS′ .”
Quote in the paperSubstantial claimMay combine two claims.
Interpretationown data§ DiscussionExposure to uncontrollable stressors biases controllability estimation, establishing relevance for neuropsychiatric disorders involving altered perceptions of controllability.
“Furthermore, exposure to uncontrollable stressors biases this process assessed by the explore-and-predict task, hence establishing its relevance for the study of neuropsychiatric disorders involving altered perceptions of controllability 9,12,13,35 .”
Quote in the paperSubstantial claim0 of 4 sources checked- cites [9] Cheng, C., Cheung, S. F., Chio, J. H. & Chan, M.-P. S. Cultural meaning of perceived control: a meta-analysis of locus of control and psychological symptoms across 18 cultural regions. Psychol. Bull. 139, 152–188 (2013). source
Not checked in the free run.
- cites [12] Gillan, C. M. et al. Obsessive–compulsive disorder patients have a reduced sense of control on the illusion of control task. Front. Psychol. 5, 204 (2014). source
Not checked in the free run.
- cites [13] Diener, C., Kuehner, C., Brusniak, W., Struve, M. & Flor, H. Effects of stressor controllability on psychophysiological, cognitive and behavioural responses in patients with major depression and dysthymia. Psychol. Med. 39, 77–86 (2009). source
Not checked in the free run.
- cites [35] Voss, M., Chambon, V., Wenke, D., Kühn, S. & Haggard, P. In and out of control: brain mechanisms linking fluency of action selection to self-agency in patients with schizophrenia. Brain 140, 2226–2239 (2017). source
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductnone given§ Participants.Prediction trials were pseudo-randomly ordered with the constraint that each state would be tested a similar number of times.
“Prediction trials were pseudo-randomly ordered with the constraint that each state would be tested a similar number of times.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductnone given§ BlindingNo blinding procedure was used.
“No blinding procedure was used”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductnone given§ Stress induction task.Data collection and analysis were not performed blind to the conditions of the experiments.
“Data collection and analysis were not performed blind to the conditions of the experiments.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based). May combine two claims. The quote only partly states this claim.
Study design and conductnone given§ Participants.One additional participant was excluded a posteriori from the fMRI experiment, and four participants were excluded from the stress experiments together with their yoked counterparts; no subjects were excluded from the behavioral experiment, and six additional subjects were excluded for the analysis of brainstem responses.
“One additional participant was excluded a posteriori from the fMRI experiment, and four participants were excluded from the stress experiments, together with their yoked counterparts (see Supplementary Methods for details on exclusion and inclusion criteria).”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
Study design and conductnone given§ Model fitting procedures.All statistical inferences based on whole-brain analyses satisfied the standard multiple comparison threshold (P FWE < 0.05) at the cluster level unless otherwise indicated.
“All statistical inferences based on whole-brain analyses satisfied the standard multiple comparison threshold (P FWE < 0.05) at the cluster level unless otherwise indicated.”
Quote in the paperSubstantial claimClassified as probably not a substantial claim.
Study design and conductnone given§ TimingStatistical inferences were based on whole-brain effects at standard thresholds (voxel-wise: P < 0.001 uncorrected; cluster-wise: P < 0.05 FWE).
“Statistical inferences were based on whole-brain effects at standard thresholds (voxel-wise: P < 0.001, uncorrected; cluster-wise: P < 0.05 FWe ).”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductreasoning§ Sample sizeNo explicit sample size determination was performed before the behavioral experiment because the explore-and-predict task was a new experimental paradigm with no equivalent in the literature.
“The explore-and-predict task being a new experimental paradigm with no equivalent in the literature, no explicit sample size determination was performed before the behavioral experiment.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
Study design and conductreasoning§ Participants.In the fMRI and stress tasks, feedback was delivered only after one of the two counterfactual predictions to prevent participants from inferring whether the rule was controllable based on feedback.
“In the fMRI and stress tasks, feedback was delivered only after one of the two counterfactual predictions, to prevent participants from inferring whether the rule was controllable or not based on feedback.”
Quote in the paperSubstantial claimThe quote only partly states this claim.
Study design and conductcitation§ Sample sizeFor the stress experiment, post-hoc power calculation indicates that an effect size of 0.931 (Cohen's d) requires a sample size of 26 participants per group to achieve a power of 0.95 (alpha: 0.05).
“Post-hoc power calculation indicates that this effect size requires a sample size of 26 participants per group to achieve a power of 0.95 (alpha: 0.05).”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites Wanke and Schwabe
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductnone given§ Model fitting procedures.Decoding analyses used a leave-one-run-out cross-validation scheme with 100 permutations per subject, so that classes remained balanced for training.
“We used a leave-one-run-out cross-validation scheme with 100 permutations per subject, so that classes remained balanced for training.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductnone given§ RandomizationThe lateral ordering of action colors and possible upcoming states was pseudorandomized to counterbalance lateralization effects within subjects.
“The lateral ordering of action colors (exploration trials) and possible upcoming states (prediction trials) was pseudorandomized so as to counterbalance lateralization effects within subjects.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductcitation§ Model fitting procedures.The fMRI sequencing protocol was optimized from the recommended imaging guidelines of the Human Connectome Project, with fast acquisition speed facilitating detection and removal of non-neuronal contributions to BOLD changes.
“This state-of-the-art sequencing protocol was optimized from the recommended imaging guidelines of the Human Connectome Project, with the fast acquisition speed facilitating the detection and removal of non-neuronal contributions to BOLD changes (http://protocols. humanconnectome.org/HCP/3T/imaging-protocols.html).”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductreasoning§ Model fitting procedures.The mixed-effects analyses on BOLD signals were not used for statistical inference but merely for visualization purposes.
“Importantly, this approach was not used for statistical inference (since doing so would constitute double dipping) but merely for visualization purposes.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductcitation§ Sample sizeIn order to maximize accuracy of individual parameter estimation, the number of trials was maximized without exceeding 40 minutes, often considered a point of diminishing returns for task-based fMRI studies.
“In order to maximize the accuracy of individual parameter estimation, we maximized the number of trials without exceeding the threshold of 40 minutes often considered as a point of diminishing returns for task-based fMRI studies Evan Lee, 2019, Communication Biology).”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites Evan Lee, 2019
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductown data§ Sample sizeThe stress experiment included 54 participants in total.
“For the stress experiment, we included 54 participants in total.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductnone given§ TimingRule C2 was the only rule allowing state repetition, a feature that was taken into account in the analyses.
“Note that rule C2 was the only rule allowing state repetition, a feature that was taken into account in our analyses.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
Study design and conductnone given§ Stress induction task.Participants were randomized in blocks of four where the controllable condition of a yoked pair was always administered first to create the schedule for the yoked counterpart in the uncontrollable condition.
“We randomized participants in blocks of four where the controllable condition of a yoked pair was always administered first to create the schedule for the yoked counterpart in the uncontrollable condition.”
Quote in the paperSubstantial claimMay combine two claims.
Study design and conductreasoning§ Stress induction task.In the controllable condition, all participants underwent the whole schedule with a minimum of 24 trials and were able to acquire the correct response for each cue.
“As such, all participants underwent the whole schedule with a minimum of 24 trials and were able to acquire the correct response for each cue.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductcitation§ Model fitting procedures.Model fitting was performed using a variational Bayesian estimation procedure using the well-validated VBA toolbox.
“Model fitting was performed using a variational Bayesian estimation procedure using the well-validated VBA toolbox 58 .”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [58] Daunizeau, J., Adam, V. & Rigoux, L. VBA: a probabilistic treatment of nonlinear models for neurobiological and behavioural data. PLoS Comput. Biol. 10, e1003441 (2014). source
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductreasoning§ Model fitting procedures.Since testing sets were not constrained to have balanced classes, balanced accuracies were used when reporting searchlight analysis results.
“Since we did not constrain the testing sets to have balanced classes, balanced accuracies were used when reporting the results of the searchlight analysis (performed within an 8 mm sphere) at the whole-brain level.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim. May combine two claims.
Study design and conductreasoning§ RandomizationThe ordering of blocks was counterbalanced across subjects in the explore-and-predict task, so that controllability and rule reversal effects could not be explained by time.
“The ordering of blocks was counterbalanced across subjects in the explore-and-predict task, so that controllability and rule reversal effects could not be explained by time.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductnone given§ Additional informationNo volume censoring was needed.
“No volume censoring needed.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductnone given§ Additional informationHead motion was minimized using tight cushions (<1.5mm).
“Head motion was minimized using tight cushions (<1.5mm)”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductnone given§ Additional informationA complete leave-one-run out cross-validation procedure was used.
“A complete leave-one-run out cross-validation procedure was used.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
Study design and conductnone given§ Additional informationWithin each fold, controllable and uncontrollable targets were equalized using a bootstrapping approach (10 draws per fold).
“Within each fold, controllable and uncontrollable targets were in equalized using a bootstrapping approach (10 draw per fold).”
Quote in the paperSubstantial claim
The paper relies on this without support.
Little is known about the algorithms by which the brain dynamically estimates the degree to which a task is controllable.
“Yet, little is known about the algorithms by which the brain estimates dynamically the degree to which a task is controllable.”
The paper relies on this without support.
The origins of the maladaptive strategy induced by uncontrollable stressors remain poorly understood.
“Indeed, exposure to uncontrollable stressors is known to induce passive responses to subsequent controllable stressors, but the origins of this maladaptive strategy remain poorly understood.”
The authors state this as an assumption.
The data are normally distributed following arcsin transformation.
“assuming normal distribution of the data following arcsin transformation”
The paper relies on this without support.
Subjective controllability can be derived from whether counterfactual predictions differ.
“Pairs of predictions always probed the two actions available for a given state (for example, blue followed by yellow in the circle state), to derive subjective controllability from counterfactual responses.”
The paper relies on this without support.
Electric stimuli at the individually determined 'just bearable, but not yet painful' intensity are sufficiently aversive to induce stress.
“First, individual levels of intensity of the electric stimulus for the manipulation task were determined using a stepwise procedure in which the intensity of the stimulus was gradually increased until participants reported a 'just bearable, but not yet painful' experience of shock.”
The paper relies on this without support.
BOLD signal changes in the identified regions reflect the neural computations of controllability inference.
“Parametric analysis of BOLD responses showed that the mPFC (b) and the nucleus accumbens (c) encoded negatively the difference term δ SS′δ SAS′ used to update controllability.”
The paper relies on this without support.
The arbitrator variable ω indexes reliance on the spectator model.
“resulting in increased reliance on the spectator model when making predictions, as indexed by the reduction of ω at the group level”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In previous studies, dmPFC has been linked to confidence and the sense of agency.
“In previous studies, dmPFC has been linked to confidence and the sense of agency.7”
From Two prefrontal-subcortical circuits estimate controllability and reflect its impact on learning (2026) · cites this paper as [24]
The Discussion (page unavailable) links dmPFC activity to subjective controllability, which is related to a sense of agency. It does not link dmPFC to confidence, and the dmPFC finding described there is this paper’s own result rather than a finding it attributes to previous studies. (LLM, conf=0.78)
Experimental — accuracy not yet measured
Statistics that don’t add up
Each reported p-value recomputed from its test statistic and degrees of freedom.
p-value fits an unstated one-tailed test: (z = 2.87, p = .002
MinorThe paper reports p = .002 here; recomputing from the test statistic and its degrees of freedom gives p = .004 (two-tailed). The reported value is what a one-tailed test gives, but the paper does not say near this result that a one-tailed test was used.
Reported p = .002 · recomputed p = .004
“(z = 2.87, P = 0.002”
Inconsistent p-value: t(52) = 0.81, p = .43
MinorThe paper reports p = .43 here; recomputing from the test statistic and its degrees of freedom gives p = .422 (two-tailed). The check cannot tell whether the statistic or the p-value is the misprint.
Reported p = .43 · recomputed p = .422
“t(52) = 0.81, P = 0.43”
Inconsistent p-value: t(52) = 0.56, p = .59
MinorThe paper reports p = .59 here; recomputing from the test statistic and its degrees of freedom gives p = .578 (two-tailed). The check cannot tell whether the statistic or the p-value is the misprint.
Reported p = .59 · recomputed p = .578
“t(52) = 0.56, P = 0.59”
Inconsistent p-value: t(53) = 2.29, p = .025
MinorThe paper reports p = .025 here; recomputing from the test statistic and its degrees of freedom gives p = .026 (two-tailed). The check cannot tell whether the statistic or the p-value is the misprint.
Reported p = .025 · recomputed p = .026
“t(53) = 2.29, P = 0.025”
Statistics recomputed
Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.
- Doesn’t add upt(53) = 2.29reported p = .025 · recomputed p = .026
“t(53) = 2.29, p = .025”
- Doesn’t add upt(52) = 0.81reported p = .43 · recomputed p = .422
“t(52) = 0.81, p = .43”
- Doesn’t add upt(52) = 0.56reported p = .59 · recomputed p = .578
“t(52) = 0.56, p = .59”
- Doesn’t add upz = 2.87reported p = .002 · recomputed p = .004
“(z = 2.87, p = .002”
- Consistentt(49) = 2.6reported p = .012 · recomputed p = .012
“t(49) = 2.6, p = .012”
- Consistentt(31) = 1.85reported p = .07 · recomputed p = .074
“t(31) = 1.85,p = .07”
- Consistentz = 1.69reported p = .045 · recomputed p = .091
“(z = 1.69,p = .045”
The paper states a one-tailed test
- Consistentt(52) = -2.45reported p = .018 · recomputed p = .018
“t(52) = -2.45,p = .018”
- Consistentt(52) = 1.86reported p = .068 · recomputed p = .069
“t(52) = 1.86,p = .068”
- Consistentz = 2.6reported p = .009 · recomputed p = .009
“z = 2.60, p = .009”
- ConsistentF(1, 52) = 5.97reported p = .018 · recomputed p = .018
“F (1,52) = 5.97, p = .018”
- Consistentt(49) = 4.39reported p = .001 · recomputed p < .001
“t(49) = 4.39, p = .001”
The paper prints p at a floor (such as p < .001)
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Stress accuracy test reports degrees of freedom for the wrong sample size
MinorThe figure reports 54 stress participants but gives t(49) for the accuracy test. A one-sample t test with 54 participants has 53 degrees of freedom, as reported for another stress-sample test.
“stress: n = 54, t(49) = 11.64,P = 3.1 × 10 -16”
“stress: 53.2 ± 10.3%, t(53) = 2.29”
Jensen-Shannon model is labelled H instead of JS
MinorThe Figure 3 caption assigns the Jensen-Shannon divergence model the H suffix, which the paper elsewhere assigns to the uncertainty model. The Jensen-Shannon model is otherwise labelled JS.
“two models estimating controllability based on the uncertainty (SS′-SAS′-H) or the Jensen-Shannon divergence (SS′-SAS′-H) of SAS′ transitions”
“the uncertainty of SAS′ transitions (SAS′-SS′-H, lower during periods of high controllability) and the Jensen-Shannon divergence of counterfactual SAS′ transitions (SAS′-SS′-JS, higher during periods of high controllability)”
Stress-induction task has both three and four cues
MinorThe Results describe three possible cues in the controllable stress task, whereas Methods specify four cues and say participants learned a response for each.
“Participants in the controllable group learned to avoid the shock following one of the three possible cues by pressing the correct response button (out of six alternatives).”
“In the controllable condition, a total of four cues (different in shape and colour) were presented for at least six repetitions each following a pre-programmed pseudo-randomized schedule.”
The citation lists haven’t been fetched yet
References 58 · Cited by 60
Research works printed but not linked: 2
- Kim, D., Park, G. J., O’Doherty, J.-D. & Lee, S. W. Task complexity interacts with state-space uncertainty in the arbitration between model-based and model-free learning. Nat. Commun. 10, 5738 (2019). · DOI
- Mohamed, S. & Jimenez Rezende, D. Variational information maximisation for intrinsically motivated reinforcement learning. Adv. Neural Inf. Process. Syst. 28, 2125–2133 (2015).
- Authors
- Ligneul, R., Mainen, Z. F., Ly, V., & Cools, R.
- Affiliations
- Radboud University Nijmegen (NL)
- Leiden University (NL)
- Corresponding author
- Ligneul, Romain
- Cools, Roshan
- Venue
- Nature Human Behaviour
- Published
- 2022-03-10
- Volume
- 6
- Issue
- 6
- Pages
- 812-822
- ISSN
- 2397-3374 (ISSN-L)
- OpenAlex
- W4220838973
- Semantic Scholar
- 69c7b90f02eff8ea1479d4e6bbc7786d32fa93be
- Cited by
- 60
- References
- 58
- Citations by year
- 2026: 14 · 2025: 22 · 2024: 14 · 2023: 5 · 2022: 4 · 2021: 1
- FWCI
- 5.80
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- Nature Human Behaviour (publishedVersion)
- PubMed (publishedVersion)
- Leiden Repository (Leiden University) (submittedVersion)
- Radboud Repository (Radboud University) (submittedVersion)
- HAL (submittedVersion)
- Data Archiving and Networked Services (DANS) (submittedVersion)
- Funders
- Fondation Fyssen (N/A)
- Brain Research Foundation (N/A)
- Universiteit Leiden (N/A)
- Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO 453-14-005)
- Topics
- Life Sciences › Neuroscience › Cognitive Neuroscience › Neural and Behavioral Psychology Studies
- Life Sciences › Neuroscience › Cognitive Neuroscience › Memory and Neural Mechanisms
- Life Sciences › Neuroscience › Cellular and Molecular Neuroscience › Neurotransmitter Receptor Influence on Behavior
- Keywords
- stress, anxiety, prediction error, actor model, spectator model, statistical learning, striatum, medial prefrontal cortex, posterior cingulate cortex, temporoparietal cortex, decision-making, action selection, BOLD signal
- MeSH terms
- Anxiety, Brain, Brain Mapping, Humans, Learning, Prefrontal Cortex
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
