Caption reader

stale · v4 open awaiting approval

What do the figure captions assert?

for Contributions of insula and superior temporal sulcus to interpersonal guilt and responsibility in social decisions · this layer across all papers · json

Awaiting approval

Waiting for approval. That is a statement about the record, not about whether anyone has read this: people read the corpus without stamping what they read, and only a stamp leaves a trace. Approval is an operation on a version, not a step of its own — it is recorded against the version it was granted to, so running this layer again does not carry it forward.

Out of date

These inputs changed after this ran:

runs/gadeke-2026-guilt-insula/prepared.json

What it produced25 claims

Read from runs/gadeke-2026-guilt-insula/caption-reader.output.json · 17 KB. agent captionmodel supplied:runs/gadeke-2026-guilt-insula/caption-reader.answer.v4.json

Participants chose the risky option (lottery) slightly more often in the Solo condition than in the Social condition in Study 1, but not in Study 2.

panel fig2a, fig2dclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “Participants chose the risky option slightly more often in the Solo condition than in the Social condition in Study 1 ( A ) but not in Study 2 ( D ).”
Risk premiums did not differ between the Solo and Social conditions in either study.

panel fig2b, fig2eclaim_type empiricalrole controlconfidence highevidence_verified trueevidence_verified_against slice

evidence “( B, E ) Risk premiums did not differ between Solo and Social conditions.”
notes “Null result presented as evidence against social-context-driven changes in risk aversion.”
Values of the risk aversion parameter ρ were broadly consistent with risk premiums but showed that participants were slightly more risk averse in the Social than in the Solo condition in Study 1 only.

panel fig2c, fig2fclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “Values of the risk aversion parameter ρ in the Solo and Social conditions were broadly consistent with Risk premium values, but showed that participants were slightly more risk averse in the Social than in the Solo condition in Study 1 only (see Results).”
Participant momentary happiness varied with the rewards received by the participant.

panel fig3a, fig3eclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “Happiness varied with rewards received by the participant ( A, E ) and by the partner ( B, F ).”
Participant momentary happiness varied with the rewards received by the partner.

panel fig3b, fig3fclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “Happiness varied with rewards received by the participant ( A, E ) and by the partner ( B, F ).”
The Responsibility Redux model, taking into account expected, previous and current rewards, reward prediction errors for both participant and partner, and decision-maker, predicted the variations in participants' momentary happiness well.

panel fig3c, fig3gclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “A computational model taking into account expected, previous and current rewards, reward prediction errors for both participant and partner, and decision-maker (Responsibility Redux model, see Results) predicted the variations in participants’ momentary happiness well”
notes “Model fit also referenced to Table 1; the model-based regressors used in the fMRI analyses depend on this fit.”
Responsibility for low lottery outcomes for the partner decreased participant happiness more than the same outcomes following partner choices, fitting the definition of interpersonal guilt.

panel fig3d, fig3hclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “Crucially, responsibility for low lottery outcomes for the partner decreased participant happiness more than the same outcomes following partner choices (see Results), which fits the definition of interpersonal guilt.”
notes “Significance of the guilt effect indicated by stars: ***p < 0.001; **p < 0.01; numeric coefficients are not in the caption.”
The parameters of the Responsibility Redux model could be recovered from synthetic data generated from each participant's estimated parameters, supporting the stability of the fitted parameters.

panel fig3s1claim_type assessmentrole methodologicalconfidence tentativeevidence_verified trueevidence_verified_against slice

evidence “Stability of the estimated parameters of the temporal difference models was evaluated by attempting to recover parameters from synthetic data created using each participant’s real estimated parameters.”
notes “The caption describes the recovery procedure and the regression of recovered on actual parameters but does not state the recovery result numerically.”
A set of regions showed a greater BOLD response when participants chose the risky (lottery) rather than the safe option, irrespective of Social or Solo condition.

panel fig4aclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “( A ) Regions showing a greater response when participants chose the risky (lottery) rather than the safe option, irrespective of Social or Solo condition.”
Regions showed a greater BOLD response when participants chose for both themselves and their partner rather than just for themselves (Social > Solo).

panel fig4bclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “( B ) Regions showing a greater response when participants chose for both themselves and their partner rather than just for themselves (Social > Solo).”
Linear mixed model coefficients indicate that precuneus and TPJ were most active when participants chose the lottery in the Social condition.

panel fig4cclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “Coefficients of linear mixed models (LMMs) indicate that two of these regions, precuneus and TPJ, were most active when participants chose the lottery in the Social condition.”
notes “Caption states all coefficients and differences are significantly different from 0 (see Appendix 1—table 4).”
Brain regions were more active during receipt of the outcomes of lotteries than of safe choices across all conditions.

panel fig4dclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “( D ) Brain regions more active during receipt of the outcomes of lotteries than safe choices (all conditions).”
Insula ROIs mirrored the behavioural guilt effect: voxels responded more to low lottery outcomes for the partner when these resulted from the participant's rather than the partner's choices, even when responses to high outcomes were subtracted.

panel fig4eclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “voxels here responded more to low lottery outcomes (L) for the partner when these resulted from participant’s rather than the partner’s choices, even when responses to high outcomes were subtracted (L–H).”
notes “Caption states all coefficients and differences are significantly different from 0 (see Appendix 1—table 6).”
A mass-univariate, voxel-wise analysis showed a cluster within the left insula ROI with higher responses to low lottery outcomes for the partner when these resulted from participant rather than partner choices.

panel fig4fclaim_type empiricalrole controlconfidence highevidence_verified trueevidence_verified_against slice

evidence “A cluster of voxels within the left insula ROI showed higher responses to low lottery outcomes for the partner if these resulted from participant rather than partner choices.”
notes “Convergent voxel-wise confirmation of the ROI-based insula guilt effect in panel E, described as a compatible result.”
Activation in bilateral ventral striatum was explained by a computational model-based regressor coding participant rewards.

panel fig4gclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “( G ) Activation in bilateral ventral striatum explained by a computational model-based regressor coding participant rewards.”
A cluster in the left superior temporal sulcus region showed a higher response to partner reward prediction errors resulting from participant rather than partner choices.

panel fig4hclaim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “one cluster in the left superior temporal sulcus region showed a higher response to partner reward prediction errors resulting from participant rather than partner choices.”
notes “Restricted to brain regions sensitive to outcomes of risky choices.”
The left superior temporal sulcus cluster responded to model-based regressors coding participant reward prediction resulting from participant and partner choices across both sessions of the experiment.

panel fig4iclaim_type empiricalrole empiricalconfidence tentativeevidence_verified trueevidence_verified_against slice

evidence “( I ) Response in this cluster to the computational-model-based regressors coding participant reward prediction resulting from participant and partner choices, for both sessions of the experiment.”
notes “Caption describes what is plotted (coefficients with 95% confidence intervals) rather than stating a directional result.”
Functional connectivity between the left anterior insula (seed) and a cluster in the right inferior frontal gyrus at the time of choice varied as a function of condition (Social vs. Solo) and choice (Risky or Safe).

panel fig5claim_type empiricalrole empiricalconfidence tentativeevidence_verified trueevidence_verified_against slice

evidence “Changes in functional connectivity between the left anterior insula (seed) and a cluster in the right inferior frontal gyrus at the time of the choice as a function of condition (Social vs. Solo) and choice (Risky or Safe).”
notes “Caption states the analysis but not the direction or significance of the connectivity effect.”
Connectivity between the left superior temporal sulcus (seed) and a cluster in the left inferior frontal gyrus was highest when participants made Safe choices for themselves and Risky choices for both players, an opposite pattern that did not survive correction for multiple comparisons.

panel fig5s1claim_type empiricalrole controlconfidence highevidence_verified trueevidence_verified_against slice

evidence “connectivity was highest when participants made Safe choices for themselves and Risky choices for both players (p uncorrected = 0.001, T = 4.44, Z = 4.30, 35 voxels, peak at MNI [–48 14 6]).”
notes “This cluster did not survive corrections for multiple tests; reported at an uncorrected threshold.”
Among the computational models fitted to momentary happiness data, the Responsibility Redux model achieved the best (lowest) AIC in both studies (Study 1 AIC –1499; Study 2 AIC –1195).

panel table1claim_type assessmentrole methodologicalconfidence tentativeevidence_verified trueevidence_verified_against slice

evidence “Responsibility Redux 4 0.361 0.331 –999 –1499”
notes “Table lists R2, adjusted R2, BIC and AIC for five models per study; the best-fitting model is inferred from the lowest AIC values (Study 2 Responsibility Redux AIC –1195).”
In mixed-effects regressions on choices, the Social condition significantly increased choice of the risky option in Study 1 but not in Study 2.

panel app1table1claim_type empiricalrole empiricalconfidence tentativeevidence_verified trueevidence_verified_against slice

evidence “Condition Social 0.14* 0.03^ 0.01 0.01”
notes “Row values are Study 1 probit (0.14*), Study 1 linear (0.03^), Study 2 probit (0.01), Study 2 linear (0.01); column identity is read from the table header.”
The linear mixed model containing all three two-way interaction terms (Model 5) best explained the happiness data in both studies, and its crucial partnerHigh:participantDecided (guilt) interaction was significant.

panel app1table2claim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “In both studies, Model 5 ( Equation 9 in the Results section of the main text), which contained all three two-way interaction terms, explained the data best, so its parameters for the crucial partnerHigh:participantDecided interaction are reported in the main text.”
notes “Model 5 partnerHigh:participantDecided coefficient is 0.39*** in Study 1 and 0.31** in Study 2.”
During the outcome phase, responses to low lottery outcomes were higher in the Social than the Partner condition in both left and right insula and lower in the right middle temporal cortex.

panel app1table9claim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “Social 0.41*** 0.18*** –0.12**”
notes “Columns are InsulaL (0.41***), InsulaR (0.18***) and MidTempR (–0.12**).”
The difference in response between low and high lottery outcomes was greater in the Social than the Partner condition in left insula, right insula, and right middle temporal cortex.

panel app1table10claim_type empiricalrole empiricalconfidence highevidence_verified trueevidence_verified_against slice

evidence “Social 0.44*** 0.19*** 0.67***”
notes “Columns are InsulaL (0.44***), InsulaR (0.19***) and MidTempR (0.67***).”
Participants rated their partners highly on sympathy, cooperation, honesty, openness, and sociability in both studies.

panel app1table11claim_type empiricalrole controlconfidence highevidence_verified trueevidence_verified_against slice

evidence “How honest did they seem? 9.05 (1.11) 9.34 (1.10)”
notes “Manipulation check on partner perception; ratings across the five items range roughly 8.35–9.34 across Studies 1 and 2.”

#	claim	panel	claim_type	role	addresses	evidence	confidence	span	notes	evidence_verified	evidence_verified_against
1	Participants chose the risky option (lottery) slightly more often in the Solo condition than in the Social condition in Study 1, but not in Study 2.	fig2a, fig2d	empirical	empirical	—	Participants chose the risky option slightly more often in the Solo condition than in the Social condition in Study 1 ( A ) but not in Study 2 ( D ).	high	—	—	true	slice
2	Risk premiums did not differ between the Solo and Social conditions in either study.	fig2b, fig2e	empirical	control	—	( B, E ) Risk premiums did not differ between Solo and Social conditions.	high	—	Null result presented as evidence against social-context-driven changes in risk aversion.	true	slice
3	Values of the risk aversion parameter ρ were broadly consistent with risk premiums but showed that participants were slightly more risk averse in the Social than in the Solo condition in Study 1 only.	fig2c, fig2f	empirical	empirical	—	Values of the risk aversion parameter ρ in the Solo and Social conditions were broadly consistent with Risk premium values, but showed that participants were slightly more risk averse in the Social th…	high	—	—	true	slice
4	Participant momentary happiness varied with the rewards received by the participant.	fig3a, fig3e	empirical	empirical	—	Happiness varied with rewards received by the participant ( A, E ) and by the partner ( B, F ).	high	—	—	true	slice
5	Participant momentary happiness varied with the rewards received by the partner.	fig3b, fig3f	empirical	empirical	—	Happiness varied with rewards received by the participant ( A, E ) and by the partner ( B, F ).	high	—	—	true	slice
6	The Responsibility Redux model, taking into account expected, previous and current rewards, reward prediction errors for both participant and partner, and decision-maker, predicted the variations in p…	fig3c, fig3g	empirical	empirical	—	A computational model taking into account expected, previous and current rewards, reward prediction errors for both participant and partner, and decision-maker (Responsibility Redux model, see Results…	high	—	Model fit also referenced to Table 1; the model-based regressors used in the fMRI analyses depend on this fit.	true	slice
7	Responsibility for low lottery outcomes for the partner decreased participant happiness more than the same outcomes following partner choices, fitting the definition of interpersonal guilt.	fig3d, fig3h	empirical	empirical	—	Crucially, responsibility for low lottery outcomes for the partner decreased participant happiness more than the same outcomes following partner choices (see Results), which fits the definition of int…	high	—	Significance of the guilt effect indicated by stars: *p < 0.001; p < 0.01; numeric coefficients are not in the caption.	true	slice
8	The parameters of the Responsibility Redux model could be recovered from synthetic data generated from each participant's estimated parameters, supporting the stability of the fitted parameters.	fig3s1	assessment	methodological	—	Stability of the estimated parameters of the temporal difference models was evaluated by attempting to recover parameters from synthetic data created using each participant’s real estimated parameters…	tentative	—	The caption describes the recovery procedure and the regression of recovered on actual parameters but does not state the recovery result numerically.	true	slice
9	A set of regions showed a greater BOLD response when participants chose the risky (lottery) rather than the safe option, irrespective of Social or Solo condition.	fig4a	empirical	empirical	—	( A ) Regions showing a greater response when participants chose the risky (lottery) rather than the safe option, irrespective of Social or Solo condition.	high	—	—	true	slice
10	Regions showed a greater BOLD response when participants chose for both themselves and their partner rather than just for themselves (Social > Solo).	fig4b	empirical	empirical	—	( B ) Regions showing a greater response when participants chose for both themselves and their partner rather than just for themselves (Social > Solo).	high	—	—	true	slice
11	Linear mixed model coefficients indicate that precuneus and TPJ were most active when participants chose the lottery in the Social condition.	fig4c	empirical	empirical	—	Coefficients of linear mixed models (LMMs) indicate that two of these regions, precuneus and TPJ, were most active when participants chose the lottery in the Social condition.	high	—	Caption states all coefficients and differences are significantly different from 0 (see Appendix 1—table 4).	true	slice
12	Brain regions were more active during receipt of the outcomes of lotteries than of safe choices across all conditions.	fig4d	empirical	empirical	—	( D ) Brain regions more active during receipt of the outcomes of lotteries than safe choices (all conditions).	high	—	—	true	slice
13	Insula ROIs mirrored the behavioural guilt effect: voxels responded more to low lottery outcomes for the partner when these resulted from the participant's rather than the partner's choices, even when…	fig4e	empirical	empirical	—	voxels here responded more to low lottery outcomes (L) for the partner when these resulted from participant’s rather than the partner’s choices, even when responses to high outcomes were subtracted (L…	high	—	Caption states all coefficients and differences are significantly different from 0 (see Appendix 1—table 6).	true	slice
14	A mass-univariate, voxel-wise analysis showed a cluster within the left insula ROI with higher responses to low lottery outcomes for the partner when these resulted from participant rather than partne…	fig4f	empirical	control	—	A cluster of voxels within the left insula ROI showed higher responses to low lottery outcomes for the partner if these resulted from participant rather than partner choices.	high	—	Convergent voxel-wise confirmation of the ROI-based insula guilt effect in panel E, described as a compatible result.	true	slice
15	Activation in bilateral ventral striatum was explained by a computational model-based regressor coding participant rewards.	fig4g	empirical	empirical	—	( G ) Activation in bilateral ventral striatum explained by a computational model-based regressor coding participant rewards.	high	—	—	true	slice
16	A cluster in the left superior temporal sulcus region showed a higher response to partner reward prediction errors resulting from participant rather than partner choices.	fig4h	empirical	empirical	—	one cluster in the left superior temporal sulcus region showed a higher response to partner reward prediction errors resulting from participant rather than partner choices.	high	—	Restricted to brain regions sensitive to outcomes of risky choices.	true	slice
17	The left superior temporal sulcus cluster responded to model-based regressors coding participant reward prediction resulting from participant and partner choices across both sessions of the experiment…	fig4i	empirical	empirical	—	( I ) Response in this cluster to the computational-model-based regressors coding participant reward prediction resulting from participant and partner choices, for both sessions of the experiment.	tentative	—	Caption describes what is plotted (coefficients with 95% confidence intervals) rather than stating a directional result.	true	slice
18	Functional connectivity between the left anterior insula (seed) and a cluster in the right inferior frontal gyrus at the time of choice varied as a function of condition (Social vs. Solo) and choice (…	fig5	empirical	empirical	—	Changes in functional connectivity between the left anterior insula (seed) and a cluster in the right inferior frontal gyrus at the time of the choice as a function of condition (Social vs. Solo) and …	tentative	—	Caption states the analysis but not the direction or significance of the connectivity effect.	true	slice
19	Connectivity between the left superior temporal sulcus (seed) and a cluster in the left inferior frontal gyrus was highest when participants made Safe choices for themselves and Risky choices for both…	fig5s1	empirical	control	—	connectivity was highest when participants made Safe choices for themselves and Risky choices for both players (p uncorrected = 0.001, T = 4.44, Z = 4.30, 35 voxels, peak at MNI [–48 14 6]).	high	—	This cluster did not survive corrections for multiple tests; reported at an uncorrected threshold.	true	slice
20	Among the computational models fitted to momentary happiness data, the Responsibility Redux model achieved the best (lowest) AIC in both studies (Study 1 AIC –1499; Study 2 AIC –1195).	table1	assessment	methodological	—	Responsibility Redux 4 0.361 0.331 –999 –1499	tentative	—	Table lists R2, adjusted R2, BIC and AIC for five models per study; the best-fitting model is inferred from the lowest AIC values (Study 2 Responsibility Redux AIC –1195).	true	slice
21	In mixed-effects regressions on choices, the Social condition significantly increased choice of the risky option in Study 1 but not in Study 2.	app1table1	empirical	empirical	—	Condition Social 0.14* 0.03^ 0.01 0.01	tentative	—	Row values are Study 1 probit (0.14*), Study 1 linear (0.03^), Study 2 probit (0.01), Study 2 linear (0.01); column identity is read from the table header.	true	slice
22	The linear mixed model containing all three two-way interaction terms (Model 5) best explained the happiness data in both studies, and its crucial partnerHigh:participantDecided (guilt) interaction wa…	app1table2	empirical	empirical	—	In both studies, Model 5 ( Equation 9 in the Results section of the main text), which contained all three two-way interaction terms, explained the data best, so its parameters for the crucial partner…	high	—	Model 5 partnerHigh:participantDecided coefficient is 0.39* in Study 1 and 0.31 in Study 2.	true	slice
23	During the outcome phase, responses to low lottery outcomes were higher in the Social than the Partner condition in both left and right insula and lower in the right middle temporal cortex.	app1table9	empirical	empirical	—	Social 0.41* 0.18* –0.12**	high	—	Columns are InsulaL (0.41*), InsulaR (0.18) and MidTempR (–0.12*).	true	slice
24	The difference in response between low and high lottery outcomes was greater in the Social than the Partner condition in left insula, right insula, and right middle temporal cortex.	app1table10	empirical	empirical	—	Social 0.44* 0.19* 0.67***	high	—	Columns are InsulaL (0.44*), InsulaR (0.19) and MidTempR (0.67**).	true	slice
25	Participants rated their partners highly on sympathy, cooperation, honesty, openness, and sociability in both studies.	app1table11	empirical	control	—	How honest did they seem? 9.05 (1.11) 9.34 (1.10)	high	—	Manipulation check on partner perception; ratings across the five items range roughly 8.35–9.34 across Studies 1 and 2.	true	slice

This cell has more than one version on the ledger, and their bytes are kept, so the current version can be held against an earlier one — the view a prompt change or a model swap needs.

v4 → v4 comparison

Aligned by the claim text — exact wording first, then a token overlap — because two versions share no slugs. 25 matched; 0 only in v4; 0 only in v4. v4 runs/gadeke-2026-guilt-insula/caption-reader.output.v4.json · 25 claims · v4 runs/gadeke-2026-guilt-insula/caption-reader.output.json · 25 claims

Matched

Participants chose the risky option (lottery) slightly more often in the Solo condition than in the Social condition in Study 1, but not in Study 2.

empirical fig2a, fig2d

Participants chose the risky option (lottery) slightly more often in the Solo condition than in the Social condition in Study 1, but not in Study 2.

empirical fig2a, fig2d
Risk premiums did not differ between the Solo and Social conditions in either study.

control fig2b, fig2e

Risk premiums did not differ between the Solo and Social conditions in either study.

control fig2b, fig2e
Values of the risk aversion parameter ρ were broadly consistent with risk premiums but showed that participants were slightly more risk averse in the Social than in the Solo condition in Study 1 only.

empirical fig2c, fig2f

Values of the risk aversion parameter ρ were broadly consistent with risk premiums but showed that participants were slightly more risk averse in the Social than in the Solo condition in Study 1 only.

empirical fig2c, fig2f
Participant momentary happiness varied with the rewards received by the participant.

empirical fig3a, fig3e

Participant momentary happiness varied with the rewards received by the participant.

empirical fig3a, fig3e
Participant momentary happiness varied with the rewards received by the partner.

empirical fig3b, fig3f

Participant momentary happiness varied with the rewards received by the partner.

empirical fig3b, fig3f
The Responsibility Redux model, taking into account expected, previous and current rewards, reward prediction errors for both participant and partner, and decision-maker, predicted the variations in participants' momentary happiness well.

empirical fig3c, fig3g

The Responsibility Redux model, taking into account expected, previous and current rewards, reward prediction errors for both participant and partner, and decision-maker, predicted the variations in participants' momentary happiness well.

empirical fig3c, fig3g
Responsibility for low lottery outcomes for the partner decreased participant happiness more than the same outcomes following partner choices, fitting the definition of interpersonal guilt.

empirical fig3d, fig3h

Responsibility for low lottery outcomes for the partner decreased participant happiness more than the same outcomes following partner choices, fitting the definition of interpersonal guilt.

empirical fig3d, fig3h
The parameters of the Responsibility Redux model could be recovered from synthetic data generated from each participant's estimated parameters, supporting the stability of the fitted parameters.

methodological fig3s1

The parameters of the Responsibility Redux model could be recovered from synthetic data generated from each participant's estimated parameters, supporting the stability of the fitted parameters.

methodological fig3s1
A set of regions showed a greater BOLD response when participants chose the risky (lottery) rather than the safe option, irrespective of Social or Solo condition.

empirical fig4a

A set of regions showed a greater BOLD response when participants chose the risky (lottery) rather than the safe option, irrespective of Social or Solo condition.

empirical fig4a
Regions showed a greater BOLD response when participants chose for both themselves and their partner rather than just for themselves (Social > Solo).

empirical fig4b

Regions showed a greater BOLD response when participants chose for both themselves and their partner rather than just for themselves (Social > Solo).

empirical fig4b
Linear mixed model coefficients indicate that precuneus and TPJ were most active when participants chose the lottery in the Social condition.

empirical fig4c

Linear mixed model coefficients indicate that precuneus and TPJ were most active when participants chose the lottery in the Social condition.

empirical fig4c
Brain regions were more active during receipt of the outcomes of lotteries than of safe choices across all conditions.

empirical fig4d

Brain regions were more active during receipt of the outcomes of lotteries than of safe choices across all conditions.

empirical fig4d
Insula ROIs mirrored the behavioural guilt effect: voxels responded more to low lottery outcomes for the partner when these resulted from the participant's rather than the partner's choices, even when responses to high outcomes were subtracted.

empirical fig4e

Insula ROIs mirrored the behavioural guilt effect: voxels responded more to low lottery outcomes for the partner when these resulted from the participant's rather than the partner's choices, even when responses to high outcomes were subtracted.

empirical fig4e
A mass-univariate, voxel-wise analysis showed a cluster within the left insula ROI with higher responses to low lottery outcomes for the partner when these resulted from participant rather than partner choices.

control fig4f

A mass-univariate, voxel-wise analysis showed a cluster within the left insula ROI with higher responses to low lottery outcomes for the partner when these resulted from participant rather than partner choices.

control fig4f
Activation in bilateral ventral striatum was explained by a computational model-based regressor coding participant rewards.

empirical fig4g

Activation in bilateral ventral striatum was explained by a computational model-based regressor coding participant rewards.

empirical fig4g
A cluster in the left superior temporal sulcus region showed a higher response to partner reward prediction errors resulting from participant rather than partner choices.

empirical fig4h

A cluster in the left superior temporal sulcus region showed a higher response to partner reward prediction errors resulting from participant rather than partner choices.

empirical fig4h
The left superior temporal sulcus cluster responded to model-based regressors coding participant reward prediction resulting from participant and partner choices across both sessions of the experiment.

empirical fig4i

The left superior temporal sulcus cluster responded to model-based regressors coding participant reward prediction resulting from participant and partner choices across both sessions of the experiment.

empirical fig4i
Functional connectivity between the left anterior insula (seed) and a cluster in the right inferior frontal gyrus at the time of choice varied as a function of condition (Social vs. Solo) and choice (Risky or Safe).

empirical fig5

Functional connectivity between the left anterior insula (seed) and a cluster in the right inferior frontal gyrus at the time of choice varied as a function of condition (Social vs. Solo) and choice (Risky or Safe).

empirical fig5
Connectivity between the left superior temporal sulcus (seed) and a cluster in the left inferior frontal gyrus was highest when participants made Safe choices for themselves and Risky choices for both players, an opposite pattern that did not survive correction for multiple comparisons.

control fig5s1

Connectivity between the left superior temporal sulcus (seed) and a cluster in the left inferior frontal gyrus was highest when participants made Safe choices for themselves and Risky choices for both players, an opposite pattern that did not survive correction for multiple comparisons.

control fig5s1
Among the computational models fitted to momentary happiness data, the Responsibility Redux model achieved the best (lowest) AIC in both studies (Study 1 AIC –1499; Study 2 AIC –1195).

methodological table1

Among the computational models fitted to momentary happiness data, the Responsibility Redux model achieved the best (lowest) AIC in both studies (Study 1 AIC –1499; Study 2 AIC –1195).

methodological table1
In mixed-effects regressions on choices, the Social condition significantly increased choice of the risky option in Study 1 but not in Study 2.

empirical app1table1

In mixed-effects regressions on choices, the Social condition significantly increased choice of the risky option in Study 1 but not in Study 2.

empirical app1table1
The linear mixed model containing all three two-way interaction terms (Model 5) best explained the happiness data in both studies, and its crucial partnerHigh:participantDecided (guilt) interaction was significant.

empirical app1table2

The linear mixed model containing all three two-way interaction terms (Model 5) best explained the happiness data in both studies, and its crucial partnerHigh:participantDecided (guilt) interaction was significant.

empirical app1table2
During the outcome phase, responses to low lottery outcomes were higher in the Social than the Partner condition in both left and right insula and lower in the right middle temporal cortex.

empirical app1table9

During the outcome phase, responses to low lottery outcomes were higher in the Social than the Partner condition in both left and right insula and lower in the right middle temporal cortex.

empirical app1table9
The difference in response between low and high lottery outcomes was greater in the Social than the Partner condition in left insula, right insula, and right middle temporal cortex.

empirical app1table10

The difference in response between low and high lottery outcomes was greater in the Social than the Partner condition in left insula, right insula, and right middle temporal cortex.

empirical app1table10
Participants rated their partners highly on sympathy, cooperation, honesty, openness, and sociability in both studies.

control app1table11

Participants rated their partners highly on sympathy, cooperation, honesty, openness, and sociability in both studies.

control app1table11

Only in v4 dropped

Nothing — every v4 claim survived.

Only in v4 added

Nothing — v4 added no claim.

How it is defined

A model answers this layer, so the prompt is the layer. It is reproduced below from the committed file, and it is a declared input — editing it makes every run that used it stale.

extract/prompts/caption-reader.mdthe prompt it runs undernot in the repository

The declaration names this path and the repository does not have it. An input that does not exist hashes to nothing, so it cannot make a run stale — the layer is declared to depend on something it is not in fact tracking.

extract/prompts/contract/vocabulary.mdthe prompt it runs undernot in the repository

extract/prompts/contract/schema-candidate.mdthe prompt it runs undernot in the repository

Artifacts

runs/gadeke-2026-guilt-insula/caption-reader.output.json

Versions

From the run ledger. There is no changelog beside it to keep in step.

v4 · 2026-09-12 · Claude Opus 4.8 (1M context) (subagent, via --answer)

caption-reader under the current chain (#96)

cd extract && python3 -m elife_extract.cli caption-reader --paper gadeke-2026-guilt-insula --answer runs/gadeke-2026-guilt-insula/caption-reader.answer.v4.json
v3 · 2026-09-11 · claude-opus-5 (subagent, via --answer)

first run under the prompt contract (#56 item 4); Opus 5 subagents for every layer; answered through --dump-prompt/--answer, not the configured backend

cd extract && python3 -m elife_extract.cli caption-reader --paper gadeke-2026-guilt-insula --answer runs/gadeke-2026-guilt-insula/caption-reader.answer.v3.json
v2 · 2026-09-11 · claude-sonnet-5 (subagent, via --answer)

first run under the prompt contract (#56 item 4); Sonnet 5 subagents read, Opus 5 subagents reconciled, reviewed and inferred edges; answered through --dump-prompt/--answer, not the configured backend

cd extract && python3 -m elife_extract.cli caption-reader --paper gadeke-2026-guilt-insula --answer runs/gadeke-2026-guilt-insula/caption-reader.answer.v2.json
v1 · 2026-09-10 · deepseek/deepseek-chat backfilled from the artifact

backfilled from runs/manifest.json

This layer across the corpus

Across the corpus

9 not run · 1 stale·a paper links to its own cell, where this layer's output for it is rendered

Paper	State	Version	Last run	Output	Cell
A three-dimensional immunofluorescence atlas of the …	not run	—	—	—	json
Distinct representational properties of cues and con…	not run	—	—	—	json
Computational modelling identifies key determinants …	not run	—	—	—	json
Contributions of insula and superior temporal sulcus…caption-reader under the current chain (#96)	stale	v4	2026-09-12	caption-reader.output.json	json
Spatially targeted inhibitory rhythms differentially…	not run	—	—	—	json
Feedback of peripheral saccade targets to early fove…	not run	—	—	—	json
iGABASnFR2 is an improved genetically encoded protei…	not run	—	—	—	json
A deep learning pipeline for mapping in situ network…	not run	—	—	—	json
Self-association enhances early attentional selectio…	not run	—	—	—	json
Impaired excitability of fast-spiking neurons in a n…	not run	—	—	—	json

Inputs and outputs

Reads, besides its dependencies: extract/prompts/caption-reader.md · declared, and not in the repository — it hashes to nothing, so it cannot make a run stale
extract/prompts/contract/vocabulary.md · declared, and not in the repository — it hashes to nothing, so it cannot make a run stale
extract/prompts/contract/schema-candidate.md · declared, and not in the repository — it hashes to nothing, so it cannot make a run stale
Produces: runs/{paper}/caption-reader.output.json

One per paper — the table above links each one that exists.
Views: list — rendered above, over the 25 claims in the artifact
table — rendered above, over the 25 claims in the artifact
comparison — on the cell page, two versions aligned by the matcher, wherever the ledger holds more than one

Running it

The command comes from the declaration, so this text and what actually runs cannot diverge. pipeline.py run also runs the unmet dependencies first.

python3 scripts/pipeline.py run <paper> caption-reader

Underneath, that runs cd extract && python3 -m claim_graphs.cli caption-reader --paper {paper}.