YOUR QUESTION
Does a larger pupil mean better working memory?
Larger is not a general rule for better memory.
Four papers examine different pupil measures and tasks. Resting size, task-evoked dilation, overload and remembered brightness must be read separately; associations in one condition do not supply a personal score or causal benefit.
Resting size
A resting baseline did not track working-memory capacity; a reliable baseline is not automatically predictive.
Task-evoked response
Positive associations occur in selected loads, trials and sessions; the measurement window matters.
Overload
More items can lead to a plateau and decline, rather than continued dilation.
Remembered content
In the PNAS task, dark and bright memory items reverse the direction of association.
4 distinct publications · different measures and tasks · no pooled participant count or clinical score
A conditional link with memory precision
Across the three experiments in this paper, better recall precision was associated with larger pupils for dark memory items, but smaller pupils for bright items.
- Analysis sample
- 68 participants · Healthy young volunteers; three distinct experiment samples
- Design
- Within-person correlations compared by brightness, with experiment as a between-person factor
These are within-person associations. They do not show that enlarging a pupil improves memory, or provide an individual prediction threshold.
Inspect original sourceStudy conditions & reported results
- Sample roles
- 68 = 22 + 23 + 23, derived from distinct final samples. This total applies to the cross-experiment correlation comparison, not every endpoint in the paper.
- Time window
- Within-trial pupil measurements before the recall probe; the early-probe delay in Experiment 3
- F(1, 65) = 9.69, P = 0.003, η2p = 0.113Within-person error–pupil correlations compared by brightness across three experimentsAuthor-reported; not recomputed from participant data. Confidence interval not reported here.
The paper does not list individual correlation coefficients or their confidence intervals here. The mechanism and causal direction remain undetermined.
Finite reviewable record with limits. Automatic checking does not certify scientific truth.
A pupil difference is not a demonstrated memory gain
In Experiment 2, dark versus bright cued memory items differed in mean pupil size. The brightness main effect on recall precision was not detected in the same 23 participants.
- Analysis sample
- 23 participants · Healthy volunteers aged 21–29 in Experiment 2
- Design
- Auditory spatial retrocue; grating orientation recall; task-irrelevant brightness
The two tests address different outcomes. Non-detection is not proof of no effect or equivalence, and the cue-validity advantage is a separate factor.
Inspect original sourceStudy conditions & reported results
- Sample roles
- 25 recruited; 2 excluded for eye-tracking problems; 23 in the final analysis. Do not add participants from the other experiments to this endpoint.
- Time window
- From 500 ms after cue onset to the recall probe, during a blank retention interval
- t(22) = 2.1, P = 0.048Experiment 2: mean pupil size for validly cued dark versus bright memory itemsAuthor-reported; not recomputed from participant data. Confidence interval not reported here.
- F(1, 22) = 0.13, P = 0.72, η2p = 0.009Experiment 2: main effect of memory-item brightness on recall precisionAuthor-reported; not recomputed from participant data. Confidence interval not reported here.
The mean-pupil test has P = 0.048. Its effect size and confidence interval are not reported here; the temporal analysis is a separate test.
Finite reviewable record with limits. Automatic checking does not certify scientific truth.
Resting size did not track working-memory capacity
In 102 participants, resting pupil size was not associated with letter–number sequencing performance (r = 0.05, p = 0.65). BF01 = 7.26 favored the null under the authors’ model.
- Analysis sample
- 102 participants · Mostly young Groningen university students
- Design
- Between-person association; one working-memory task; arbitrary pupil units
This limits a baseline-size rule; it does not refute task-evoked or brightness-dependent effects.
Inspect original sourceStudy conditions & reported results
- Sample roles
- 102 completed the relevant analysis; 224 is the combined sample for other paper analyses.
- Time window
- Two-minute resting recordings before and after tasks
- r = 0.05, p = 0.65Resting baseline versus letter–number sequencingAuthor-reported; not recomputed from participant data. Confidence interval not reported here.
- BF01 = 7.26Authors’ Bayesian correlation modelAuthor-reported; not recomputed from participant data. Confidence interval not reported here.
Null-model support depends on the model, task and sample; no universal absence or clinical threshold follows.
Finite reviewable record with limits. Automatic checking does not certify scientific truth.
A positive link at one load and time window
For nine-digit sequences, pupil dilation during the three-second retention interval correlated with recall (r(71) = .397).
- Analysis sample
- 73 participants · Young adult volunteers performing auditory digit recall
- Design
- Auditory sequences of 5, 9 or 13 digits; baseline-corrected pupil response
The coefficient is specific to this endpoint, not all loads or a causal memory gain.
Inspect original sourceStudy conditions & reported results
- Sample roles
- 86 completed the task; two technical and eleven artifact exclusions left 73 pupil analyses.
- Time window
- Three seconds after the final digit, before recall
- r(71) = .397Nine-digit retention dilation versus recallAuthor-reported; not recomputed from participant data. Confidence interval not reported here.
Do not reuse this coefficient for five- or thirteen-digit endpoints; a corresponding P value or interval is not supplied here.
Finite reviewable record with limits. Automatic checking does not certify scientific truth.
Greater load did not produce indefinite dilation
The mean response rose during early encoding, plateaued around digits 7–10, and declined at digits 11–13.
- Analysis sample
- 73 participants · The same pupil-analysis cohort as the retention card
- Design
- Memory compared with passive listening
Load, successful recall and individual capacity are different quantities; this trajectory is not a personal capacity test.
Inspect original sourceStudy conditions & reported results
- Sample roles
- 73 pupil analyses; the two cards share one study and are not independent samples.
- Time window
- Sequential encoding under the longest digit load
A group trajectory, not a general pupil-to-performance mapping.
Finite reviewable record with limits. Automatic checking does not certify scientific truth.
Some phasic effects persisted; baseline stability was insufficient
Phasic responses related to performance in some trial types and sessions. Stable baseline measures did not predict performance in these n-back variants.
- Analysis sample
- Varies by experiment and session · Young adults in two n-back experiments
- Design
- Short and long n-back variants; trial-specific pupil measures
Trial type, task duration and session matter. Reliability does not establish prediction or clinical validity.
Inspect original sourceStudy conditions & reported results
- Sample roles
- Experiment 1: sessions N64 and N63, paired N56. Experiment 2: sessions N42 and N42, paired N40. Repeated sessions are not added as independent people.
- Time window
- Pre-trial/pre-experimental baseline and stimulus-evoked response; two sessions weeks apart
Qualitative findings only. Disputed baseline coefficients and formulas are excluded; no raw-data replication is claimed.
Finite reviewable record with limits. Automatic checking does not certify scientific truth.
Direct opposition to the brightness mechanism
Not foundThe added papers limit a universal “larger is better” rule through different measures and tasks. They do not directly replicate or refute the PNAS remembered-brightness mechanism. No direct opposing replication of that mechanism is included; this is a targeted selection, not a systematic review.
Null or non-monotonic findings in a different measure limit generalization; they are not opposing replications of the brightness mechanism. Non-detection does not establish equivalence.
When this answer applies
Which measure?
Resting size, pre-trial baseline, baseline variability and evoked change are different measurements.
Which task and sample?
Digit recall, n-back and visual-memory precision differ. Endpoint samples and repeated sessions cannot be pooled as independent people.
What cannot be inferred?
No clinical prediction, universal memory score or causal improvement follows. Four publications are not four replications of one mechanism.
Trace the answer to its source
6 records from 4 distinct papers. Cards, experiments and repeated sessions are not counted as independent publications.
[1]Modulation of the pupillary response by the content of visual working memory.Proceedings of the National Academy of Sciences of the United States of America · 2019
CC BY 4.0 · Article licence verified in the frozen source. Summaries and Chinese explanations are assistant-written. English excerpts are unchanged.
Snapshot retrieved: 2026-10-06. Source-specific limits and checking method appear below.
Exact excerpts for this answer
Brightness changes the direction of association
In other words, smaller errors (better memory precision) were associated with larger pupil sizes for dark items (negative relation) and with smaller pupil sizes for bright items (positive relation).
Where in the source?
Discussion
/article/body/sec[2]/p[3]Frozen text character span: [28106, 28304) · Unicode code points
MED_31636213_B051_0_528
Comparison of within-person correlations
F(1, 65) = 9.69, P = 0.003, η2p = 0.113
Where in the source?
Results / Memory Error Is Predicted by the Degree of Pupillary Modulation.
/article/body/sec[1]/sec[4]/p[2]Frozen text character span: [25530, 25569) · Unicode code points
MED_31636213_B047_0_606
Experiment 2: mean pupil difference
t(22) = 2.1, P = 0.048]. As in Exp. 1, darker items elicited a larger mean pupil size (Fig. 2E).
Where in the source?
Results / Exp. 2: Pupillary Response According to Task-Irrelevant Brightness of Attended Stimuli. / Pupil traces.
/article/body/sec[1]/sec[2]/sec[2]/p[4]Frozen text character span: [17515, 17611) · Unicode code points
MED_31636213_B031_0_407
Experiment 2: recall precision test
brightness of the probed item (dark vs. bright) (see SI Appendix, Fig. S1 for mixture modeling of response error). Precision measures showed reliable advantages of spatial relative to neutral retrocues (Fig. 2B) [a main effect of cue validity F(1, 22) = 7.8, P = 0.011, η2p = 0.26]. There was no effect of side [F(1, 22) = 1.7, P = 0.2, η2p = 0.06] or brightness of the probed item [F(1, 22) = 0.13, P = 0.72, η2p = 0.009]
Where in the source?
Results / Exp. 2: Pupillary Response According to Task-Irrelevant Brightness of Attended Stimuli. / Behavioral measures.
/article/body/sec[1]/sec[2]/sec[1]/p[1]Frozen text character span: [14754, 15176) · Unicode code points
MED_31636213_B025_0_770
A functional role remains untested
However, whether the pupillary response plays an active functional role during the retention of items in working memory remains to be tested.
Where in the source?
Discussion
/article/body/sec[2]/p[6]Frozen text character span: [30864, 31005) · Unicode code points
MED_31636213_B054_0_861
Check sample counts and measurement conditions
Distinct participants across experiments
Participants were different for each experiment
Materials and Methods · [33603, 33650)
/article/body/sec[3]/p[1]Experiment 1: final sample
leaving 22 volunteers
Materials and Methods / Experiment 1. / Participants. · [34258, 34279)
/article/body/sec[3]/sec[1]/sec[1]/p[1]Experiment 2: final sample
leaving 23 volunteers
Materials and Methods / Experiment 2. / Participants. · [40073, 40094)
/article/body/sec[3]/sec[2]/sec[1]/p[1]Experiment 3: final sample
leaving 23 volunteers
Materials and Methods / Experiment 3. / Participants. · [41608, 41629)
/article/body/sec[3]/sec[3]/sec[1]/p[1]Source version & checking method
Assistant reading and deterministic checks of licence, original locations, reported statistics and endpoint-specific samples. No independent scientific certification or clinical validation.
Original frozen XML retained. Online corrections, retractions and external supplements were not rechecked for this paper.
Raw XML SHA-256: 5ecc72423d7d23928b432a54415c4748aa5556147810a486aba40b61b02269a0
Frozen text SHA-256: 7e54153b5dbe3cbb9cbd3c9d9e9dc40745d4e870c3d4053954ba28eaf56fbeb7
PMID 31636213 · PMCID PMC6842592
[2]Baseline Pupil Size Seems Unrelated to Fluid Intelligence, Working Memory Capacity, and Attentional ControlJournal of Cognition · 2024 · Published article
CC BY 4.0 · Article-level CC BY 4.0 verified in publisher rights metadata. Summaries and translations are assistant-written; short English excerpts are unchanged.
Snapshot retrieved: 2026-10-07. Source-specific limits and checking method appear below.
Exact excerpts for this answer
Resting baseline and working-memory performance
r = 0.05, p = 0.65
Where in the source?
Results / Baseline Pupil Size, Working Memory Capacity and Attentional Control
Published HTML paragraph
#xml-article · p[62] (zero-based)Normalized paragraph character span: [132, 150)
Paragraph SHA-256: a2484e2646959e06f2107f241e2f9d08ee505e1bdeb0f9465603932c3078659a
baseline-correlation
Bayesian comparison under the authors’ model
BF01 = 7.26
Where in the source?
Results / Baseline Pupil Size, Working Memory Capacity and Attentional Control
Published HTML paragraph
#xml-article · p[62] (zero-based)Normalized paragraph character span: [287, 298)
Paragraph SHA-256: a2484e2646959e06f2107f241e2f9d08ee505e1bdeb0f9465603932c3078659a
baseline-bayes
Check sample counts and measurement conditions
Working-memory analysis sample
102 participants
Results / Baseline Pupil Size, Working Memory Capacity and Attentional Control · [157, 173)
#xml-article · p[61]Measurement units for the relevant cohort
arbitrary units
Methods / Groningen / Pupil recording · [323, 338)
#xml-article · p[18]Source version & checking method
Assistant reading and deterministic checks of licence, original locations, reported statistics and endpoint-specific samples. No independent scientific certification or clinical validation.
Published results checked against the public page; participant data and supplements were not downloaded or recomputed.
Publisher HTML SHA-256: 6af43c2dc54a99945f3983955db20287ba21087bc8dc0d43be2498dda96a6fd7
[3]EEG and pupillometric signatures of working memory overloadPsychophysiology · 2023 · Published article
CC BY 4.0 · Publisher-supplied Crossref metadata assigns CC BY 4.0 to the version of record. Summaries and translations are assistant-written; short English excerpts are unchanged.
Snapshot retrieved: 2026-10-07. Source-specific limits and checking method appear below.
Exact excerpts for this answer
Plateau at intermediate load
plateau from 7 to 10
Where in the source?
Discussion / 4.1 Pupil size temporal dynamics
Public publisher text; direct browser fetch was blocked by human verification
Local publisher-text snapshot line: 391
Normalized paragraph character span: [317, 337)
Paragraph SHA-256: 963242ed3b7df0f76b8030075c5f66e48fd3c7c1bdb12f944c8febc02403592f
overload-plateau
Decline at the highest load
decline from 11 to 13
Where in the source?
Discussion / 4.1 Pupil size temporal dynamics
Public publisher text; direct browser fetch was blocked by human verification
Local publisher-text snapshot line: 391
Normalized paragraph character span: [350, 371)
Paragraph SHA-256: 963242ed3b7df0f76b8030075c5f66e48fd3c7c1bdb12f944c8febc02403592f
overload-decline
Nine-digit retention: performance association
r(71) = .397
Where in the source?
Results / 3.4 Individual differences in physiological functioning and recall accuracy
Public publisher text; direct browser fetch was blocked by human verification
Local publisher-text snapshot line: 359
Normalized paragraph character span: [335, 347)
Paragraph SHA-256: 2c355a036f710cb27eb76517d8af029e71b0ced87de984dc60d57770be582db6
retention-correlation
Check sample counts and measurement conditions
Endpoint-specific pupil analysis sample
73 participants
Method / 2.4 Pupillometry · [214, 229)
Publisher-text snapshot L248Source version & checking method
Assistant reading and deterministic checks of licence, original locations, reported statistics and endpoint-specific samples. No independent scientific certification or clinical validation.
Direct publisher access requested human verification; results were checked in publicly readable publisher-page text. The article licence was checked independently in Crossref. No supplements or participant data were downloaded.
Public publisher-text snapshot SHA-256: 284281f1e6b3c1841d612477b9f0c6c516d69639f049b6101517e44bde489261
[4]Investigating the test–retest reliability and behavioral correlates of pupil responses in the n-back taskScientific Reports · 2026 · Version of record · 3 June 2026
CC BY 4.0 · Article-level CC BY 4.0 verified in Rights and permissions. Summaries and translations are assistant-written; short English excerpts are unchanged.
Snapshot retrieved: 2026-10-07. Source-specific limits and checking method appear below.
Exact excerpts for this answer
Phasic association depends on trial type
hit-related TEPRs
Where in the source?
Interim discussion
Published HTML paragraph
#Sec17-content · p[1] (zero-based)Normalized paragraph character span: [236, 253)
Paragraph SHA-256: 2b84e050d037a247dd48b2da4c751ea4c63d37defcbe679db2ea70daec5a1966
retest-phasic
Reliability does not establish performance prediction
they did not predict task performance in either of our experiments
Where in the source?
General discussion / Implications of our findings regarding the use of pupillometry as an individual difference measure
Published HTML paragraph
#Sec30-content · p[10] (zero-based)Normalized paragraph character span: [159, 225)
Paragraph SHA-256: cba5d56209a4bf07d369a1a2e4ddf1c2891ec1049328748af09afad2b00bc3cb
retest-baseline
Check sample counts and measurement conditions
Experiment 1: first-session sample
N = 64 for session 1
Experiment 1 / Methods / Participants · [82, 102)
#Sec5-content · p[2]Experiment 2: first-session sample
N = 42 for session 1
Experiment 2 / Methods / Participants · [48, 68)
#Sec18-content · p[2]Source version & checking method
Assistant reading and deterministic checks of licence, original locations, reported statistics and endpoint-specific samples. No independent scientific certification or clinical validation.
The June 3 version is used. Inconsistent baseline-variability coefficients and the ambiguous false-alarm formula are excluded. No supplementary statistics ZIP or participant-data recomputation was used.
Publisher HTML SHA-256: 5b31b126eab5a144fc26d9fa37ae12ac1ca037a6d177bfc38b271de8383bc6e1
Use the graph as a secondary view
The existing explorer uses 70 records from 24 sources. This demo uses a separate collection of 6 records from 4 papers; these records have not been added to the explorer or its scientific release.
Open the existing explorer ↗Local research prototype · saved answer, not live model output · not for personal assessment or diagnosis
mentalkg-answer-prototype-20261007.2