arxiv: 2603.11512 · v2 · submitted 2026-03-12 · 💻 cs.HC · cs.CV

Recognition: 2 theorem links

· Lean Theorem

From Pen Strokes to Sleep States: Detecting Low-Recovery Days Using Sigma-Lognormal Handwriting Features

Chisa Tanaka , Andrew Vargo , Anna Scius-Bertrand , Andreas Fischer , Koichi Kise

Authors on Pith no claims yet

Pith reviewed 2026-05-15 12:37 UTC · model grok-4.3

classification 💻 cs.HC cs.CV

keywords handwriting analysissleep recoverysigma-lognormal modelneuromotor featuresautonomic monitoringhealth sensingwearable validationpersonalized classification

0 comments

The pith

Handwriting dynamics can reveal daily fluctuations in sleep recovery.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper tests whether routine handwriting captures within-person changes in recovery after sleep. Researchers extracted features from the Sigma-Lognormal model of pen-stroke velocity and trained personalized classifiers to flag low-recovery days, defined by the bottom quartile of four wearable-measured sleep metrics. In a 28-day field study of 13 students, the models beat the 0.25 PR-AUC baseline for heart-rate variability, lowest heart rate, average heart rate, and total sleep time. Performance held across different writing tasks and times of day, suggesting the recovery signal sits in ordinary movement dynamics rather than task-specific content.

Core claim

Sigma-lognormal parameters derived from everyday pen strokes encode neuromotor information that allows binary classification of low-recovery days, with leave-one-day-out PR-AUC significantly above the 0.25 baseline for all four sleep-related variables after correction.

What carries the argument

The Sigma-Lognormal model, which decomposes pen-stroke velocity profiles into lognormal functions to extract parameters reflecting neuromotor command generation.

If this is right

Recovery signals appear in general handwriting movements, independent of writing task or recording time.
Cardiac metrics (HRV and heart-rate measures) yield stronger detection than sleep duration alone.
Routine digital writing could serve as a passive, device-free channel for tracking autonomic recovery.
Handwriting analysis can move beyond character recognition and clinical diagnosis to everyday physiological monitoring.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same motor features might be extractable from typing or drawing on tablets for similar recovery tracking.
If the signal generalizes, note-taking apps could add daily recovery estimates without extra sensors.
Larger studies would be needed to separate recovery effects from individual writing style differences.
This approach could offer a low-cost alternative for populations that do not use wearables.

Load-bearing premise

The assumption that the extracted handwriting features specifically track sleep-related autonomic recovery instead of mood, task load, or stable personal motor habits, and that results from 13 students extend to other people.

What would settle it

A replication in which the same features yield PR-AUC no higher than 0.25 once daily mood, stress, and task demands are measured and controlled.

Figures

Figures reproduced from arXiv: 2603.11512 by Andreas Fischer, Andrew Vargo, Anna Scius-Bertrand, Chisa Tanaka, Koichi Kise.

**Figure 2.** Figure 2: Devices used for data collection mondon [7]. On our dataset, we achieved a mean signal-to-noise ratio (SNR) of 24.1 ± 4.4 dB, indicating good reconstruction quality of the Sigma-Lognormal model. From this model, we extract five basic features shown in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Figure 3: Visualization of writing speed for three participants (P1, P7, P13). Each [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Stroke order instructions for handwriting tasks [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

**Figure 5.** Figure 5: Distribution of target variables for each participant. Box plots show [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗

**Figure 6.** Figure 6: Classification performance by task (each target variable). Each point rep [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗

**Figure 7.** Figure 7: Classification performance by timing (each target variable). Each point [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗

read the original abstract

While handwriting has traditionally been studied for character recognition and disease classification, its potential to reflect day-to-day physiological fluctuations in healthy individuals remains unexplored. This study examines whether daily variations in sleep-related recovery states can be inferred from online handwriting dynamics. % We propose a personalized binary classification framework that detects low-recovery days using features derived from the Sigma-Lognormal model, which captures the neuromotor generation process of pen strokes. In a 28-day in-the-wild study involving 13 university students, handwriting was recorded three times daily, and nocturnal cardiac indicators were measured using a wearable ring. For each participant, the lowest (or highest) quartile of four sleep-related metrics -- HRV, lowest heart rate, average heart rate, and total sleep duration -- defined the positive class. Leave-One-Day-Out cross-validation showed that PR-AUC significantly exceeded the baseline (0.25) for all four variables after FDR correction, with the strongest performance observed for cardiac-related variables. Importantly, classification performance did not differ significantly across task types or recording timings, indicating that recovery-related signals are embedded in general movement dynamics. These results demonstrate that subtle within-person autonomic recovery fluctuations can be detected from everyday handwriting, opening a new direction for non-invasive, device-independent health monitoring.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Handwriting sigma-lognormal features beat baseline at flagging low-recovery days in 13 students, but unmeasured daily confounders remain the main open question.

read the letter

The core result is that sigma-lognormal parameters extracted from everyday pen strokes gave PR-AUC above the 0.25 baseline for detecting the lowest quartile of HRV, lowest heart rate, average heart rate, and sleep duration, with cardiac variables performing best. The study ran 28 days in the wild with three handwriting samples per day from 13 students and used a wearable ring for ground truth. Leave-one-day-out cross-validation on per-person quartile labels produced the reported lift, and performance held across task types and recording times. That last check is useful because it suggests the signal is not just tied to a specific writing prompt or time of day. The approach is new in applying these neuromotor features to day-to-day autonomic recovery rather than disease detection or recognition tasks. The authors also kept the modeling simple and personalized, which fits the small N. The main limitation is the absence of any measurement or statistical control for mood, perceived stress, caffeine, or same-day physical activity. Those variables are known to move both sleep metrics and fine-motor output, and with only about seven positive days per participant the risk of picking up correlated but non-causal patterns is real. The sample is also narrow—university students—so generalizability is unclear. The citation pattern looks standard for the subfield and does not appear to over-claim prior work. This paper is for researchers working on passive health sensing or HCI applications to wellness tracking. A reader who wants to see whether handwriting can serve as a cheap, device-free recovery signal will find the setup and the positive PR-AUC numbers worth examining, even if the causal story needs more work. I would send it to peer review. The empirical result is concrete enough to justify referee time, and the confounder issue is fixable with additional variables or larger controls rather than a fatal flaw.

Referee Report

3 major / 2 minor

Summary. The manuscript claims that subtle within-person variations in sleep-related autonomic recovery can be detected from everyday online handwriting using features extracted from the Sigma-Lognormal model of pen-stroke generation. In a 28-day in-the-wild study with 13 university students, handwriting was sampled three times daily while nocturnal HRV, lowest/average heart rate, and sleep duration were recorded via wearable ring; per-participant quartile thresholds defined low-recovery (positive) days, and leave-one-day-out cross-validation yielded PR-AUC values significantly above the 0.25 baseline for all four metrics after FDR correction, with no significant differences across task types or recording times.

Significance. If the central result holds after addressing confounder controls and sample-size limitations, the work would open a promising direction for non-invasive, device-independent daily health monitoring by showing that neuromotor features of routine handwriting encode recovery fluctuations indexed by cardiac and sleep metrics. The in-the-wild design, personalized modeling, and task-independence findings are particular strengths that enhance potential translational value.

major comments (3)

[Methods] Methods section (data collection and labeling): the study records no daily measures or statistical controls for mood, perceived stress, caffeine intake, or physical activity—variables known to covary with both wearable recovery metrics and fine-motor output. With only ~7 positive days per participant under quartile labeling, any unmeasured confounder could inflate PR-AUC without the Sigma-Lognormal features specifically encoding autonomic signals.
[Results] Results section (cross-validation performance): N=13 with per-person models and leave-one-day-out CV on ~28 days (~7 positive instances) leaves the reported PR-AUC gains vulnerable to overfitting and limited generalizability; the manuscript should report effect sizes, confidence intervals, and external validation on an independent cohort to substantiate the claim that performance exceeds baseline for cardiac variables.
[Discussion] Discussion section (feature interpretation): the claim that Sigma-Lognormal parameters (velocity profiles, lognormal parameters) specifically track recovery-related neuromotor processes lacks supporting ablation or feature-importance analyses comparing them to simpler kinematic descriptors; without this, the results could reflect generic motor variability rather than the targeted physiological signal.

minor comments (2)

[Abstract] Abstract: explicitly state the total number of handwriting samples collected per participant and the precise quartile threshold application for each of the four metrics.
[Methods] Figure captions and methods: clarify the exact Sigma-Lognormal parameter set retained as features and any preprocessing steps applied to stroke velocity profiles.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback. We address each of the major comments below, indicating the revisions we plan to make to strengthen the manuscript.

read point-by-point responses

Referee: [Methods] Methods section (data collection and labeling): the study records no daily measures or statistical controls for mood, perceived stress, caffeine intake, or physical activity—variables known to covary with both wearable recovery metrics and fine-motor output. With only ~7 positive days per participant under quartile labeling, any unmeasured confounder could inflate PR-AUC without the Sigma-Lognormal features specifically encoding autonomic signals.

Authors: We acknowledge this important limitation. The study was designed as an in-the-wild investigation without daily self-reports on mood, stress, caffeine, or activity levels to minimize participant burden. While this leaves room for potential confounders, the use of personalized models and the finding of task-independent performance suggest that the detected signals are not solely due to these factors. In the revised manuscript, we will add a dedicated paragraph in the Discussion section addressing potential confounders and their implications for interpretation. revision: partial
Referee: [Results] Results section (cross-validation performance): N=13 with per-person models and leave-one-day-out CV on ~28 days (~7 positive instances) leaves the reported PR-AUC gains vulnerable to overfitting and limited generalizability; the manuscript should report effect sizes, confidence intervals, and external validation on an independent cohort to substantiate the claim that performance exceeds baseline for cardiac variables.

Authors: We agree that the small sample size and cross-validation setup limit strong claims about generalizability. We will revise the Results section to include effect sizes (e.g., Cohen's d or similar for PR-AUC differences) and bootstrap-derived confidence intervals for the PR-AUC values. Regarding external validation, we do not have an independent cohort available in this study and will explicitly state this as a limitation, recommending it for future work. revision: partial
Referee: [Discussion] Discussion section (feature interpretation): the claim that Sigma-Lognormal parameters (velocity profiles, lognormal parameters) specifically track recovery-related neuromotor processes lacks supporting ablation or feature-importance analyses comparing them to simpler kinematic descriptors; without this, the results could reflect generic motor variability rather than the targeted physiological signal.

Authors: The Sigma-Lognormal model provides a theoretically grounded decomposition of handwriting into parameters that reflect neuromotor control, which we argue is more specific than generic kinematics. To address the concern, we will conduct and report a feature importance analysis (e.g., using permutation importance or SHAP values) on the Sigma-Lognormal parameters in the revised manuscript to demonstrate their contribution relative to simpler features like stroke duration or velocity averages. This will help substantiate the specificity of the approach. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation uses independent ground truth

full rationale

The paper extracts Sigma-Lognormal parameters from handwriting strokes as input features and trains a classifier to predict low-recovery labels derived from separate wearable ring measurements (HRV, heart rate, sleep duration). These labels are obtained independently of the handwriting data and are not used to define or fit the Sigma-Lognormal features themselves. Leave-one-day-out cross-validation evaluates generalization without reducing the prediction to a re-fit of the same inputs. No equations or steps equate the claimed detection to a self-defined quantity, and the model is presented as a pre-existing neuromotor framework rather than an ansatz derived from the current recovery data. The central empirical result therefore rests on correlation between distinct measurement streams rather than constructional equivalence.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the established Sigma-Lognormal model for stroke generation and the validity of quartile-based labeling from wearable data; no new entities are introduced.

free parameters (1)

quartile threshold
Lowest (or highest) quartile of each sleep metric defines the positive class; value 0.25 is chosen by convention.

axioms (1)

domain assumption Sigma-Lognormal model accurately captures the neuromotor generation process of pen strokes
Invoked to justify feature extraction from handwriting dynamics.

pith-pipeline@v0.9.0 · 5542 in / 1243 out tokens · 42881 ms · 2026-05-15T12:37:02.359677+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

velocity profile ... sum of lognormal distributions ... features D, t0, nblog, SNR ... RandomForest ... PR-AUC on HRV/Lowest HR quartiles
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

28-day in-the-wild study ... leave-one-day-out cross-validation ... no task/timing effect after FDR

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

14 extracted references · 14 canonical work pages

[1]

PhysiologicalReports13(16),e70527(2025).https://doi.org/10.14814/phy2.70527

Dial,M.,Hollander,M.,Vatne,E.,Emerson,A.,Edwards,N.,Hagen,J.:Validation of nocturnal resting heart rate and heart rate variability in consumer wearables. PhysiologicalReports13(16),e70527(2025).https://doi.org/10.14814/phy2.70527

work page doi:10.14814/phy2.70527 2025
[2]

Faci, N., Nguyen, H., Laniel, P., Gauthier, B., Beauchamp, M., Nakagawa, M., Plamondon, R.: Classifying the Kinematics of Fast Pen Strokes in Children with ADHD using Different Machine Learning Models, chap. 5, pp. 117–142. World Scientific (2021). https://doi.org/10.1142/9789811226830_0005

work page doi:10.1142/9789811226830_0005 2021
[3]

Journal of Sleep Research18(2), 264–271 (2009)

Jasper, I., Häußler, A., Marquardt, C., Hermsdörfer, J.: Circadian rhythm in handwriting. Journal of Sleep Research18(2), 264–271 (2009). https://doi.org/10.1111/j.1365-2869.2008.00706.x

work page doi:10.1111/j.1365-2869.2008.00706.x 2009
[4]

Psychiatry Investigation 15(3), 235–245 (2018)

Kim, H.G., Cheon, E.J., Bai, D.S., Lee, Y., Koo, B.H.: Stress and heart rate variability: A meta-analysis and review of the literature. Psychiatry Investigation 15(3), 235–245 (2018). https://doi.org/10.30773/pi.2017.08.17

work page doi:10.30773/pi.2017.08.17 2018
[5]

IEEE Transactions on Pattern Analysis and Machine Intelligence44(10), 6472– 6485 (2022)

Lai, S., Jin, L., Zhu, Y., Li, Z., Lin, L.: SynSig2Vec: Forgery-free learning of dy- namic signature representations by sigma lognormal-based synthesis and 1D CNN. IEEE Transactions on Pattern Analysis and Machine Intelligence44(10), 6472– 6485 (2022). https://doi.org/10.1109/TPAMI.2021.3087619

work page doi:10.1109/tpami.2021.3087619 2022
[6]

In: Proceedings of the 2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp ’15)

Lazar, A., Koehler, C., Tanenbaum, J., Nguyen, D.H.: Why we use and abandon smart devices. In: Proceedings of the 2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing (UbiComp ’15). pp. 635–646. ACM (2015). https://doi.org/10.1145/2750858.2804288

work page doi:10.1145/2750858.2804288 2015
[7]

Pattern Recognition42(12), 3324–3337 (2009)

O’Reilly, C., Plamondon, R.: Development of a sigma–lognormal represen- tation for on-line signatures. Pattern Recognition42(12), 3324–3337 (2009). https://doi.org/10.1016/j.patcog.2008.10.017

work page doi:10.1016/j.patcog.2008.10.017 2009
[8]

In: New Trends in Image Analysis and Processing – ICIAP 2015 Workshops, Lecture Notes in Computer Science, vol

Pirlo, G., Diaz, M., Ferrer, M., Impedovo, D., Occhionero, F., Zurlo, U.: Early diagnosis of neurodegenerative diseases by handwritten signature analysis. In: New Trends in Image Analysis and Processing – ICIAP 2015 Workshops, Lecture Notes in Computer Science, vol. 9281, pp. 290–297. Springer (2015). https://doi.org/10.1007/978-3-319-23222-5_36

work page doi:10.1007/978-3-319-23222-5_36 2015
[9]

Frontiers in Psychology4, 945 (2013)

Plamondon, R., O’Reilly, C., Rémi, C., Duval, T.: The lognormal handwriter: learning, performing, and declining. Frontiers in Psychology4, 945 (2013). https://doi.org/10.3389/fpsyg.2013.00945

work page doi:10.3389/fpsyg.2013.00945 2013
[10]

Plamondon, R.: The Lognormality Principle: A Personalized Survey, chap. 1, pp. 1–39. World Scientific (2021). https://doi.org/10.1142/9789811226830_0001

work page doi:10.1142/9789811226830_0001 2021
[11]

Frontiers in Artificial Intelligence5, 787179 (2022)

Prange, A., Sonntag, D.: Modeling users’ cognitive performance using dig- ital pen features. Frontiers in Artificial Intelligence5, 787179 (2022). https://doi.org/10.3389/frai.2022.787179

work page doi:10.3389/frai.2022.787179 2022
[12]

Frontiers in Physiology4, 294 (2013)

Tobaldini, E., Nobili, L., Strada, S., Casali, K., Braghiroli, A., Montano, N.: Heart rate variability in normal and pathological sleep. Frontiers in Physiology4, 294 (2013). https://doi.org/10.3389/fphys.2013.00294

work page doi:10.3389/fphys.2013.00294 2013
[13]

Neuroscience and Behavioral Physi- ology54, 961–966 (2024)

Venevtseva, Y., Melnikov, A., Nesterova, S.: Sleep and fine motor skills: the influ- ence of sex and the level of physical activity. Neuroscience and Behavioral Physi- ology54, 961–966 (2024). https://doi.org/10.1007/s11055-024-01697-5

work page doi:10.1007/s11055-024-01697-5 2024
[14]

Behavioral Sleep Medicine 17(2), 124–136 (2019)

deZambotti,M.,Rosas,L.,Colrain,I.,Baker,F.:Thesleepofthering:Comparison of the ¯Oura sleep tracker against polysomnography. Behavioral Sleep Medicine 17(2), 124–136 (2019). https://doi.org/10.1080/15402002.2017.1300587

work page doi:10.1080/15402002.2017.1300587 2019