VeriWright · Process & Academic Integrity
Reading the rhythm of thought.
A finished text hides everything that made it. VeriWright uses a digital pen to record the writing process — timing, pauses, hesitation, pen movement — and reads it as what it is: the trace of a cognitive process. Where the writer planned, where language came easily, where it did not, where they went back. In the age of generative AI, that trace is also the honest answer to “how was it made?”
Live demonstration · synthetic data
Watch the digital pen write.
A student writes a sentence by hand with the digital pen — drawn stroke by stroke. The finished text looks ordinary, but replaying the process reveals a planning pause, a moment of contemplation, and a revision. Synthetic illustration; switch the language below.
How it works
Pen data → process replay → teacher review.
Capture
A digital pen records handwriting with timing, pauses, and pen movement while the student works on paper.
Replay
The session is reconstructed so the writing process — not just the final text — can be reviewed.
Review
The teacher interprets the process with context, using it as one input among many.
What it looks at
Kinematic signals of the writing process.
The pen measures movement. The value is what movement stands in for: writing is thinking made physical, and the places where it slows, lifts, hesitates or doubles back are where the cognitive work happened.
Rhythm of thought
Speed and acceleration across a session — fast writing can suggest fluency; slow writing, processing.
Marks of contemplation
Pauses and in-air pen movement — the “thinking time” while a writer searches for a word or structures an idea.
Process analytics
Planning, translating, revision, and metacognition inferred from how a text is built up over time — the four processes the dashboard below lays on a timeline.
Supportive diagnostics
Objective, quantitative indicators (e.g., grip-control stability) that a specialist can consider.
VeriWright is a support tool, not a medical device, and does not provide a diagnosis.
Teacher dashboard · synthetic data
The writing process, read as a thinking process.
Writing research describes composing as a cycle: planning what to say, translating it into language, monitoring what has landed on the page, and revising when the two do not match. Those processes are invisible. The pen is not. VeriWright times every pause, hover and retrace, and lays the most likely process on a timeline — with the evidence attached, so a teacher can question it.
The session as a cognitive process
Which process the pen record points to, minute by minute — and what it appeared to cost.
Table view
| Minute | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Cognitive load | 48 | 55 | 63 | 68 | 66 | 61 | 70 | 82 | 88 | 79 | 66 | 58 | 54 | 57 |
| Sustained attention | 62 | 71 | 78 | 84 | 86 | 83 | 74 | 66 | 58 | 64 | 75 | 81 | 84 | 79 |
| Process | Planning | Translating | Monitoring | Translating | Revising | Translating | Monitoring | |||||||
Process row is rounded to whole minutes; the chart uses the exact segment boundaries.
Share of the session by process
The same 14 minutes, totalled.
Underneath: the pen was writing for 54% of the session, paused for 23%, and moving above the paper for 15%. Those raw shares are what the four processes are inferred from.
Worth a look
Prompts for the teacher — never conclusions.
- Warning · a long search at minute 8. Still translating, but load ran above this student’s own band for over a minute. Was it the idea that was hard, or the word for it?
- Attend · monitoring turned into revising once. The pen went back over the text at minute 10 and a phrase was rewritten — worth asking what the student noticed.
- Good · planning came before writing. The first 1.5 minutes were spent planning rather than starting immediately — a pattern worth naming out loud to the student.
Turma 7ºB — cognitive load on the same task
One cell per student, this session. Outlined cells sit outside their own usual band.
From pen evidence to cognitive process
Every label on the timeline is an inference with a visible reason. This is the whole of it — there is nothing else behind the screen.
- A long pause, pen lifted, before the first wordPlanningThe writer is organising the whole before committing to a sentence.
- Bursts of writing separated by short pauses at clause boundariesTranslatingThe plan is being turned into language; the pauses fall where language is chunked.
- A long pause inside a phrase, pen hovering above the paperTranslating, under strainA word or a construction is being searched for — the moment vocabulary support would pay off.
- The pen travelling back over finished text without writingMonitoringRereading to check what is on the page against what was intended.
- Crossing out and rewritingRevisingMonitoring found a mismatch and the writer acted on it — the visible end of an invisible check.
How each indicator is derived
Cognitive load
Length and placement of pauses, time with the pen in the air, and how much writing speed varies inside a single sentence.
Sustained attention
Continuity of on-task writing: how long the pen keeps producing text between interruptions, and how evenly those interruptions are spread.
Grip-pressure variability
Variation in pen pressure and small-amplitude tremor across the session, compared with the student’s own earlier sessions.
Writing fluency
Consistency of stroke velocity and letter formation — steady movement versus repeated stopping and restarting mid-word.
Research foundation · Germany
Built on published digital-pen science.
Reconstructing writing from pen kinematics — and reading cognitive state from low-level pen signals — is an established research line, not something we invented in isolation. VeriWright is developed on that base, in collaboration with the German digital-pen research community: an embedded-AI handwriting tutor at KIT, the reference sensor dataset at Fraunhofer IIS, and pen-signal work on cognitive state and writer authentication at DFKI.
Handwriting tutor on embedded AI
Karlsruhe Institute of Technology · Institute for Information Processing Technologies (ITIV)
Project · KIHT
Kaligo-based Intelligent Handwriting Teacher — a Franco-German joint project building an intelligent handwriting-learning device around a sensor pen that writes on ordinary paper. Inertial sensors reconstruct the writing trajectory with no tablet and no special surface; ITIV’s part is making those AI algorithms run on the pen’s embedded hardware.
The sensor signal, benchmarked
Fraunhofer Institute for Integrated Circuits (IIS)
Dataset · OnHW
Online Handwriting Recognition from Sensor-Enhanced Pens — raw time-series from a multi-sensor digital pen released as a public benchmark. Triaxial acceleration, gyroscope, magnetometer and force, sampled at 100 Hz; the OnHW-chars set alone holds 31,275 letters from 119 writers.
Cognitive state & authorship
German Research Center for Artificial Intelligence · Interactive Machine Learning Lab
Papers · Digital pens in education
Two lines meet here. One asks whether low-level pen signals can predict task difficulty and performance in primary-school children. The other — Detection of contract cheating in pen-and-paper exams through the analysis of handwriting style — separates writing style from content in a neural architecture to confirm, passively and continuously, that the person writing is the person enrolled.
Responsible AI & inclusion
Designed to support teachers and protect students.
Teacher decides
No automated grading, discipline, or placement. The teacher is always the final decision-maker.
Rights & review
Students and families can request explanation, correction, appeal, and human review.
Data minimisation
Minimal collection, limited retention, deletion by default. No re-use for unrelated model training, advertising, or sale.
Equity & access
Equivalent alternative activities for students who do not participate; bias reviewed across language, disability, and writing style.
Run a controlled pilot with us.
Our elementary POC validates technical operability, teacher workflow, and data quality & safety — it does not claim academic effect or AI-detection accuracy. The application collects school-staff information only; no child data is collected here.