Linewize

Linewize ANZ (part of the Qoria family) — Keystroke-level and full-screen wellbeing corpus from student devices — and uniquely, captured offline. Monitor risk-assesses everything a student types across Google Workspace, Microsoft 365, offline documents, web chat and social media, whether the device is online or not, buffering on device and uploading when the device reconnects. Each detected behaviour becomes an alert rated 1-5 on frequency and severity, and every high-level alert is reviewed by a trained human moderator who confirms or rejects it for tone, context and pattern-over-time before…

Linewize offers (Alternative, Sentiment) — Two very different layers. The alert layer is small and precious: est. tens of thousands of human-adjudicated alerts per year for this division, since only high-level alerts draw moderator review. The capture layer is enormous and almost certainly not licensable: every keystroke and screen event across all monitored students during school hours, which at even 10K students is a continuous drafting stream. The commercially interesting asset is the alert layer — the adjudicated extremes plus the rejected flags — not the raw stream..

Highest-value use is the human-adjudicated false-positive set: real drafted text that a model flagged and a trained human rejected, which is the calibration data content-moderation and youth-safety classifiers most need and least able to obtain. Also: graded-severity training data for risk models that must rank rather than merely detect; cyberbullying and offensive-language corpora in adolescent register across Australian, New Zealand and British English; drafting-context data (what students write in documents and chat versus search) for early-intervention and wellbeing models; gaming-compulsion signals from a category peers rarely measure; longitudinal deterioration series for public-health research; and offline-capture methodology for privacy-preserving edge inference.

The data is with 4 years of history.