The evidence behind reviewed records.
How we study documentation burden, human review, and continuity in frontline care — and what the data says about getting it right.
Human review, measured
We track how AI drafts and human corrections interact, not just model accuracy in isolation.
Provenance first
Every claim a model makes is traced to its source context, so findings are auditable.
In real settings
We design studies to run with partner care homes on real shifts, not synthetic benchmarks.
Research agenda & briefs
Illustrative research agenda · not published studies
Human-in-the-loop documentation: accuracy and accountability in AI-drafted care records
Measuring correction rates, time-to-approval, and error types when supervisors review AI-drafted forms.
Discuss this brief →Documentation burden on night shifts: a time-and-motion analysis in long-term care
Where the minutes go between an event and an official record, and how much is spent re-telling the same story.
Discuss this brief →Operational memory: a framework for continuity across shift changes
Defining what context must carry forward — and in what form — for the next shift to act safely.
Discuss this brief →Provenance and audit trails for AI-assisted clinical documentation
A technical approach to tracing every field back to its source, and why it matters for accountability.
Discuss this brief →