Nursing handover
Metrics and safety contract exist. A dedicated matched baseline-vs-CareOS task runner is not built yet. Do not imply participant evidence exists.
Evidence before expansion
Does CareOS actually return time to care?
The studies deliberately measure workflow behavior instead of asking whether the demo feels impressive. Speed only counts when safety and source verification remain intact.
Synthetic only. No names, no patient data, no clinical efficacy claim.
Physician: fragmented systems vs CareOS
Two matched synthetic Infectiology cases. The participant performs the same morning-review task once across simulated KIS/LIS/medication/lab/documents and once in a unified source-linked CareOS view. Odd/even participant codes counterbalance order and case assignment.
Automatically captured: task time, systems opened, context switches and source opens. Observer-scored: errors, missed pending work, corrections, verification behavior, effort and safety stops.
Start physician paired session →CareOS vs CareOS + bounded agent draft
This older A/B test does not represent the fragmented real-world baseline. It asks a narrower question: once source-linked CareOS context exists, does adding a source-linked agent draft improve the workflow without creating recommendation/source-truth confusion?
Run clinician agent A/B →Discharge / case management
Metrics and cross-provider workflow contract exist. A dedicated matched baseline-vs-CareOS task runner is not built yet.
Evidence rules
1. Paired matched cases. Each participant sees baseline and CareOS with different comparable synthetic variants.
2. Counterbalanced order. Both baseline-first and CareOS-first participants must appear in a highlighted set.
3. Safety beats speed. Wrong patient, missed pending work, unsupported/stale claims, draft-as-recommendation or verification collapse block a result.
4. Minimum evidence. No directional role/workflow result before ≥5 complete safe pairs.
5. No clinical-efficacy leap. A faster synthetic workflow only earns the next research question.
The machine-readable contract and aggregation logic live in app/time_returned_to_care.py and scripts/aggregate_time_returned_to_care.py.