Back to Dashboard
Study Protocol CLAUDE.md
A compact operating protocol for running the AGS study without accidentally biasing the intervention or damaging data integrity.
Participant procedure
- Confirm eligibility and explain study purpose in neutral language.
- Open the participant app and allow the participant to control the device whenever possible.
- Let AGS present language, privacy, and consent screens.
- Do not mention condition assignment or expected support behavior.
- Observe only enough to support safety and logistics.
- After completion, allow optional survey submission.
Phase rules
| Phase | Allowed | Restricted |
|---|---|---|
| Calibration/pilot | Adjust thresholds, content, templates, locators, extension instructions, and capture fallback. | Do not treat pilot outcomes as main effect evidence. |
| Main | Monitor, export, resolve alerts, withdraw/exclude with reason. | No threshold/content changes unless protocol permits and audit explains. |
| Closeout | Resolve alerts, freeze exports, document exclusions, preserve config snapshot. | No retroactive operational edits without audit trail. |
Condition protection
- Participants must not know whether they are control, governed adaptive, or always adaptive.
- Facilitators should not promise that help will appear.
- Overlays must remain visually indistinguishable across conditions.
- Manual facilitator help should be logged in study notes if it changes the task outcome.