Operator access to the device fleet. Credentials are the Supabase Auth operator users provisioned at deploy (role=operator).
Demo mode renders the full console against built-in, IP-safe sample readouts so you can see the surface before any device has paired. Everything is clearly badged DEMO.
ITB-PILOT-001 · WS3 · builds against device-API contract v1 (E4/E5) + itb-pilot-readout/v1
This console lets you watch the study phones, pair new ones, and read each subject's day-by-day estimates. Everything here is an observational estimate — a prompt to help you decide who to check in with, never a diagnosis and never a dose instruction.
Engagement — who is in the study and reporting in.
Measurement quality — whether we can trust what's coming in.
Estimated state is the headline: On (the medication looks active), Off (looks inactive), or Indeterminate (not enough clean signal to call). Confidence is how sure that estimate is, on a five-step scale from very low to very high — on the trend chart, a smaller point means lower confidence.
Before you rely on a reading, the phone checks itself three ways:
The combined badge reads Trusted only when all three clear. Degraded means a confidence penalty; Not comparable means a different carry position than the baseline; Unsupported means the phone can't deliver the measurement at all; Gates pending means the phone hasn't reported them yet.
Calibration is the subject's personal baseline — readings are most reliable while it's fresh, and read Stale after a few days. Missed counts recent days with no completed session. Status tracks the phone: awaiting pairing → paired → online / offline.
You can pair a device and send a subject a templated, non-prescriptive nudge — request a follow-up, a check-in, a reminder. The clinical decision stays in your own system; this channel only nudges. There is no dose entry, and the console never shows a numeric score or any model internal — only qualitative bands.
These estimates are exploratory monitoring — an early read to guide your attention, not proof of anything. Whether the technology works is decided by the pre-registered analysis plan, never by this dashboard.
Click a subject to open the readout trend, the measurement-trust breakdown, adherence, and notification controls.
The tiles above summarise the latest session. Below: Measurement trust is the three quality gates for that reading; Estimated medication state is the day-by-day trend (a larger point = higher confidence); Adherence is the subject's own check-in — self-reported, not a measurement; Derived qualitative metrics are opaque band-only slots whose meaning stays on-device. Clinician actions send a templated nudge only — never a dose. All bands, no numeric scores.
Three on-device honesty gates for the latest session: can the phone deliver the measurement, did it sample fast enough, and was it carried where the signal is visible (vs the calibration baseline).
Observational estimate per session. Point size reflects confidence band. No numeric scores.
Opaque metric slots, 5-band scale (very low → very high). Mapping to meaning stays on-device.
Templated, non-prescriptive notifications only. The clinical act happens in your own system — this channel only nudges. No dose entry.