Insights
The three metrics this product is judged on (PRD §8), computed from live data.
1 · Import accuracy
Clean imports
2 of 2 saved with zero manual correction
Median paste → save
Target: under 60s for a ~15-exercise list
Median list size
Exercises per imported routine
| Routine | Exercises | Corrections | Paste → save | Engine |
|---|---|---|---|---|
| Knee Rehab - Medium | 5 | 0 | 41s | rules |
| Knee Rehab - Strength & Control | 14 | 0 | 41s | rules |
2 · Adherence, last 4 weeks
Sessions logged
against 16 prescribed
Roster adherence
Shared sessions ÷ prescribed sessions
Patients at risk
Below half their prescribed frequency
| Patient | Logged | Prescribed | |
|---|---|---|---|
| Derek Wong | 14 | 16 | |
| Mei Tan | 0 | — |
3 · Does the tool outlive the acute phase?
The metric that matters most needs 4+ weeks of data per patient before it means anything. The roster table above is the input; the retention curve — share of patients still logging in week 5, week 8, week 12 — should be the headline chart once there is a real cohort. It is deliberately not faked here.
Event stream health
Events are stored first-party with a salted actor hash and no free-text. See src/lib/analytics.ts for the taxonomy.