The rest, one by one.
The other use cases you raised, ranked by how much they add once the spine exists, and how much lift each one takes to build. Most sit on top of the same context layer, so they get cheaper, not harder, as we go.
Tier 2 near-term: feeds and sharpens the spine
Behaviour pattern detectionspots the avoidance and drop-off patterns automatically · #5How Nudges, calls and app logs get timestamped and tagged, then a lightweight model flags recurring pre-lapse sequences for a counsellor to review.
Medium lift
Emotional tone detectionreads sentiment across messages and calls · #6How Message-tone scoring runs on each chat and call log nightly, flagging shifts toward frustration or withdrawal for the Nudge Friend to see.
Medium lift
Smart nudge enginedrafts the nudge itself, human-approved · #2How Approved nudge templates plus the patient's three profiles go into a prompt that drafts the message text, and every counsellor edit sharpens the next draft.
Medium lift
Feedback loop & survey analysisfolds survey and outcome data back in · #10How A nightly job scores each patient's survey answers and glucose or adherence changes, then adjusts their nudge weighting before the next cycle runs.
Medium lift
Tier 3 later: higher effort or clinical risk
Buddy voice callsreaches patients who don't live in the app · #15
Infra-heavy
Conversational Friend botpatient-facing chat: needs safety rails · #3
Safety-gated
Progress & HbA1c predictionforecasts outcomes once the data is deep enough · #12 · #1
Needs data
Content, themes & gamificationgenerated content, cohort themes, streak mechanics · #7 · #8 · #13
Later
Training sim, Play, infra & safetycounsellor training, sandbox, platform hardening · #9 · #14 · #19 · #20
Separate track
Nothing here is off the table. It's a running order: we build the spine, prove it on your
real caseload, then take these one at a time, each reusing what's already there.
Start with the Nudge Friend view