Smart Scheduling KB — Synthesis for the 10–11 Deck Wrap¶
Date: 2026-06-03 (for the 6/04 patent working meeting) · Owner: Gene What this is: the distilled output of the five KB deep-research docs — the strongest, citeable facts, each tagged to the slide it bolsters, plus the reframes the research forces and the counsel hand-off list. Read this, then we drop the evidence onto the deck.
One-line read: the literature validates our design, the prior-art map narrows our claim to the integration, and the Jan-2026 regulatory shift moves the wellness line in our favor. Lead the patent on the system — clinically-curated tolerance × cross-engine event × provenance × dose-context × de-escalation — not on prediction or personalization.
A · The 12 sharpest facts (each ready to quote)¶
Evidence of need / why this beats a dumb reminder 1. No single tactic wins; combinations do. Largest recent systematic review (128 studies, JAGS 2025): 75.8% of interventions helped but "mainly in the short term," and no single type consistently won. → reminders are the floor, not the differentiator. (KB01, Scotti 2025) 2. Behavioral structure beats bare reminders by a wide margin. Older-adult meta-analysis: implementation-intention + monitoring g=1.22, health-belief education g=0.88; "reminders alone insufficient without behavioral coaching." (KB01, Jeon 2022) 3. Confirmation > passive alert — proven at scale. 46,581-person RCT: a one-way reminder alone had no significant active-period effect; prediction + commitment drove the result. → external proof for confirmation-first. (KB01, Dai 2017) 4. Reminders can backfire for belief-driven skipping. Cue prompts only fix forgetting; for intentional non-adherence they're "ineffective and potentially counterproductive." → detect intent, route to a human, don't nag harder. (KB01, Necessity-Concerns meta-analysis) 5. Routine-anchoring builds durable habit (~66-day median to automaticity); routine cues work as well as clock cues and survive better. → the literal grounding for "learns your rhythm." (KB01, Lally 2010 / Keller 2021)
The tolerance table (our distinctive core) 6. "Forgiveness ≈ duration of action − dosing interval" — a named, cited clinical concept, driven by PK persistence and PD persistence. Aspirin is the proof PK half-life alone misleads: ~15-min half-life, ~2-day platelet effect. → why the window must be clinically curated, not naive half-life math. (KB02, Assawasuwannakit 2015, PMC4394614) 7. A weekly biologic and a q8h drug genuinely behave differently. KB02 has 27 drug-class rows with cadence, half-life, qualitative forgiveness, missed-dose handling, and food/timing — every value flagged for Dr. B sign-off. The high-stakes set (insulin, warfarin/DOACs, methotrexate-weekly-never-daily, denosumab rebound, sulfonylurea-with-meal) gets explicit-confirm + escalate, never dosing logic. (KB02 §2/§4)
The defensible space / FTO 8. The closest prior art is real and must be named: US9070267B2 (Alarm.com, priority 2011) pre-arts miss-prediction + adaptive/withheld reminders + caregiver escalation — and even gestures at "therapeutic windows." Add Medisafe's patented JITI, a cloud-ANN reminder patent (US 12040066), US10255412B2 (generic ± window), and the RL/JITAI corpus. Do not lead on prediction or personalization. (KB03 §2a) 9. The open ground is the integration: clinically-curated PK/PD tolerance behavior + the cross-engine medication-change event with the fall barometer (no competitor or patent pairs scheduling with a fall engine) + provenance-tiered scheduling + dose-time context payload + de-escalation (everyone else escalates). (KB03 §2b) 10. Dose-time context is confirmed white space. Systematic review (Springer 2020): missed-dose & administration guidance in patient leaflets/SPCs is "inconsistent or absent." We fill it from DailyMed/SPL data. (KB03 §2b / KB02 §3)
The build + the horizon 11. The architecture is groundable and buildable on free, authoritative standards: RxNorm (identity) → DailyMed/SPL + openFDA (dose context) → our pharmacist-curated PK/PD tolerance table served via a RAG layer behind a deterministic safety rail → FHIR Dosage + iCal RRULE (schedule) → HealthKit (WWDC25 Medications API) / Health Connect + OS notifications (delivery) → shared med-change event bus to the Balance Meter. NCPDP SCRIPT feeds the pharmacist-verified provenance tier. (KB03 §3, KB04 §2) 12. The regulatory line just moved our way. Jan-2026: FDA widened the general-wellness/CDS lane (non-invasive sensing + wellness insight can ship without clearance; positioning, not capability, decides classification — wellness products may even tell a user to "seek evaluation") while CMS lowered RPM/RTM thresholds (2–15 days, 10-min, new code 99445). The wellness↔reimbursement boundary is now a staged business decision. Adherence (PDC) is ~11% of Part D Star weight — the buyer's rationale. (KB05 §3 — directly serves Elliot's "let the lawyers decide the line")
B · Slide-by-slide upgrades (current 11-slide deck)¶
| Slide | Drop in | Source |
|---|---|---|
| S3 Evidence of need | Facts 1–4: the g=1.22 vs "reminders-alone insufficient," the 46,581-RCT confirmation result, the "reminders backfire for belief-driven skipping." Pair with our own field evidence (confirmation>reminders; fatigue, not forgetting). | KB01 |
| S5 Cadence/tolerance table | Fact 6 (the forgiveness definition + aspirin PK/PD proof) as the header concept; cite that KB02 carries 27 curated classes pending Dr. B. Keep the table illustrative, not exhaustive. | KB02 |
| S6 Learned rhythm | Fact 5 (routine-anchoring, ~66-day habit) — gives the rhythm feature a citation, not just intuition. | KB01 |
| S7 Pre-emptive + cross-engine | Frame prediction as table-stakes done well (AUC 0.75–0.85 is mature), and put the weight on the cross-engine med-change event (Fact 9) as the novel piece. | KB03/KB04 |
| S8 What to protect | Already reframed to "the system, not any single trick" — reinforce with Facts 8+9: name the Alarm.com reference as why we claim the integration. | KB03 |
| S9 Prior art | Fact 8 verbatim — the honest crowded-vs-open table. This is the slide counsel will linger on. | KB03 |
| S10 Wellness, not device | Fact 12 — the Jan-2026 FDA widening + the "positioning decides classification" line. Reframes the wellness lane from a constraint to a staged optionality. | KB05 |
| S11 Open & next | The FTO ask (Section C below) + the build-first order (KB04 §6: ground → rail → instrument → deterministic → learn). | KB03/KB04 |
| (optional new) Vision close | Fact 12 + the acquisition angle: the durable asset is longitudinal multimodal baselines on older adults + a pre-engineered RPM/SaMD ramp — not the reminder. | KB05 §5 |
C · Counsel hand-off list (the FTO ask, sourced)¶
- Search the crowded corridor head-on — assume miss-prediction + adaptive timing are blocked broadly: US9070267B2 (Alarm.com, closest art), Medisafe JITI family, US 12040066 (cloud-ANN reminders), US10255412B2 (dosing window), US8040236B2 + Google meal-reminder filing, and the RL/JITAI literature.
- Claim the combination + the un-arted cores — clinically-curated PK/PD tolerance behavior × cross-engine med-change event × provenance-tiered scheduling × dose-context payload. Prediction/de-escalation become dependent/implementation claims, not the independent claim.
- ⚖ The CDS/device line — PK/PD-curated tolerance + miss-prediction sit near the CDS boundary. Counsel opines where "convenience reminder" ends and "regulated CDS/device" begins. Note the Jan-2026 guidance widened this lane (KB05 §3) — the line moved toward us.
- Portfolio question — file Smart Scheduling + Balance Meter separately or as one portfolio? The shared med-change event + shared provenance model argue for a coordinated filing.
D · What the research changes in our thinking¶
- Reframe prediction/personalization as table-stakes, not the claim. (Already done on S8; reinforce on S7/S9.) The Alarm.com 2011 reference makes this non-negotiable.
- De-escalation is a genuine, un-obvious differentiator. Everyone escalates; the literature + our field evidence say the failure mode is fatigue/burden, not forgetting — so removing nags as a routine proves out is the contrarian, defensible move.
- The wellness lane is an asset, not a limitation. Jan-2026 FDA + CMS shifts mean wellness-first with an RPM/SaMD-ready architecture is the optimal posture — fast to ship, and pre-engineered optionality into reimbursement. This is the strongest possible framing for Elliot's "shoot for the stars, let the lawyers draw the line."
- Build order is now explicit: curated drug KB → RAG layer → deterministic safety rail → event schema + de-ID boundary → pharmacist HITL queue → then deterministic scheduler → then learning. Ground before you generate; rail before you emit; instrument before you learn.
- Metrics discipline: confirmation rate (primary + bandit reward), false-alert/alarm-fatigue rate as a guardrail (a scheduling win that raises it is a loss), time-to-enter-a-med. PDC/MPR for benchmarking only — they measure possession, not ingestion.
E · Sign-off / open flags carried forward¶
- KB02 is DRAFT pending pharmacist (Dr. B) sign-off — every half-life, forgiveness tier, and the NTI list (confirm digoxin/phenytoin/lithium). Numbers are design scaffolding, not production truth.
- A few KB02 primary sources were paywalled (AARDEX, two Springer/ASH full texts) → cross-check against FDA/DailyMed labels before sign-off.
- All five KB docs are fully cited (26 / 28 / ~40 / 26 / 30 sources respectively).