Rx360 — Master Product Roadmap to Completion¶

Date: 2026-07-05 · Owner: Gene Lang, PharmD — Director of Pharmacy Operations Built by: multi-agent Workflow (5 workstream roadmaps → synthesis → adversarial critique → 7 fixes applied). Companion: Rx360_Workstream_Roadmaps_2026-07-05.md (the per-workstream detail this synthesizes). Status: 🟡 Draft planning artifact — dates are relative (phase names, not calendar) because every date-bearing source doc is stale-flagged. Not a commitment.

1. The one-paragraph thesis¶

A "completed" Rx360 product is a 100% wellness medication-and-safety system for seniors and their caregivers, in market and generating revenue, resting on four working pillars: (a) a wearable line — Mark I (RxSIDEKICK, medication reminders + manual SOS/circle, shipping) evolving to Mark II (RxBALANCE, automatic fall detection + the Balance Meter fall-risk read); (b) a Medication-Intelligence "brain" (curated drug KB → knowledge graph → grounded retrieval → deterministic clinical rail) that supplies drug-truth to every other engine; (c) algorithm products on top of the brain — the Fall Barometer / Balance Meter and Smart Scheduling (PK/PD-aware reminders) — that are the patented differentiation; and (d) a commercial motion (Star Ratings / RTM / MTM / B2B) funded by pilot-generated usability evidence and a separate USC-led adherence-evidence arm. The shortest honest path is not "build everything" — it is a deliberately sequenced spine: run the Mark I usability pilot to completion → land the human clinical/legal sign-offs that everything is frozen behind → build the brain's terminology-spine-through-clinical-rail once → light up the algorithm engines on that shared substrate → open the first commercial door. Today the product is mostly designed, partly shipping (Mark I only), and blocked at a small number of human gates — the work is far less "invent" than "get the right people to sign, then build in the already-specified order."

2. Phase gates (the whole product, phased)¶

Five phases, today → commercial. Dates are deliberately relative — every date-bearing plan in the source docs (PP-18, the KG effort table) is stale-flagged or unresourced, so committing calendar dates here would be inventing them.

Phase A — Pilot-Ready & Unfrozen (the "confirm reality + unlock the gates" phase)¶

Goal: Re-baseline the pilot against today, clear the front-gate blockers, and unfreeze the two sign-off-frozen models so downstream build can legally start. Must-be-true exit criteria: - IRB classification confirmed in writing (NHSR vs. exemption Cat 2/3 vs. expedited), dated, recorded in STATE/DECISIONS — no more [STALE] flags on PP-4/PP-18. - Anchor-site↔Rx360 BAA executed (PP-84 / PP-23 unblocked, T-014 closed). - Gene + Matilda sign-off on the fall-risk model + 17-PMID evidence set → canonical Balance Meter model UNFROZEN, weights baked, disclosure language upgraded to "clinically endorsed." Per Matilda's 6/23 review the clinical model is essentially signed off already — this is a bounded, scoped session, not an open clinical re-litigation. Outstanding surface = (1) the 20/20 cap-rebalance decision, (2) baking in the three A5 refinements, and (3) the diabetes / muscle-relaxant / anticholinergic weight calls. One item genuinely reopens a settled number: the session must resolve whether the ≥2-CNS-active term is added as a graded interaction — which would restore a second interaction bonus alongside cognitive×polypharmacy. Do not treat "one interaction bonus" as settled; the outcome must reconcile against the patent disclosure, which describes two example bonuses plus the contemplated CNS term. - Dr. B pharmacist sign-off pipeline stood up (even if not all facts cleared) so preliminary→validated can begin. - Pilot-execution reality confirmed (hard prerequisite, not a scheduling detail): the entire PP-18 calendar (screening Apr 20, Pilot 1 May 18, Pilot 2 Jun 15) is now fully in the past with no CHANGELOG entry confirming any phase ran, and STATE calls pilot-ops stale/parked. Before any later phase can claim a "completed take-home phase," confirm in writing (a) whether ANY pilot phase actually ran, and (b) that ≥10 functional Mark I take-home units actually exist (nothing in the corpus confirms Mark I shipped in pilot quantity). If neither is confirmed, the honest critical path has an extra hardware-and-execution link — do not hide it inside "run the pilot." - Two open DECISIONS forks resolved (both gate the Phase-C/D external artifacts): the band-words fork ("Worth a Look" vs "Notable Change" — needs Gene + lexicon) and the market-size fork ("$300B" vs "$528B" — needs one primary CMS/KFF-locatable number). Shipping a B2B one-pager or pitch with an unlocked band word or market number is exactly the SeniorLife-style claims risk in risk #5. - USC counterpart named and a written confirmation of whether a PDC/adherence protocol even exists (the arm currently has no protocol motion in-doc). The USC arm is tracked as a parallel workstream, not a wellness-product phase gate. - PP-18/PP-4 re-dated to real current status; the four soft directions (acquisition-exit, caregiver-first, tagline, adherence-demoted-below-fall) LOCKED.

Phase B — Pilot Executed + Brain Foundation Built¶

Goal: Run the Mark I usability pilot end-to-end; build the brain's foundation (terminology spine → seeded graph → retrieval → rail) and the deterministic layers that don't need the learning loop or Mark II. Must-be-true exit criteria: - Mark I usability pilot completed through the wearable take-home phase (usability + safety endpoints in hand; no efficacy/PDC claim; adherence tagged exploratory-UX only) — contingent on the Phase-A confirmation that a pilot phase actually ran and ≥10 functional take-home units exist. - Brain Phases 0–3 built: terminology spine (RxNorm/RxClass/OpenFDA + crosswalks + provenance/governance), seeded clinical graph (FRID-load query first), grounded retrieval layer, and the deterministic clinical rail in code (fail-closed, allow-listed, clinical-line leakage = 0). - Bout/walking-window detector prototype works: raw Mark I IMU → gait-feature vector on-device (the single most important unbuilt prototype). - Layer-1 baseline calculator built (STEADI + meds, mechanism-locked) and validated against the 754-senior / senior cohort. - Deterministic Smart Scheduling core built on the signed-off KB/02 tolerance table.

Phase C — Evidence + Mark II Hardware¶

Goal: Turn a designed Mark II into a real device with automatic fall detection and the full Balance Meter; produce the pilot results package; open the first commercial conversation. Must-be-true exit criteria: - Mark II BOM locked (barometer required, dedicated wear-detect required, PPG standard-vs-Enhanced resolved), board bring-up done, sensors enumerating. - Firmware v0.3 — fall DETECTION working on Mark II (barometer altitude-drop + 9-intent voice cascade + auto-emergency-calling on no-response). - Full Balance Meter calc chain live (gait features → 30-day baseline → Layer-2 deviation → 3-band output) — gated on the Phase-A unfreeze. - Pilot results package produced (usability + feasibility), scrubbed against SeniorLife claim patterns. - First commercial conversation logged (Outcomes/TDS BD outreach + one MA-plan Star-Rating one-pager) and the non-pharmacy-vendor fee-routing legal question answered. - FDA CDS Software guidance pulled and the gated clinical dashboard validated against it (if a pharmacist-facing surface ships).

Phase D — Commercial / Star-Rating (revenue)¶

Goal: Convert research into a pipeline; first paid B2B contract and first reimbursed claim. Must-be-true exit criteria: - First signed B2B contract (most likely MA-plan PMPM or MTM lead-gen via TDS) — first revenue recognized. - RTM/MTM billing live through a partner (98975/98977/98980 or 99605/99606) off Rx360-generated data. - ≥2 expansion sites (Sites 2–5) live on the anchor-validated playbook (per-site BAA + IRB amendment). - Cross-engine medication-change event live across Smart Scheduling + Balance Meter (the flagship novelty, now that both engines exist). - Learning layers (contextual-bandit timing; brain product-integration/learning loop) shipping behind the pharmacist gate.

Phase E — Scale / Acquisition-Optionality (commercial horizon)¶

Goal: Prove the platform scales and close the acquisition narrative. Must-be-true exit criteria: - Mark II ecosystem accessories (Base Station Mark II, Pill Bottle RFID/OCR, Smart Pill Dispenser) reach shippable spec + prototype. - Serving/scale tier (on-device → BAA private-cloud → cloud) + evals/MLOps (clinician gold set, drift/canary/rollback, GMLP+PCCP) in place. - Full-sensor Balance Meter calibration closes remaining literature-grounded weights with in-house validation data. - USC-co-authored publication + valuation/comparables package (R6) — acquisition-optionality loop closed. - Form-factor-independence demonstrated (algorithm runs on a non-wrist path) — the "algorithm-as-moat" thesis proven.

3. The critical path¶

The chain that gates the whole product — each link cannot start until the prior one clears:

IRB classification confirmed  ──►  BAA executed  ──►  [CONFIRM: did any pilot phase run? do ≥10 units exist?]  ──►  Mark I usability pilot runs to completion
        (Kristen/Monika)          (counsel/PP-23)          (unconfirmed prerequisite — PP-18 calendar all past)   (needs ≥10 functional Mark I units)
                                                                          │
                                                                          ▼
                                              Pilot results package + first commercial conversation ──► Revenue (Phase D)

Gene + Matilda fall-model sign-off  ──►  canonical model UNFREEZES  ──►  Balance Meter Layer-2 / Mark II 3-band output ships
   (the hard freeze)                                                              │
                                                                                  ▼
Dr. B pharmacist sign-off pipeline  ──►  preliminary→validated (pilot drug set)  ──►  brain can EMIT clinical content to members
   (throughput-limited)                                                           │
                                                                                  ▼
Bout/walking-window detector prototype  ──►  on-device gait feature vector  ──►  Layer-2 gait signal ──► (feeds the unfrozen model)
   (the top unbuilt engineering item)

Two parallel spines, one shared endpoint. The ops/evidence spine (IRB → BAA → pilot → results → revenue) and the product spine (fall-model unfreeze + pharmacist sign-off + bout detector → Balance Meter → Mark II) run largely in parallel but converge: the pilot is the only place any threshold moves from mechanism-locked to numeric, and Mark II's value story (fall detection + calibrated band) is what the commercial motion ultimately sells.

Single biggest bottleneck (current): the human sign-off cluster — specifically (1) the IRB classification (stale, unconfirmed, blocks every pilot phase) and (2) Gene + Matilda's fall-model sign-off. Note the fall-model gate is near-closed, not from-scratch: per Matilda's 6/23 review the clinical model is "essentially signed off," and the remaining surface is a bounded set of decisions — the 20/20 cap-rebalance, baking the three A5 refinements, and the diabetes/muscle-relaxant/anticholinergic weight calls (plus the ≥2-CNS-active call that may restore a second interaction bonus). It is a freeze (nothing is baked into canonical yet, so it still blocks the Balance Meter / Mark II spine and the brain's FRID-load query), but it is a short scoped session, not an open clinical re-litigation. Neither gate is an engineering problem; both are calendar-and-decision problems. Behind them sits (3) Dr. B pharmacist sign-off throughput — the brain can be fully built and still emit nothing clinical to a member until this clears the pilot's drug set. The one engineering item genuinely on the critical path is the bout/walking-window detector — nothing in the Balance Meter's live-gait layer can be implemented until it exists, and its feasibility on a wrist sensor in seniors is the product's deepest technical unknown.

4. Cross-workstream dependency map¶

From (produces)	To (needs it)	What	When needed
Gene + Matilda	Fall Barometer, Med-brain (FRID query), Mark II hardware, Smart Scheduling	Fall-model + 17-PMID evidence sign-off → unfreeze canonical model	Phase A (gates everything downstream)
Kristen / Monika / Advarra	Pilot/Clinical, all pilot phases	IRB classification confirmed in writing	Phase A (top gate)
Counsel (via PP-23)	Pilot (PHI phases), all revenue	Executed anchor-site↔Rx360 BAA	Phase A
Dr. B (pharmacist)	Med-brain, Smart Scheduling, Balance Meter	`preliminary→validated` sign-off on KB/02 + FRID/forgiveness tables	Phase A→B (pipeline up A, cleared B)
Hardware (Emil)	Pilot (wearable phase), Fall Barometer	≥10 functional Mark I units; bout detector compute budget	Phase B
Firmware (Mike + hardware)	Fall Barometer, Balance Meter	On-device IMU gait-feature vector (bout detector output)	Phase B
Med-brain (Faraz/Sana)	Smart Scheduling, Balance Meter (Layer-1 med layer), Pilot (med capture)	Curated KB + RAG + rail + medication object + FRID-load query	Phase B (Smart Sched + BM full fidelity)
Fall Barometer / Balance Meter side	Smart Scheduling, Med-brain	Canonical `medication-change` event contract (owned here); FRID-load spec	Phase B→D
Smart Scheduling	Balance Meter, patent portfolio	Shared provenance model + shared med-change event schema (co-defined)	Phase B (coordinate filings)
Mike (AI architect)	Med-brain	Graph-DB / retrieval / rail / eval recommendations	Phase A (gated on his NDA)
Counsel	Smart Scheduling, patent portfolio	FTO search + CDS/wellness-line call + coordinated-vs-separate filing	Phase A→B
Pilot	Fall Barometer, Smart Scheduling, hardware	Field data to move thresholds mechanism-locked → numeric; battery/GNSS curves	Phase B→C
Pilot results + GTM research	Peyman / acquisition thesis, B2B pitch	Results package + Star-Rating one-pager + revenue model	Phase C→D
USC	(parallel workstream — not a wellness-product phase gate)	Separate PDC/adherence IRB arm (own owner, own IRB) + co-authored manuscript	Parallel track; pub feeds Phase E — arm runs on its own timeline, does not gate Phase B close

5. Top risks to completion¶

The two frozen models are single points of blockage on the whole product spine. Gene+Matilda's fall-model sign-off gates the Balance Meter, Mark II's calibrated band, and the brain's highest-value first query (FRID-load). The gate is bounded (per Matilda's 6/23 review the model is "essentially signed off" — remaining surface is the 20/20 cap-rebalance, three A5 refinements, and the diabetes/muscle-relaxant/anticholinergic weight calls), so this is a scoped session, not a wholesale re-review. But the interaction-bonus count is NOT settled and must not be presented as such: the 6/30 evidence handoff lists "≥2 CNS-active +2" as an already-thumbs-up-able graded interaction and explicitly warns that adding it restores a second interaction bonus alongside cognitive×polypharmacy — and Matilda's review still has the ≥2-CNS-active floor as an item to bake. So the very session this roadmap schedules will likely move the model from one to two interaction bonuses. That outcome must be reconciled against the patent disclosure, which describes two example bonuses plus the contemplated CNS term. Mitigation: run the tightly-scoped evidence sign-off session first in Phase A, bounded to the 6/30 evidence deltas + the enumerated weight calls; explicitly decide the ≥2-CNS-active term and update disclosure language to match; don't bundle it with broader model debate.
The pilot may not have started at all — this is an execution risk, not a scheduling one. The entire PP-18 calendar (screening Apr 20, Pilot 1 May 18, Pilot 2 Jun 15) is now fully in the past (today is 2026-07-03) with no CHANGELOG entry confirming any phase ran, and STATE calls pilot-ops stale/parked. Yet Phases B–D all rest on a "completed take-home phase," and the critical path assumes ≥10 functional Mark I units exist — which nothing in the corpus confirms. Treating this as "the pilot slipped and needs re-dating" understates it: the honest question is did any phase run, and does the take-home hardware even exist? If not, there is an extra hardware-and-execution link the roadmap must not bury inside "run the pilot to completion." Mitigation: make the two-part confirmation — (a) did any pilot phase run, (b) do ≥10 functional Mark I take-home units exist — an explicit Phase-A prerequisite before re-dating PP-18/PP-4 and before any downstream phase claims a completed take-home phase; and confirm the IRB determination in writing as the paired front-gate.
The preliminary-content bottleneck becomes the critical path. You can build the entire brain (graph + retrieval + rail) and still emit zero clinical facts to a member until Dr. B clears the pilot's drug set. If sign-off throughput isn't resourced now, engineering finishes and the product stays mute. Mitigation: stand up the sign-off pipeline in Phase A, scope v1 validation to the pilot's actual drug list (not all 27 classes), and resource reviewer cadence explicitly.
Wrist-worn gait fidelity in seniors is genuinely uncertain — and it underpins the Mark II value story. The pipeline doc is candid that the nearest precedent (Apple) is weakest exactly on the metrics that matter (double-support, asymmetry, sit-to-stand), and Rx360 has no in-house gold standard to validate against. Mitigation: prioritize cadence (the one wrist-validated fall-risk signal), flag weak metrics low-confidence, prove the bout detector on-bench in Phase B before committing Mark II claims, keep competitor-outperformance claims internal until Rx360 has its own data.
A claims-line violation in a shared artifact triggers a SeniorLife-style warning letter. SeniorLife (8/21/25) is the exact analog; a single "proactively identifies fall risk" or an implied "Mark I does automatic fall detection / calibrated balance score" in a deck, results doc, or B2B one-pager re-classifies the device or draws enforcement. The docs-site audit already caught product docs still selling "automatic fall detection." Mitigation: mandatory Wellness-Lexicon + SeniorLife-pattern scrub on every outbound; hold the locked band words and neutral med-card copy; keep the usability/efficacy wall (separate USC arm) crisp; never let Mark I imply fall detection.

(Honorable mention — resourcing/dates: the brain's build timeline and Mike's onboarding are undated and unresourced, and the USC adherence arm has no protocol motion in-doc. These pace Phases B–C but are TBD / need Gene, not structural risks.)

6. Immediate next actions (the next ~6 weeks)¶

Prioritized across all workstreams, drawn from each roadmap's "Now" band. The first three are the front-gate unblocks; the rest can run in parallel.

Confirm the IRB determination in writing (Kristen/Monika/Advarra) and record it CONFIRMED in STATE + DECISIONS; clear the [STALE] flags on PP-4/PP-18. Top blocker — nothing pilot-side runs until this lands.
Push the anchor-site↔Rx360 BAA to execution (unblock PP-23, close T-014). Gene can push this directly since he owns the pharmacy.
Run the Gene + Matilda fall-model evidence sign-off session — a bounded, scoped session (the model is "essentially signed off" per 6/23), covering the 20/20 cap-rebalance, the three A5 refinements, the diabetes/muscle-relaxant/anticholinergic weight calls, and the ≥2-CNS-active decision (which may restore a second interaction bonus — reconcile against the patent disclosure) — to unfreeze the canonical model and unblock the Balance Meter, Mark II, and the brain's FRID-load query.
Confirm pilot-execution reality, then re-date PP-18/PP-4. First confirm in writing whether ANY pilot phase actually ran and that ≥10 functional Mark I take-home units exist — the whole PP-18 calendar is in the past and pilot-ops is stale/parked, so this is a "did it start" question, not just a re-scheduling one. Then re-date PP-18/PP-4 to real current status (which phases ran, which slipped) with a corrected critical-path table — and scrub while you're in the file: replace the anchor-site initials with "anchor site" (Hard Rule #2), keep the study/participant/"Study ID"/"Pilot 1/2" language internal-ops-only (acceptable internally, but must never appear in any consumer-facing or externally-shared derivative — Hard Rule #6), and note that April 20 is consumer discovery, not a "study."
Stand up the Dr. B pharmacist sign-off pipeline and scope its first pass to the pilot's actual drug set (not all 27 classes) — the pacing gate for member-facing clinical content.
Lock the four soft directions + two open forks via /lock-decision: the four soft directions (acquisition-exit strategy, caregiver-first messaging, "Stay Healthy Stay Independent" tagline, adherence-demoted-below-fall, open 2+ months) plus the two ⚠ OPEN DECISIONS forks that gate the Phase-C/D external artifacts: the band-words fork ("Worth a Look" vs "Notable Change" — needs Gene + lexicon) and the market-size fork ("$300B" vs "$528B" — needs one primary CMS/KFF-locatable number). A one-pager or pitch shipped with an unlocked band word or market number is the SeniorLife-style claims risk in risk #5.
Send the staged brain packages (D-010 v2 + D-011 Meenakshi guidance) to unblock the Faraz/Sana input lane — Gene's explicit send-approval required; nothing posts without it.
Onboard Mike + close his NDA, then have him drive his four owned decisions (graph DB, retrieval stack, rail enforcement, eval/observability) to written recommendations.
Start the UMLS/SNOMED license request (named Phase-0 blocker, real lead time) — or make the Gene call to defer SNOMED and anchor v1 on RxNorm/RxClass.
Kick off the bout/walking-window detector prototype (Mike + firmware) — the single most important unbuilt engineering item; nothing in the live-gait layer can proceed without it.
Confirm the non-provisional patent filing status (STATE shows "believed FILED, no serial" — UNCONFIRMED) before anything IP-derived ships externally, and commission the Smart Scheduling FTO search with counsel. Independent of serial-number confirmation: the filed disclosure is known to predate the 6/29 orthostatic-bonus removal + 6/30 evidence set + gait-DSP work — the working model and the filed disclosure now differ, and a delta memo is already staged. Route that delta to counsel as a continuation input now; don't wait on the serial number.
Name a USC counterpart and confirm whether a PDC/adherence protocol even exists — the arm currently has no protocol motion in-doc. Track it as a parallel workstream with its own owner and IRB; do not let it gate the wellness-product phases.
Resolve the delta-gate threshold conflict on paper (Sensor/DataFlow ~18%/25% vs. Canonical ~12%/15%) — one reconciled mechanism, numbers stay pilot-gated — so no two docs cite conflicting %.

Sequencing note: items 1–3 are the true unlocks and should get Gene's calendar first (they're other people's availability, not build hours); 4–6 are days of Gene's own work; 7–13 run in parallel once the gates are moving. Every outbound in this list respects the hard rule — nothing posts to Jira/Drive/Confluence/Gmail without Gene's explicit chat approval.