Person-centered care plan

The controlling public source for this draft is the CMS GUIDE Request for Applications, Appendix B, Section 2 ↗. CMS requires a beneficiary-led person-centered care plan based on the comprehensive assessment, health-related social-needs screening, and home-visit findings; it requires the plan to address the beneficiary's goals, strengths, preferences, needs, services, supports, providers, payment responsibility, and caregiver education and support when applicable. CMS also requires revision when circumstances, goals, preferences, or needs change, incorporation into the EHR, and sharing with an outside PCP and other providers as applicable.

The literal work sequence, Proxi labor assumptions, source hierarchy, negative-evidence examples, and closure tests below are Proxi operating design, not additional CMS requirements. This remains a draft until the Participant's executed GUIDE Participation Agreement and current operating instructions are checked and the named clinical, compliance, privacy, and operational owners approve it.

Medical SOP purpose and service boundary#

This SOP turns the 52 existing Care Plan tasks into one operating sequence for:

1. opening an initial plan, scheduled review, or event-driven revision;
2. assembling the assessment, social, home, clinical, medication, provider, service, caregiver, and prior-plan inputs;
3. eliciting the beneficiary's goals, strengths, preferences, needs, and choices;
4. identifying realistic clinical, functional, behavioral, psychosocial, safety, community, caregiver, coordination, and payment options;
5. drafting, reviewing, correcting, and finalizing the written plan;
6. incorporating the plan into the EHR and sharing it with permitted recipients; and
7. turning accepted content into owned work, following the results, and reopening the plan when needed.

The comprehensive assessment supplies assessed facts and clinician findings; this SOP does not repeat that assessment. Medication reconciliation remains in 05_SOP_Medication_Management.md. Outside-provider coordination and transitions remain in 04_SOP_Care_Coordination.md. Community-service connection remains in 15_SOP_Referral_and_Services.md. Caregiver services remain in 16_SOP_Caregiver_Education_and_Support.md. The Care Plan names and carries those decisions and services without taking over the professional act or external service.

Authority and role assignment#

The Puerto Rico L2 navigator lane performs no clinical or medical act. Every L3 or L4 act routes to Proxi's separate U.S. clinical workforce: an RN, LCSW or other behavioral clinician, NP/PA/CNS, physician/MD, or other appropriately licensed professional for the specific act. Before that professional acts, confirm the beneficiary's current physical location and the professional's authority for that jurisdiction and act. L0-L2 may continue permitted collection, education, navigation, and connection work when no clinical act is occurring.

Cross-pillar preparation and handoffs follow 25_GUIDE_Eight_Pillar_Service_Integration.md. This cross-reference does not add a general navigator or whole-plan clinical approval beyond the task-specific authority below.

Scoped clinical contributions and accountable plan completion#

A Care Plan may contain work from several professionals. Two kinds of accountability must remain separate:

• A scoped attributable clinical contribution is an answer to an exact clinical question or approval of an exact clinical plan element by a professional whose location authority and scope cover that act. It identifies the question, facts reviewed, professional identity and credential, decision, date, limitations, effective timing, monitoring/follow-up, and unresolved matters. It does not approve unrelated plan content.
• Accountable plan completion is the whole-plan review performed by one Participant-authorized Care Plan owner. That person confirms that the beneficiary-led review occurred, every applicable section has a truthful disposition, every accepted action has an owner, every clinical element carries a current contribution from the correct professional, conflicts remain visible until properly resolved, and the final plan is internally coherent.

The accountable Care Plan owner does not diagnose, prescribe, reconcile medication, decide clinical safety, or override a clinician unless that person separately holds and is exercising the authority required for the exact act. When reconciling two clinical positions itself requires clinical judgment, the affected element remains unresolved until the professional authorized for that question supplies the disposition.

The public GUIDE RFA requires a beneficiary-led Care Plan based on the comprehensive assessment and an interdisciplinary team that includes a dementia-proficient Part B E/M clinician. It does not state that every Care Plan needs a generic RN co-signature or whole-plan clinical blessing. Proxi therefore uses the current authorized clinical work produced by the assessment and other service episodes, obtains new scoped contributions only when needed, and keeps plan-level completion separate. Whether Proxi's L2 may serve as the Participant-authorized Care Plan owner, or whether the acquired Participant requires another role or an additional clinical attestation, remains subject to O-003 and O-PA-001.

Decide whether current clinical work is enough#

For an initial plan, annual review, or focused revision, no new clinical contribution is needed only when all five statements below are true:

1. The work does not add, remove, or change the clinical meaning of a plan element. A change limited to wording, scheduling, logistics, preference, a community service, a provider of the same service, payment, or participation is not by itself a clinical change.
2. No new or corrected clinical fact has arrived since the most recent scoped contribution on the affected element. This includes a symptom, transition, medication event, safety signal, or late/corrected result.
3. No carried clinical decision is past a recheck or monitoring window stated by the deciding clinician. If the clinician stated no window, the next comprehensive assessment is the default recheck; passage of time alone does not make the decision invalid between assessments.
4. No beneficiary or caregiver clinical question, request for a clinical alternative, or refusal needing an explanation of clinical consequences remains unanswered.
5. No unresolved clinical recommendation conflicts with the element.

When all five are true, the planning record names the source contributions used and states why the occasion required zero incremental Care Plan clinical minutes. When any statement is false, only the affected element goes to the professional authorized for that exact question. The beneficiary-led, logistical, community-service, and other unaffected work continues; one clinical question does not turn the whole plan into a clinical review.

Work one or several clinical questions#

For one clinical question, Proxi assembles the exact question, current source-linked facts, beneficiary goals and preferences, prior decisions, and required response. It sends the question to the least-cost professional whose license, scope, and beneficiary-location authority cover the act. Only the affected element waits. The Care Plan owner inserts the returned decision without reinterpretation and confirms that the response answered the question asked.

For two or more clinical questions, Proxi separates the work by professional authority and sends independent questions in parallel. One question waits for another only when the first answer is genuinely required to decide the second. Each response is added as its own scoped contribution; one professional's response or silence does not decide another professional's question and does not stop unrelated plan work.

If two recommendations conflict, both remain visible. The navigator and Care Plan owner coordinate the resolution but do not choose the clinical answer. The professional authorized for the disputed act supplies the disposition. Silence, an office-staff response, or an old note is not concurrence. After the approved number of attempts, the authorized GUIDE clinician either resolves the exact issue within that clinician's own authority or the element remains truthfully open with the safest already-authorized current course and a named follow-up. O-048 owns the unresolved attempt and disposition policy; medication-change concurrence continues to follow O-005.

If a late or corrected clinical result contradicts content already distributed, the affected part of the plan is revised and every recipient of the superseded content receives the authorized correction. Unrelated plan content is not reopened. O-049 owns the final recipient, urgency, communication, and correction-evidence policy. A result that merely confirms the current plan is attached as additional evidence and does not create a new clinical review.

Required care-planning record#

The working record must retain, as applicable:

• the planning occasion, trigger, source, date, and whether the work is initial, review, or revision;
• beneficiary identity, current contact route, communication and accessibility needs, and participation preference;
• the beneficiary's direction about caregiver involvement and any established representative authority;
• each assessment or record used, its author or source, event date, receipt date, and whether it is current, stale, missing, or conflicting;
• the beneficiary's own words for goals, strengths, preferences, refusals, and requested changes;
• caregiver and clinician statements kept separate from beneficiary statements when they differ;
• each accepted recommendation or action, named owner, intended provider or program, expected payment responsibility, timing, and open dependency;
• each clinical decision and the qualified professional who made it, without replacing the source record with an AI summary;
• each unresolved question, refusal, disagreement, failed handoff, and planned follow-up;
• the current plan date, review with the beneficiary, EHR incorporation, each authorized recipient, delivery result, and acknowledgment or access evidence; and
• the actual outcome of every carried-forward action, including failure, denial, waitlist, withdrawal, or no response.

Unknown, stale, conflicting, or late-arriving information stays visibly unresolved. A concise care-plan draft is not permission to discard the source facts or flatten disagreements into false consensus.

Universal safety interrupt#

If planning reveals a possible immediate safety issue, new or worsening symptom, medication problem, abuse or neglect concern, suicidal statement, behavioral-health crisis, conflicting clinical instruction, or request for urgent medical advice:

1. L1 or L2 records the report verbatim, confirms callback information and current location when safely obtainable, and stops routine advice or plan finalization for the affected issue.
2. The approved clinical, safeguarding, or emergency route is invoked. L0-L2 do not determine urgency, causality, diagnosis, treatment, or disposition.
3. A professional in the separate U.S. L3 pool may assess only within actual location-based authority and scope. A professional in the separate U.S. L4 pool or the applicable treating clinician makes medical decisions; emergency or protective services perform their own acts.
4. A failed connection is actively escalated. An alert, attempted call, or sent message is not a completed safety handoff.
5. The safety disposition and remaining Care Plan work are documented separately. Completing the immediate handoff does not silently close the underlying need.

Care-plan lifecycle map#

Human-workload calibration#

The numbers below are provisional active-human-work estimates, not CMS requirements, billing units, encounter-length requirements, staffing guarantees, or field observations. Software and AI processing is shown as work performed but counts as zero human minutes. Beneficiary, caregiver, representative, interpreter, outside-provider, payer, and community-organization time is excluded unless Proxi actually pays that person.

The audited model treats one planning conversation as the parent event and separates software work, Philippines administrative exceptions, Puerto Rico relationship work, clinical judgment, and outbound release.

At the settled labor rates, the Puerto Rico portion is approximately $4.03 low, $12.47 typical, and $27.50 high per initial Care Plan. Philippines administrative labor is $0 low, $0-$0.46 for a typical triggered 0-5-minute exception, and $1.19 at the 13-minute high cap. Clinical and one-minute V1 release events remain separate because the actual role and number of triggered questions or packets determine their cost.

All clinical minutes in this SOP belong to the separate U.S. L3/L4 workforce. They are never Puerto Rico navigator minutes. The Puerto Rico 11 / 34 / 75-minute scenarios cover only the nonclinical human relationship, clarification, facilitation, and coordination work that remains after preparation. The low route does not buy a navigator merely to bless a clear authenticated choice or clean direct plan review.

Zero incremental Care Plan clinical minutes does not mean the beneficiary had no clinical work. It means the current assessment, medication-reconciliation, transition, behavioral, or other authorized source contribution already answers every clinical question needed for faithful incorporation into the plan, so the same paid clinical work is not counted again. A new fact, stale or inapplicable decision, changed recommendation, conflict, unanswered question, or requested clinical alternative opens a new scoped contribution.

The labor ledger counts one actual human event once. If one clinical review, administrative pursuit episode, or beneficiary/caregiver interaction supplies evidence to Care Plan and another GUIDE service, link that event to both obligations without duplicating its paid minutes. A linked event satisfies only the parts its actual content and evidence support.

Puerto Rico L2 workload is 11 / 34 / 75 active minutes for low / typical / high initial-plan scenarios; 75 is the high case. A clear authenticated choice or clean direct review adds no L2 approval step.

The annual reviewed-no-change or changed-plan route is the same human event as the annual care-plan/relationship allowance inside the navigator tier totals in 22_GUIDE_Human_Workload_and_Staffing_Estimates.md; it is counted once. The separate U.S. clinical minutes below are generated only by clinical questions the planning work itself produces and are not added again when the clinical work already belongs to assessment, medication, transition, behavioral, or another service episode.

Clinical-work pattern by Care Plan case#

Typical Puerto Rico navigator timeline — 34 active minutes#

The 34 minutes are active navigator time, not a mandated call length. They may occur in one interaction or multiple beneficiary-preferred contacts. If a clinical question appears, the prepared packet routes directly to the authorized clinician and that clinician's time is recorded separately.

Low and high Puerto Rico navigator timelines#

Per-task human-minute breakdown#

Read every numeric cell as low / typical / high active human minutes. The phase subtotals are the additive staffing ledger for the parent planning interaction. A Philippines value marked with an asterisk is a possible slice for that trigger, but all Philippines slices together are capped at one 13-minute initial-plan exception episode. Clinical values are nonadditive: one prepared review may resolve several task rows. Task 42 replaces the changed-plan route when the plan is truly unchanged. Task 52 applies to a later revision and is not part of the initial-plan total.

Participation and source setup#

Puerto Rico phase subtotal: 2 / 4 / 8 minutes. Philippines starred values are possible slices of one shared exception episode and are never added row by row.

Beneficiary direction#

Puerto Rico phase subtotal: 7 / 15 / 25 minutes. Philippines starred values are possible slices of one shared exception episode and are never added row by row.

Options and decisions#

Puerto Rico phase subtotal: 0 / 9 / 20 minutes. The low route presents the prepared comparison, captures a clear authenticated choice, and performs clean practical matching without buying a navigator validation. Philippines starred values are possible slices of one shared exception episode and are never added row by row.

Draft and beneficiary review#

Puerto Rico phase subtotal: 0 / 4 / 18 minutes on the changed-plan route; the low route uses an approved direct section-by-section review, and task 42 is the alternative no-change route. Philippines starred values are possible slices of one shared exception episode and are never added row by row.

Incorporation, distribution, and first follow-up#

Puerto Rico phase subtotal: 2 / 2 / 4 minutes on the initial route; task 52 is later-revision work. Philippines starred values are possible slices of one shared exception episode and are never added row by row.

Detailed Care Plan task cards#

Each card answers what happens, when it happens, what information may support it, what Proxi can prepare or perform, which human act remains, the low/typical/high human-minute allocation, what proves completion, and what must not be mistaken for completion.

1. Open the care-planning work#

When. Open after receipt of a completed comprehensive assessment for the initial plan; after an authenticated beneficiary request for review or change; or after a distinct event reports changed circumstances, goals, preferences, or needs. A possible immediate safety signal starts the separate safety pathway before routine planning continues.

What and how much. Create exactly one care-planning episode for the triggering event, linked to the beneficiary, trigger type, source, observed/received times, applicable prior plan, and reason for initial planning, review, or revision. Exact duplicate triggers attach provenance to existing work; they do not create duplicate planning episodes.

Data. Direct: a completed comprehensive-assessment handoff; an authenticated beneficiary or decision-specific authorized-representative request; or an authoritative care-team event explicitly requiring plan review. Corroborating: a new encounter, transition, assessment delta, caregiver report, or provider message that may indicate change but still needs source/subject validation. Identity-only: message headers, routing metadata, or an unmatched alert identify a candidate event but not the correct beneficiary or materiality.

Potential Proxi work. Detect eligible triggers, match the beneficiary and prior plan, deduplicate exact repeats, classify the stated trigger without deciding its materiality, assemble the work packet, and put it on the assigned care-plan queue. Preserve every source and timestamp.

Human role. No Proxi human is needed on the clean path. L1 resolves an unmatched administrative source under the shared exception. L2 handles an authenticated change request whose meaning remains unclear. L3/L4 handles a clinical or safety signal within scope. The beneficiary or caregiver may supply the trigger; their time is not Proxi labor.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 1*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. One case/episode identifier; beneficiary identity; trigger source, type, and timestamps; prior-plan link if any; duplicate/new disposition; and assigned work owner. If the trigger is an exact duplicate, retain the linked prior event and duplicate proof.

What does not prove completion. A generic queue alert, an unmatched message, assessment upload without correct-beneficiary linkage, AI “material change” classification, or a second case created from the same trigger.

Edge cases/open decisions. CP-E11, CP-E15, CP-E21, O-004, and O-010. Minimum tests: missing source; stale earlier event; conflicting simultaneous triggers; exact duplicate; late older fact; corrected fact; post-terminal material event; immediate safety content.

2. Confirm the beneficiary#

When. Complete before asking person-specific questions, displaying protected information, inviting another participant, or attaching source records to the planning episode. Repeat when identifiers or contact information conflict or have changed.

What and how much. Establish one matched beneficiary and one current contact profile sufficient for the intended interaction. Resolve or explicitly hold every material identity conflict; do not merge two people because names or addresses resemble one another.

Data. Direct: authenticated beneficiary confirmation plus the Participant’s authoritative enrollment/EHR identifiers; or a representative acting within established decision-specific authority. Corroborating: payer demographics, prior authenticated encounters, prior successful contact, and current address/phone from another authoritative source. Identity-only: caller ID, device ownership, emergency-contact fields, email display name, and demographic similarity.

Potential Proxi work. Match identifiers, prefill current known contact data, detect collisions and changed demographics, ask a bounded verification set, and block the interaction when the evidence does not uniquely identify the beneficiary. AI may transcribe responses; it may not waive an identity conflict.

Human role. No Proxi human is needed for an exact clean match. The beneficiary or authorized representative confirms the information. L1 resolves administrative discrepancies under the shared exception. Privacy/records staff handles unresolved or suspected cross-person contamination; L2 is not the default identity adjudicator.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Matched beneficiary identifiers; authentication method; current confirmed contact fields and date; source attribution; confirming person; representative authority when applicable; and disposition of every conflict.

What does not prove completion. Name and date-of-birth similarity alone, possession of a phone, caller ID, an emergency-contact relationship, a payer record alone, or an AI confidence score.

Edge cases/open decisions. CP-E01, CP-E02, CP-E15, CP-E20, and O-004. Minimum tests: missing identifier; stale contact; conflicting addresses; duplicate names; wrong-beneficiary source; late demographic update; deceased/disenrolled signal; failed authentication.

3. Confirm who leads the plan#

When. At the initial planning episode and whenever participation ability, requested support, or representative information changes or is challenged.

What and how much. Establish one current participation record stating that the beneficiary leads the initial plan and revisions, how the beneficiary wants to participate, what support is requested, and whether a representative is proposed or established for any defined decision. Keep beneficiary participation, communication assistance, and legal decision authority as separate facts.

Data. Direct: the authenticated beneficiary’s current participation choice; and, when applicable, current representative documentation whose identity, scope, conditions, effective/expiry dates, and decision applicability are established. Corroborating: prior care-plan notes, EHR representative indicators, and observed need for communication support. Identity-only: caregiver labels, emergency-contact status, portal proxy access, next-of-kin fields, and physical accompaniment.

Potential Proxi work. Present the approved plain-language explanation; offer communication/support choices; capture the beneficiary’s words; surface existing representative records; compare scope and dates; and route missing, partial, expired, conditional, or conflicting authority. Proxi may not decide capacity, legal authority, or who controls a disputed decision.

Human role. The beneficiary leads and chooses the participation method. L2 spends the calibrated two minutes confirming understanding and the participation record within the combined planning conversation. A properly authorized representative participates only within established scope. L2 verifies facts but privacy/legal/clinical authority resolves disputed authority or capacity under O-011.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 1 / 1 / 2; Separate U.S. L3/L4 clinical: 0 / 0 / 5–15.

Completion evidence. Beneficiary-led status; authenticated statement; participation method and requested supports; representative status; exact documentary source and scope if applicable; confirmation date; and any unresolved authority issue held separately.

What does not prove completion. Dementia diagnosis, caregiver attendance, emergency-contact status, portal proxy access, a historical representative flag, AI inference of capacity, or silence.

Edge cases/open decisions. CP-E03, CP-E05, CP-E06, CP-E10, O-011, and O-PA-001. Minimum tests: capable beneficiary; requested assistance without transferred authority; unsupported incapacity assertion; expired/partial/conditional authority; multiple representatives; later restoration or correction.

4. Ask whether to involve a caregiver#

When. After beneficiary leadership is established and before collecting caregiver input, inviting the caregiver, or sharing plan information. Reconfirm when the beneficiary changes the choice, the caregiver changes, or permissions are narrowed or revoked.

What and how much. Produce one current participation profile for each proposed caregiver: involved or not involved; allowed planning topics; allowed contact channels and timing; whether the caregiver may contribute information, receive information, or both; and any restrictions. A participation preference is not itself the complete legal disclosure basis.

Data. Direct: the authenticated beneficiary’s current caregiver-participation choices; or a representative’s choice within established scope. Corroborating: prior participation, caregiver-assessment linkage, and prior successful joint contacts. Identity-only: emergency-contact fields, caregiver lists, shared address, phone ownership, portal access, and family relationship.

Potential Proxi work. Ask structured questions, capture free-text limits, read the profile back, prevent invitations outside the selected scope, retain dates and source, and detect conflict with older preferences or queued communications. AI cannot infer permission from family relationship or historical involvement.

Human role. The beneficiary makes the choice. L2 spends the calibrated two minutes confirming the profile during the combined conversation. The caregiver supplies contact information or indicates willingness only after the beneficiary permits involvement. L2 handles relationship-sensitive disagreement; privacy/legal resolves an unclear disclosure basis or disputed representative authority.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 1 / 1 / 2; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Caregiver identity; beneficiary choice; allowed roles/topics/channels; disclosure/participation distinction; confirming person and date; and any unresolved permission or authority matter.

What does not prove completion. Caregiver presence, shared residence, prior receipt of records, portal access, emergency-contact designation, lack of beneficiary objection, or the caregiver answering the phone.

Edge cases/open decisions. CP-E03, CP-E04, CP-E05, CP-E06, CP-E07, CP-E14, O-011, O-013, and later distribution decisions O-012/O-014. Minimum tests: no caregiver; unknown caregiver; caregiver declined; partial-topic permission; beneficiary/caregiver disagreement; multiple caregivers; revoked permission; late change after invitation.

5. Capture communication needs#

When. Before the planning exchange and whenever language, hearing, vision, cognition, literacy, device access, preferred channel, or safe-contact conditions change.

What and how much. Create one current communication profile covering preferred language; communication modality; accessible format; reading/processing accommodations; interpreter or assistive-support need; preferred and fallback channels; safe times; and message/voicemail restrictions. Record unknowns rather than defaulting to English, portal, or caregiver mediation.

Data. Direct: the beneficiary’s or decision-specific authorized representative’s current stated needs and choices; and a qualified accessibility/interpreter assessment when one exists. Corroborating: prior successful modality, prior interpreter use, documented accommodation, and observed failed comprehension. Identity-only: demographic language fields, device settings, diagnosis, address, and prior message-delivery metadata.

Potential Proxi work. Offer text, voice, audio-visual, large-text, plain-language, and approved language options; retain original and translated content; test the selected channel; arrange approved interpreter/accessibility support; and flag where automation cannot provide usable access. AI translation is a draft unless the approved policy establishes its permitted use.

Human role. The beneficiary states what is usable. L2 uses up to one minute in the typical or high combined interaction to check that the profile is complete and usable; no separate L2 minute is allocated on the low route. A qualified interpreter or accessibility professional participates when required. L1 handles bounded contact cleanup only; L2 does not substitute as an unqualified interpreter.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 1 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Current profile; source and date; tested or confirmed primary/fallback mode; required accommodation; interpreter/support status; restrictions; and failed-channel disposition.

What does not prove completion. A stored preferred-language field, successful SMS delivery, device ownership, caregiver interpretation, a translated document without usability confirmation, or absence of a recorded accommodation.

Edge cases/open decisions. CP-E02, CP-E03, CP-E06, CP-E19, O-004, and O-011. Minimum tests: stale language; conflicting modality; low literacy; hearing/vision limitation; inaccessible portal; interpreter unavailable; duplicate profiles; late accommodation request.

6. Schedule the planning conversation#

When. Once required participants, communication needs, and care-team availability are known. Reschedule after a routine conflict, failed connection, or beneficiary request without reopening unrelated completed tasks.

What and how much. Create one appointment record with date/time/timezone, modality/location, required and invited participants, accommodation/interpreter needs, access instructions, reminders, and a fallback/rescheduling path. Do not invite a caregiver or representative beyond the confirmed participation profile.

Data. Direct: care-team availability; beneficiary-selected time/method; invited-person status; and confirmed accommodation resources. Corroborating: prior scheduling preferences, calendar availability, and prior successful contact windows. Identity-only: phone, email, address, and calendar identity do not establish permission or attendance.

Potential Proxi work. Offer eligible times, reconcile calendars and timezones, send approved invitations/reminders, include accessibility instructions, confirm connection details, and automatically reschedule routine conflicts within beneficiary-approved constraints.

Human role. No Proxi human is needed on the clean path. The beneficiary and invited people select and attend. L1 uses the shared exception for manual office/calendar or failed-contact resolution. L2 enters only when scheduling reveals distress, disagreement, authority, accessibility, or relationship-sensitive constraints.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 3*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Confirmed appointment; participant list and permission basis; timezone/modality; accommodation status; sent/received confirmations where available; and fallback instructions.

What does not prove completion. A calendar invitation alone, an offered time, an unread reminder, caregiver acceptance without beneficiary permission, or a link sent through an inaccessible channel.

Edge cases/open decisions. CP-E14, CP-E19, O-004, and O-011. Minimum tests: no common time; duplicate booking; timezone error; failed reminder; invited caregiver after revocation; interpreter unavailable; late cancellation; no response.

7. Bring in the comprehensive assessment#

When. Before plan-content development and again when a later approved assessment supersedes or materially corrects the source used for the plan.

What and how much. Assemble one current comprehensive-assessment package linked to the beneficiary and planning episode, including all completed required domains, source documents, dates, authors/performers, conclusions, open items, and corrections. Preserve domain-level unknown, declined, missing, stale, and conflicted states. Include the assessment's plan-consumable clinical direction: the clinician-authored synthesis, accepted recommendations, the professional authority behind each clinical element, and any stated recheck or monitoring window. If that direction is absent, return the gap to the assessment's clinical author; the planning team must not reconstruct clinical intent from raw findings.

Data. Direct: the finalized comprehensive assessment and its source records, with beneficiary, date, performer/author, and approval status. Corroborating: recent clinical notes, encounter data, prior assessment, and beneficiary/caregiver corrections that may trigger source review. Identity-only: document title, folder location, or unmatched assessment record does not prove correct beneficiary, completeness, recency, or approval.

Potential Proxi work. Retrieve structured fields and source notes; link every item to provenance; compare required-domain coverage; identify missing/stale/conflicted entries; display without changing clinical meaning; and route each exception to the fact owner.

Human role. No Proxi human is needed for clean retrieval. L1 uses the shared exception to pursue existing missing records. L2 resolves beneficiary/household facts. L3/L4 or the original clinical author resolves clinical findings. Proxi does not reinterpret the assessment inside care planning.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 3*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Assessment identifier; beneficiary link; completion/approval status; date; performer/author; list of required domains; source links; correction history; and explicit disposition of every missing/stale/conflicted domain.

What does not prove completion. File presence, an assessment score without source context, an unsigned draft, an old assessment, a copied summary without provenance, or AI-generated “complete” status.

Edge cases/open decisions. CP-E01, CP-E02, CP-E08, CP-E20, O-003, and O-015. Minimum tests: missing domain; stale assessment; wrong beneficiary; draft/final conflict; duplicate corrected record; late amendment; clinical approval invalidated.

8. Bring in the social-needs screening#

When. Before developing social/community plan elements and whenever a new screening, beneficiary correction, or changed social need is received.

What and how much. Assemble one current HRSN result set with every screened item’s response/status, date, source, declined/unanswered state, identified need, existing response, and still-open action. Preserve the beneficiary’s wording and do not convert a declined or unanswered item to “no need.”

Data. Direct: the completed HRSN screening with item-level responses and date; and authenticated beneficiary/authorized-representative corrections. Corroborating: prior referrals, service records, utilization patterns, navigator notes, and earlier screenings. Identity-only: ZIP-code deprivation measures, payer segment, diagnosis, or resource-directory matches do not establish the person’s need.

Potential Proxi work. Import item-level responses, retain source and date, group needs without erasing individual items, distinguish answered/declined/unanswered, detect stale/conflicting results, and connect each identified need to later care-plan work.

Human role. No Proxi human is needed for clean retrieval. L1 may pursue an existing missing source under the shared exception. L2 asks only the unresolved person-centered follow-up; it does not infer a need from neighborhood or utilization data. Clinical or safeguarding signals route to L3/L4 under approved policy.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Screening identifier/date; item-level responses and source; explicit declined/unanswered states; correction history; identified-needs list; and linked open actions or truthful no-action status.

What does not prove completion. A completed-screening checkbox, a neighborhood index, claims pattern, prior referral, generic social-risk score, or AI-filled unanswered items.

Edge cases/open decisions. CP-E01, CP-E02, CP-E11, CP-E20, O-004, and O-015. Minimum tests: unanswered item; declined item; stale screening; conflicting correction; duplicate screening; late new need; safety/abuse signal.

9. Bring in home-visit findings#

When. Before developing home-safety, function, environmental, social, or behavioral plan elements and whenever a corrected or later home-visit record arrives.

What and how much. Assemble one current home-visit finding set covering the recorded home environment, navigation, ADLs/function, and other environmental/social/behavioral observations relevant to the plan, with each item’s source, date, observer, and status. This task retrieves findings; it does not perform or retroactively invent the home visit.

Data. Direct: the completed, beneficiary-matched home-visit record with performer, date, location/context, observations, and any approved conclusions; plus authenticated beneficiary corrections to factual household information. Corroborating: caregiver report, clinical/functional notes, prior home records, and authenticated photos or video where permitted. Identity-only: home address, property listing, map imagery, residence type, or device location does not establish current in-home conditions.

Potential Proxi work. Retrieve and organize each observation; preserve author language; identify unaddressed, stale, missing, or contradicted findings; link source evidence; and route clinical/safety interpretation rather than generating it.

Human role. No Proxi human is needed for clean retrieval. L1 pursues an existing record under the shared exception. L2 resolves household facts and practical preferences. L3/L4 handles clinical or safety interpretation within authority. A remote data source does not let Proxi claim a required home visit occurred.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Home-visit record identifier; beneficiary; performer; date/context; itemized findings; source links; correction/conflict disposition; and explicit missing/stale status where applicable.

What does not prove completion. An address, property photograph, prior residence record, caregiver statement alone, an unverified video, a generic home-safety checklist, or AI reconstruction of observations.

Edge cases/open decisions. CP-E01, CP-E02, CP-E11, CP-E20, O-003, and O-015. Minimum tests: missing visit record; stale findings after move; wrong address; conflicting household report; duplicate corrected record; late safety finding; absent performer identity.

10. List current clinicians#

When. During initial planning and whenever the beneficiary reports a provider change, a new provider appears in current records, a prior relationship ends, or provider sources conflict.

What and how much. Produce one reconciled current-clinician roster covering the PCP, behavioral health provider, specialists, and other clinicians involved in care. For each candidate record the individual, practice, role, relationship status, source, as-of date, and uncertainty. “No current clinician” and “unresolved” are distinct truthful outcomes.

Data. Direct: the current statement of the capable beneficiary or decision-specific authorized representative; authenticated practice confirmation; or a current authoritative clinical/team record explicitly naming the relationship. A state/MCO PCP field is direct only for its administrative assignment (SRC-NC-MEDICAID-PCP-ASSIGNMENT). Corroborating: beneficiary-authorized payer/claims/encounter data (SRC-CMS-PATIENT-ACCESS-API, SRC-CMS-BLUE-BUTTON), referrals, orders, recent visits, and HIE records. Identity-only: NPI/NPPES and payer/public directories (SRC-CMS-NPI-FACT-SHEET, SRC-CMS-PROVIDER-DIRECTORY-API), licensure listings, addresses, phone numbers, and websites.

Potential Proxi work. Pull candidates from authorized sources; normalize names, NPIs, practices, aliases, and roles; keep administrative assignment, beneficiary-identified provider, observed utilization, and GUIDE-team membership separate; present candidates for confirmation; retain corrections; and route unresolved conflicts.

Human role. The beneficiary or authorized representative confirms lived relationships. L2 uses up to one minute only in the high combined interaction to resolve a substantive roster or relationship gap; no separate L2 minute is allocated on the low or typical route. L1 uses the shared exception for provider identity/contact verification. L3/L4 does not decide relationship identity merely because a clinical credential is present; clinical recommendation conflicts are separate work.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 3*; Puerto Rico L2 navigator: 0 / 0 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Each current clinician and practice; role; NPI when available; relationship status; source/as-of date; confirming person; and disposition of every duplicate, uncertain, ended, or conflicting candidate.

What does not prove completion. Most frequent billing provider, Medicaid assignment alone, prescribing provider, hospital attending, directory listing, NPI match, old EHR field, referral order, or model-generated “likely PCP.”

Edge cases/open decisions. CP-E01, CP-E02, CP-E09, CP-E20, O-003, and O-015. Minimum tests: no PCP; assignment/treating conflict; same-name providers; changed practice; stale claim; duplicate NPI candidate; late new-provider record; conflicting PCP/specialist positions.

11. List current services and supports#

When. During initial planning and whenever a service starts, stops, changes, is declined, becomes unavailable, or conflicts with another source.

What and how much. Produce one itemized current inventory of formal community services, Medicaid/HCBS and other programs, paid in-home supports, informal supports, and other recurring help. For every item record provider/person, service, frequency/intensity when known, status, start/end dates, payer/funding, beneficiary-reported usefulness, and verification source. Keep active, authorized-not-started, waitlisted, referred, ended, declined, and unknown statuses separate.

Data. Direct only for the status recorded: an authenticated beneficiary/authorized-representative report directly establishes the person’s report; an authorization/enrollment record establishes authorization or enrollment but not service start; current provider confirmation or a service record can establish the provider’s recorded delivery status; and an informal supporter’s authenticated statement establishes what that person reports providing. Corroborating: claims/encounters, referral records, old care plans, case-manager notes, payment records, and scheduled visits. Identity-only: resource-directory listing, geographic match, broad eligibility result, provider marketing, and contact information do not establish enrollment, availability, or service receipt.

Potential Proxi work. Prepopulate candidates, normalize providers/programs, compare dates and statuses, ask item-specific confirmation questions, preserve informal supports separately, detect possible duplication/gaps, and link each active/open item to later plan work. AI may flag a possible gap; it may not decide clinical suitability, eligibility, or payment responsibility.

Human role. The beneficiary/caregiver confirms lived service use; external programs confirm their own status. L2 uses up to one minute only in the high combined interaction to resolve a substantive practical-accuracy exception; no separate L2 minute is allocated on the low or typical route. L1 uses the shared exception for administrative status verification. Clinical suitability routes to L3/L4; coverage or program disputes route to the authorized payer/program owner.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 0 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Itemized inventory with source/date, actor/provider, status, frequency/intensity when known, funding, confirmation, discrepancies, and explicit no-service/unknown outcomes.

What does not prove completion. A referral, directory match, eligibility result, authorization without start, claim without current status, calendar entry, generic caregiver label, or “closed” case status.

Edge cases/open decisions. CP-E01, CP-E02, CP-E07, CP-E20, O-004, and O-015. Minimum tests: service authorized but not started; ended service with late claim; duplicate providers; informal/formal overlap; conflicting frequency; no caregiver; waitlist; late service start.

12. Bring in caregiver-assessment findings#

When. When an involved caregiver exists and a caregiver assessment is available; repeat when a new assessment or authenticated correction supersedes the prior source. Use the no-caregiver/unknown-caregiver branch when applicable rather than creating a fictitious caregiver record.

What and how much. Assemble one distinct current assessment snapshot per involved caregiver covering stated ability, willingness, knowledge, needs, supports, well-being, stress, and challenges. Keep caregiver information separate from beneficiary facts and distinguish caregiver self-report from observations about the beneficiary.

Data. Direct: the completed caregiver assessment attributed to the correct caregiver, with date and source, plus that caregiver’s authenticated corrections. Corroborating: prior caregiver assessment, navigator notes, accepted support services, and observed participation. Identity-only: caregiver/contact relationship, shared address, portal access, and emergency-contact status do not establish willingness, capacity, stress, or assessment completion.

Potential Proxi work. Retrieve the correct assessment; maintain caregiver/beneficiary separation; compare dates and changed responses; flag missing, stale, or conflicting items; and link stated caregiver needs to later education/support tasks without diagnosing or overriding the caregiver.

Human role. No Proxi human is needed for clean retrieval. The caregiver corrects their own information. L1 pursues an existing record under the shared exception. L2 resolves substantive missing facts or differing perspectives while preserving each speaker. L3/L4 handles clinical, behavioral, safeguarding, or crisis issues within authority.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Caregiver identity; participation basis; assessment identifier/date/source; item-level results; correction history; beneficiary/caregiver separation; and explicit no-caregiver, caregiver-unknown, unavailable, or stale status where applicable.

What does not prove completion. Caregiver contact information, prior involvement, a generic burden score without source, beneficiary report about the caregiver alone, or an AI-inferred caregiver need.

Edge cases/open decisions. CP-E01, CP-E02, CP-E03, CP-E04, CP-E05, CP-E06, CP-E07, CP-E11, O-011, O-013, and O-015. Minimum tests: no caregiver; unknown caregiver; multiple caregivers; stale assessment; beneficiary/caregiver disagreement; wrong caregiver linkage; late correction; distress/safeguarding signal.

13. Bring in the medication summary#

When. Before medication-related plan elements are drafted and whenever the authorized medication workflow issues a new clinician-reviewed summary, correction, or unresolved action that affects the plan.

What and how much. Retrieve one current clinician-reviewed medication summary for the planning episode, including its author/authority, date, source medication state, unresolved discrepancies/actions, and any plan-relevant practical support needs. Do not reconcile, recommend, discontinue, or change medication inside this task.

Data. Direct for this task: the exact current medication summary produced or approved by the role authorized under the future O-028 matrix, with source, date, and scope. Corroborating only: payer/pharmacy claims, dispense history, bottle photographs, beneficiary/caregiver reports, medication lists, and transition records; these may reveal discrepancy but do not replace the clinician-reviewed summary. Identity-only: RxNorm/drug-directory identity, NDC lookup, medication name matching, and pharmacy contact data normalize entities but do not establish the current regimen, possession, use, or clinical appropriateness.

Potential Proxi work. Retrieve the exact summary and source links; verify beneficiary, date, and content identity; display unresolved actions; compare for a newer corrected summary; and route discrepancies to the medication workflow. AI may summarize approved content without changing meaning and may not select which source is clinically correct.

Human role. No Proxi human is needed for clean retrieval. L1 may pursue existing records under the shared exception but stops on clinical content. L2 may authenticate practical use reports but does not reconcile. The prescribing-authority clinician or other role settled under O-028 owns medication decisions.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 3*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Exact summary identifier/content; beneficiary; clinical author/approver and authority; date; source medication events; unresolved actions and owners; and link to the medication case/workflow.

What does not prove completion. Claims, a dispense event, bottle photo, patient-reported use, EHR list, discharge list, AI-generated reconciliation, or “medications reviewed” checkbox without author and source.

Edge cases/open decisions. CP-E02, CP-E08, CP-E09, CP-E20, O-003, O-015, O-028, and O-030. Minimum tests: stale summary; wrong beneficiary or wrong content; conflicting discharge list; dispense after stop; no clinical author; duplicate corrected summary; late outside order; unresolved medication action.

14. Review the prior plan#

When. When a prior plan exists and the current episode is an annual review, beneficiary-requested review, or event-driven revision.

What and how much. Produce one item-by-item delta review covering every prior goal, action, owner, due/follow-up item, refusal, unresolved matter, and distribution dependency. Classify only source-supported facts such as completed, still active, declined, not obtained, changed, or unknown. Capture beneficiary-reported helpful/not-helpful outcomes without turning them into an unauthorised clinical effectiveness conclusion.

Data. Direct: the exact prior approved/current plan; linked action records and objective outcome evidence; authenticated beneficiary report of lived outcome; and authorized clinician/provider disposition for clinical content. Corroborating: claims, appointments, delivery acknowledgments, provider notes, service records, and caregiver input permitted by the beneficiary. Identity-only: a closed task flag, referral record, calendar entry, message delivery, or matching service/provider name does not establish action completion or benefit.

Potential Proxi work. Compare the prior plan to current evidence; build a side-by-side delta; link each proposed status/change to its source; detect unsupported or conflicting statuses; preserve refusals and unknowns; and draft questions for the beneficiary or responsible professional.

Human role. L2 uses up to one minute in the typical or high combined interaction to confirm lived outcomes and any nonclinical proposed status; no separate L2 minute is allocated on the low route. The beneficiary reports what happened and what mattered. L3/L4 or the responsible external clinician evaluates clinical effectiveness, safety, or changed treatment. L1 may pursue existing administrative evidence under the shared exception.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 1 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 4.

Completion evidence. Prior-plan identifier; item-level status/delta; source for every status; beneficiary-confirmed lived outcomes; unresolved/conflicting items; and linked owners/questions for later tasks.

What does not prove completion. Task closure, referral sent, appointment scheduled, claim paid, message delivered, no complaint, elapsed time, or AI determination that an intervention was ineffective.

Edge cases/open decisions. CP-E02, CP-E08, CP-E09, CP-E10, CP-E20, CP-E21, O-003, and O-010. Minimum tests: no prior plan; stale or superseded prior plan; action marked complete without evidence; beneficiary/provider conflict; clinical change; refusal; duplicate/late outcome; genuine no-change review.

15. Ask what matters now#

When. During the beneficiary-led planning conversation after communication supports and the current evidence packet are ready; repeat when the beneficiary requests review or reports changed circumstances, goals, preferences, or needs.

What and how much. Capture at least one current, source-faithful “what matters” statement in the beneficiary’s own words, plus any stated priority ordering and context. Preserve the original response alongside any plain-language summary. “I do not know,” decline, inability to answer now, and conflicting statements are truthful states requiring different follow-up—not blank fields to auto-fill.

Data. Direct: the authenticated beneficiary’s spoken, typed, selected, or otherwise accessible response. A representative’s response is direct only within established decision-specific authority and must not erase a capable beneficiary’s statement. Corroborating: invited caregiver input, prior plan, life-history notes, and observed routines may help prompt or clarify. Identity-only: diagnosis, demographics, utilization, residence, caregiver opinion, and AI prediction do not establish what matters to the beneficiary.

Potential Proxi work. Conduct an accessible text/voice interview; use approved neutral prompts; preserve transcript/audio/text provenance; summarize without replacement; read back or display the summary; accept correction; and detect contradiction, distress, safety content, or inability to engage for routing.

Human role. The beneficiary supplies the direction. L2 uses two minutes in the low interaction, three minutes in the typical interaction, and up to five minutes in the high interaction to facilitate, confirm meaning, and maintain the human navigator relationship. A caregiver contributes only when invited. L2 does not infer capacity or decide which value should prevail; clinical/safety content routes appropriately.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 2 / 3 / 5; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Original authenticated response; source/mode/date; beneficiary-confirmed summary or recorded correction/decline; priority/context where stated; and any unresolved conflict or routed signal.

What does not prove completion. AI-generated preference, caregiver-only answer without established authority, prior-plan statement copied forward, generic values checklist, silence, diagnosis-based assumption, or completed text fields without provenance.

Edge cases/open decisions. CP-E03, CP-E04, CP-E05, CP-E06, CP-E10, CP-E11, O-003, O-004, and O-011. Minimum tests: low-confidence transcript; ambiguous answer; caregiver contradiction; representative overreach; decline; inaccessible modality; immediate safety content; late beneficiary correction.

16. Identify the beneficiary's goals#

When. After “what matters” is captured and whenever the beneficiary changes a desired outcome or current evidence makes a goal impossible, completed, unsafe, or in need of professional review.

What and how much. Create one beneficiary-approved goal list. Each goal records the desired outcome in understandable language, the life/health context the beneficiary gives it, priority, and any timeframe or success description the beneficiary accepts. Do not force a generic SMART formulation or convert a clinician recommendation into the beneficiary’s goal without acceptance.

Data. Direct: the authenticated beneficiary’s explicit acceptance of each goal and wording; or a representative’s choice within established decision-specific authority. Corroborating: “what matters” statements, assessment findings, prior goals, invited caregiver suggestions, and clinician-proposed options. Identity-only: goal templates, diagnoses, risk scores, utilization, or AI-generated priorities do not establish the beneficiary’s chosen goal.

Potential Proxi work. Suggest plain-language formulations; link drafts to the beneficiary’s source statement; identify vague, duplicated, or internally conflicting wording; show alternatives; retain rejected drafts; and route a goal that entails diagnosis, treatment, restriction, medication, or clinical safety judgment.

Human role. The beneficiary chooses each goal. L2 uses two minutes in the low interaction, three minutes in the typical interaction, and up to five minutes in the high interaction to facilitate and confirm wording. Caregiver input is included only when invited. L3/L4 or the relevant external clinician supplies clinical options/constraints without taking over the beneficiary’s choice.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 2 / 3 / 5; Separate U.S. L3/L4 clinical: 0 / 0 / 4.

Completion evidence. Goal list; source statement link; beneficiary acceptance and date; priority/timeframe/success description where chosen; rejected or unresolved alternatives; and clinical-review status for affected goals.

What does not prove completion. Assessment recommendation, diagnosis-linked template, clinician’s goal alone, caregiver preference, AI-generated SMART goal, unsigned draft, or lack of objection.

Edge cases/open decisions. CP-E03, CP-E04, CP-E05, CP-E06, CP-E08, CP-E09, CP-E10, CP-E11, O-003, O-004, and O-011. Minimum tests: vague goal; mutually conflicting goals; clinical-treatment goal; caregiver-selected goal; beneficiary refusal; changing goal; capacity/authority uncertainty; safety conflict.

17. Identify strengths#

When. During the beneficiary-led planning conversation after current goals are known and whenever abilities, relationships, routines, resources, or supports materially change.

What and how much. Create one strengths inventory covering abilities, routines, relationships, coping strategies, cultural/community resources, technology, finances/resources when volunteered and relevant, and existing formal/informal supports that the beneficiary believes can help. Preserve source and do not convert an assessment deficit into an assumed strength.

Data. Direct: authenticated beneficiary statements and, when invited, caregiver/supporter statements about abilities or support they actually provide; verified current resources and relationships. Corroborating: functional assessment findings, prior successes, care-team observations, and prior plan outcomes. Identity-only: diagnosis, functional score, contact list, residence, income proxy, or generic community-resource availability does not establish a usable strength.

Potential Proxi work. Prompt for commonly overlooked strengths; organize responses without flattening the beneficiary’s language; link strengths to relevant goals as a draft; compare with current facts; and flag conflicts such as a listed support who is no longer available.

Human role. The beneficiary identifies strengths; invited caregivers/supporters may contribute. L2 uses one minute in the low interaction, two minutes in the typical interaction, and up to three minutes in the high interaction to confirm practical accuracy. L1 may verify an administrative support fact under the shared exception. Clinical interpretation is not part of this task.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 1 / 2 / 3; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Itemized strengths with speaker/source/date; beneficiary confirmation; availability/currentness; any linked goal; and explicit conflict or unknown status.

What does not prove completion. Generic strengths checklist, inferred resilience, a contact name, prior support, functional score, resource-directory match, or AI-generated positive framing.

Edge cases/open decisions. CP-E02, CP-E03, CP-E04, CP-E05, CP-E06, CP-E07, CP-E20, O-011, and O-013. Minimum tests: supporter no longer available; beneficiary/caregiver disagreement; no caregiver; stale strength; duplicate source; late loss/gain of support; inaccessible prompt.

18. Identify preferences#

When. During initial planning and whenever a preference changes, becomes relevant to a new decision, conflicts with another person’s view, or is affected by changed authority, circumstances, or options.

What and how much. Create one itemized current preference profile across care, daily routines, providers, setting, communication, risk/choice tradeoffs, and who should participate. Each preference records the exact subject/decision scope, speaker, date, strength or conditionality when stated, and whether it is accepted, declined, unknown, conflicted, or superseded. Keep preference separate from consent, legal authority, clinical appropriateness, and actual implementation.

Data. Direct: the authenticated beneficiary’s specific current choice; or a representative’s choice within established decision-specific authority. Corroborating: prior preferences, repeated behavior, invited caregiver input, and existing plan statements can prompt confirmation or expose conflict. Identity-only: EHR defaults, residence, payer network, device settings, family relationship, and AI prediction do not establish preference.

Potential Proxi work. Ask structured and free-text questions; preserve verbatim responses; compare with prior choices; distinguish conditional preference from absolute refusal; surface contradictions; read back the profile; accept corrections; and prevent an unconfirmed candidate from becoming plan content or disclosure authority.

Human role. The beneficiary chooses. L2 uses one minute in the low or typical combined interaction and up to four minutes in the high interaction to confirm scope, conditions, and conflicts. A caregiver contributes only when invited; a representative acts only within established scope. L2 preserves dissent but does not decide capacity, legal control, or clinical appropriateness.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 1 / 1 / 4; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Itemized preferences; exact scope; original statement; speaker/authentication/authority; date; confirmation; conditionality; conflict/supersession history; and any separate clinical, consent, permission, or implementation issue.

What does not prove completion. Historical preference, caregiver assumption, observed behavior alone, EHR default, lack of objection, AI prediction, residence, payer limitation, or a permission form that does not state the preference.

Edge cases/open decisions. CP-E03, CP-E04, CP-E05, CP-E06, CP-E10, CP-E14, CP-E20, O-003, O-011, and O-015. Minimum tests: low-confidence extraction; contradictory current statements; caregiver disagreement; representative overreach; partial/conditional preference; revocation; late correction; preference requesting clinically unsafe or unavailable action.

19. Identify unmet needs and concerns#

When. Perform after the current comprehensive assessment, social-needs screening, home-visit findings, prior-plan comparison, beneficiary goals, strengths, and preferences have been assembled. Repeat for an initial plan and whenever new or corrected findings, changed circumstances, a beneficiary request, or an unresolved prior need reopens the affected plan element.

What and how much. Review every applicable care-plan domain with the beneficiary and assign each identified item one truthful disposition: currently met, partially met, unmet, declined, not applicable with a recorded basis, or unresolved. Record the beneficiary's concern in their own words, the affected goal or daily-life consequence, and the source finding. Categorization organizes work; it must not create a diagnosis, severity rating, urgency class, or clinical interpretation.

Data.

• A current authenticated beneficiary statement directly proves what the beneficiary reports, experiences, wants addressed, or says is already resolved. It does not by itself prove a diagnosis or clinical severity.
• A caregiver statement directly proves that caregiver's observations and constraints only. It is not automatically the beneficiary's view and is not a clinical finding.
• A clinician-authored assessment directly supports the exact finding and date recorded within that clinician's scope. It does not prove the need remains current, that the beneficiary accepts the interpretation, or that an action occurred.
• Claims, encounters, sensor data, missed visits, and prior service records can corroborate a possible gap. They do not independently classify a current unmet need.
• An AI-extracted candidate remains a candidate until the underlying source and required human or beneficiary confirmation are preserved.

Potential Proxi work. Assemble all assessment domains and prior-plan actions; compare current services with identified needs; show the beneficiary a plain-language, source-linked list; capture corrections and free-text concerns; retain conflicting statements; and expose missing or stale inputs. Proxi may draft categories and summaries. It may not infer that silence means “met,” assign clinical urgency, or decide which clinical issue matters most.

Human role. The beneficiary supplies and authenticates lived priorities and corrections. The L2 Puerto Rico navigator conducts the person-centered review, ensures no assessment domain disappears, and separates beneficiary, caregiver, and clinician statements. L3/L4/L5 interprets a clinical, behavioral, medication, capacity, or safety finding only when that interpretation is actually required.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 1 / 4 / 4; Separate U.S. L3/L4 clinical: 0 / 0 / 4.

Completion evidence. A dated, source-linked domain inventory; each identified item and its disposition; the beneficiary's authenticated corrections; distinct caregiver and clinician inputs; every unresolved item retained as unresolved; and the identity of the navigator who completed the review.

What does not prove completion. An AI-generated problem list, an abnormal result, a diagnosis code, a risk score, a stale assessment, a copied prior-plan need, an empty field, or lack of beneficiary objection.

Edge cases/open decisions. CP-E01, CP-E02, CP-E03, and CP-E11; O-003, O-004, and O-015. Immediate safety content leaves the routine path. The final required-domain/applicability manifest remains open under O-015.

20. Obtain caregiver input#

When. Perform only after the beneficiary has chosen to involve a caregiver and the permitted participation scope is known. Repeat when the beneficiary changes that choice, the caregiver changes, permission is narrowed or revoked, or the caregiver reports materially changed observations, constraints, ability, willingness, or needs.

What and how much. Obtain one attributable caregiver input set covering observed beneficiary needs, practical barriers, routines, the caregiver's own ability and willingness to perform proposed work, and support the caregiver believes is needed. Keep the caregiver's words and concerns separate from the beneficiary's words and choices. If the beneficiary does not want caregiver participation, record that choice and do not pursue caregiver input on this basis.

Data.

• The beneficiary's authenticated participation choice proves whether and how the caregiver is invited; it does not by itself establish legal representative authority.
• The caregiver's authenticated response proves that caregiver's own observations, willingness, constraints, and preferences. It does not prove the beneficiary agrees, that the caregiver has decision authority, or that a clinical conclusion is correct.
• A prior caregiver assessment can prefill questions and identify change candidates, but it must carry its date and cannot silently replace current input.
• Contact records, emergency-contact status, portal proxy access, or household relationship identify possible participants only; none independently proves permission or authority.

Potential Proxi work. Send an accessible structured questionnaire or conduct a bounded text/voice interview; preserve the caregiver's original response; summarize it without erasing qualifications; compare it with beneficiary and assessment inputs; flag disagreement; and stop future caregiver outreach when the controlling participation basis is revoked. AI may summarize but cannot decide whose view controls.

Human role. The beneficiary controls ordinary caregiver participation unless a settled decision-specific authority rule says otherwise. The caregiver supplies their own input. L2 reviews the result in the shared planning conversation and handles distress, relationship sensitivity, or differing views. Legal/privacy/clinical owners resolve disputed authority or capacity; L2 does not.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 1 / 2; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Beneficiary participation choice and scope; caregiver identity; collection date and modality; the original response plus faithful summary; separate willingness/constraints; disagreements preserved; and any resulting open action routed without being presented as resolved.

What does not prove completion. A caregiver name in the chart, a caregiver opening a message, an old caregiver assessment, an emergency-contact field, proxy portal access, silence, or a summary that merges caregiver and beneficiary views.

Edge cases/open decisions. CP-E03 through CP-E07; O-011 and O-013. The exact treatment of caregiver-only sensitive information in the beneficiary-facing plan needs privacy/compliance resolution rather than an assumed disclosure rule.

21. Preserve differing views#

When. Trigger whenever two attributable sources disagree about a goal, preference, need, fact, proposed action, willingness, authority, clinical recommendation, or payment/operational fact. Reopen when a later statement changes, corrects, or resolves only part of the disagreement.

What and how much. Create one disagreement record per affected plan element. Preserve each statement verbatim or faithfully, identify the speaker/source and date, state the precise point of agreement and disagreement, and identify which authority or fact is still needed. Do not convert multiple views into an averaged narrative or false consensus.

Data.

• An authenticated statement proves only that speaker's stated view.
• A legal or representation document supports only the scope, conditions, dates, and decisions it actually covers.
• A clinician recommendation proves the clinician's recorded recommendation within scope; it does not erase beneficiary refusal or settle another clinician's conflicting recommendation.
• AI similarity, sentiment, confidence, or majority counts can detect possible conflict but cannot determine truth, capacity, authority, or the controlling choice.

Potential Proxi work. Detect structured and textual contradictions; display statements side by side; retain source links and dates; ask neutral clarification questions; isolate only the affected element; and prepare the issue for the correct authority. Proxi must not select the “most credible” person or rewrite disagreement as agreement.

Human role. L2 authenticates statements, facilitates discussion, and preserves dissent. The capable beneficiary's decision, representative authority, capacity, privacy, and clinical conflicts are resolved only through the settled decision-specific policy and appropriately authorized people. L3/L4/L5 handles clinical conflicts within scope.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 1 / 2; Separate U.S. L3/L4 clinical: 0 / 0 / 4.

Completion evidence. Each distinct view, speaker/source, date, affected element, authority status, resolution owner, and either an authorized resolution or an explicit unresolved disposition that remains visible.

What does not prove completion. Majority vote, caregiver status alone, the newest note, an L2 judgment about capacity, an AI consensus summary, removal of the minority view, or proceeding because one person stopped responding.

Edge cases/open decisions. CP-E04, CP-E05, CP-E06, CP-E09, and CP-E10; O-003 and O-011. The final decision-specific authority hierarchy is open.

22. Identify clinical care options#

When. Perform only when an assessment or new clinical fact identifies a clinical need for which an evaluation, treatment, monitoring approach, or professional service option must be considered for the plan. Repeat when the underlying clinical fact changes, an approval becomes invalid, recommendations conflict, or the beneficiary requests reconsideration.

What and how much. Produce a bounded, individualized option set for the exact clinical question. The set may include evaluation, treatment, monitoring, referral, continuation, or no-change options only when an authorized clinician determines they are clinically appropriate. Preserve rationale, material risks or constraints, required follow-up, and which parts remain choices for the beneficiary. This task does not prescribe, order, implement, or prove acceptance.

Data.

• An authenticated clinician decision directly establishes the option or recommendation that clinician made, for the exact facts reviewed and within actual license, location, and scope.
• A comprehensive assessment, test result, diagnosis, medication list, guideline, or prior note supplies evidence for review; none independently chooses an individualized option.
• An outside clinician recommendation remains an outside recommendation until applicability and any conflict are handled by the authorized care team.
• AI-generated or rules-matched options are draft review material only and must never be presented as clinician-selected care.

Potential Proxi work. Assemble the bounded clinical question, current facts, provenance, conflicts, beneficiary goals and preferences, prior decisions, and source documents; retrieve an approved evidence or service category; draft an option comparison for clinician review; and record the clinician's exact returned decision. Proxi cannot rank treatments by “best,” infer safety, or promote a draft to the plan.

Human role. The L3 clinical-support role may perform only functions within actual RN/LCSW/behavioral scope. L4 or L5 makes diagnosis, treatment, prescribing, or other higher-authority decisions within beneficiary-location authority. The beneficiary chooses among clinically appropriate options after explanation. The final per-element route is open under O-003.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 4–8 / 15.

Completion evidence. Exact clinical question and facts; source links; clinician identity, credentials, beneficiary-location authority and scope; dated decision; option set/rationale; limitations and follow-up; conflicts resolved or preserved; and a clear separation between clinician options and later beneficiary choice.

What does not prove completion. An AI recommendation, a generic guideline, diagnosis code, claims pattern, risk score, resource match, L2 opinion, unsigned draft, blanket plan signature, or outside note whose current applicability is unresolved.

Edge cases/open decisions. CP-E01, CP-E02, CP-E08, CP-E09, CP-E11, and CP-E19; O-003, O-004, O-015, and O-PA-001.

23. Identify functional-support options#

When. Perform when the assessment or beneficiary identifies a functional need involving activities of daily living, instrumental activities, mobility, communication, hearing, vision, cognition-related routines, home access, equipment, or skills. Recheck when function, residence, available help, equipment, or beneficiary preference changes.

What and how much. Generate a practical option set matched to each confirmed functional need. Separate nonclinical assistance and training options from therapy evaluation, prescribed equipment, or other clinical decisions. Record what the option is intended to help with, setting, burden, availability status, and what professional assessment or order is still required.

Data.

• Beneficiary/caregiver reports directly establish reported difficulty and practical context, not a clinical diagnosis or equipment prescription.
• A clinician or therapist assessment directly supports the documented functional finding within scope.
• An equipment catalog, service directory, prior authorization rule, or prior use identifies candidates and administrative conditions; it does not prove clinical suitability, present availability, successful use, or beneficiary acceptance.
• A device order proves an order was made; it does not prove delivery, fit, training, or effective use.

Potential Proxi work. Convert confirmed limitations into approved nonclinical support categories; filter current inventories by geography, accessibility, language, setting, coverage candidates, and beneficiary constraints; compare options; and identify when a therapy assessment, prescription, or clinical review is still required. Proxi may rank practical fit but cannot select a clinical device or therapy.

Human role. L2 handles the one-minute person-centered fit check and relationship-sensitive exceptions. L3/L4/L5 or the relevant therapist decides any clinical evaluation, therapy, equipment prescription, restriction, or individualized safety implication within scope. The beneficiary chooses among appropriate practical options.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 0 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 10.

Completion evidence. Confirmed functional need; source/date; option set with practical constraints and current status; beneficiary preference; any required professional assessment or order; and explicit unresolved items.

What does not prove completion. A diagnosis, ADL score alone, generic equipment link, catalog match, prior equipment claim, an order without practical-fit review, or delivery without training/use confirmation.

Edge cases/open decisions. CP-E01, CP-E02, CP-E08, and CP-E11; O-003, O-004, and O-015.

24. Identify behavioral and psychosocial options#

When. Perform when the assessment, beneficiary, or permitted caregiver identifies a behavioral symptom, emotional concern, caregiver-beneficiary interaction issue, mental-health or substance-use concern, social isolation, distress, or other psychosocial need. Immediate danger leaves this path for Task 25/CP-E11.

What and how much. Separate approved nonclinical dementia-navigation supports—education, routine/environment strategies, coaching, support groups, facilitated conversation, and referral logistics—from individualized diagnosis, psychotherapy, medication, behavioral treatment, or clinical risk decisions. Produce only the option set supported by the correct authority.

Data.

• Beneficiary and caregiver statements directly prove their reported experience or observation, not a diagnosis, intent, capacity, or risk level.
• A clinician or behavioral-health assessment directly supports the exact finding and recommendation documented within scope.
• Screening scores, utilization, messages, and AI sentiment can identify a review candidate; they do not diagnose, determine urgency, or select treatment.
• An approved education/support catalog proves available content or services, not individual clinical appropriateness or completed support.

Potential Proxi work. Collect attributable descriptions; preserve the beneficiary's and caregiver's distinct views; present approved nonclinical supports; retrieve current behavioral-health provider and program candidates; create an exact clinical review/referral packet when needed; and deliver only already-approved educational content. Proxi cannot diagnose, select psychotherapy or medication, or generate individualized behavioral instructions.

Human role. L2 supplies nonclinical dementia-navigation explanation and a brief fit check. A beneficiary-location-authorized L3 behavioral clinician/LCSW or L4/L5 clinician addresses assessment, diagnosis, individualized treatment, safeguarding, or clinical risk within actual scope. The beneficiary chooses participation in appropriate services.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 0 / 1; Separate U.S. L3/L4 clinical: 0 / 4–8 / 15.

Completion evidence. Attributable concern; separation of nonclinical and clinical options; current approved options; beneficiary choice or deferment; clinician-authored decision when required; and exact referral or follow-up dependency.

What does not prove completion. A screening score, AI sentiment label, caregiver characterization presented as diagnosis, generic crisis text, automated provider match, or L2 selection of treatment.

Edge cases/open decisions. CP-E04, CP-E06, CP-E08, CP-E09, and CP-E11; O-003, O-004, and O-011.

25. Identify safety actions#

When. Perform when a current assessment or report identifies a home, driving, wandering, fall, firearm, medication-access, abuse, neglect, exploitation, self-harm, violence, elopement, or other safety concern. A possible immediate threat invokes CP-E11 immediately and does not wait for this card's ordinary completion.

What and how much. For each non-immediate confirmed concern, establish a practical action set with the beneficiary and authorized participants: the exact concern, agreed prevention or response action, responsible person, timing, contact route, required clinical/protective authority, and what happens if the plan fails. Separate routine safeguards from restrictions, clinical risk decisions, capacity decisions, emergency actions, and legally mandated protective actions.

Data.

• A beneficiary/caregiver report directly proves the report and source, not objective risk severity or capacity.
• A home-visit observation proves what was observed at that time, not a universal risk determination.
• A clinician-authored safety/risk assessment supports only the documented assessment, date, scope, and recommended action.
• Sensor alerts, location events, falls claims, missed medications, and AI pattern detection are signals requiring the approved pathway; they do not determine urgency or authorize a restriction.
• Emergency/protective-service acknowledgement proves handoff/receipt, not resolution or Care Plan completion.

Potential Proxi work. Display source-linked concerns; invoke the preapproved response pathway; stop affected routine actions when an immediate-risk signal appears; assemble the clinical or protective review packet; draft plan language from authorized decisions and beneficiary choices; issue approved reminders; and track handoff acknowledgement. Proxi cannot decide risk class, capacity, driving cessation, confinement, protective reporting, or a clinical response.

Human role. L2 develops routine practical safeguards and supports beneficiary/caregiver participation. L3/L4/L5 determines clinical risk, clinical restrictions, treatment, safeguarding, or capacity functions within scope. The approved U.S. emergency/protective pathway acts on immediate threats. L1 stops and escalates when safety content appears.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 0 / 2; Separate U.S. L3/L4 clinical: 0 / 4–8 / 20+.

Completion evidence. Exact concern and source; urgency/authority decision made by the authorized role; beneficiary/representative choice; action, owner, timing, fallback, and accessible communication; required handoff acknowledgement; and every unresolved risk explicitly retained.

What does not prove completion. A safety checkbox, alert generation, message delivery, caregiver reassurance, AI “low risk” classification, emergency call attempt, referral creation, or unsigned safety-plan draft.

Edge cases/open decisions. CP-E04, CP-E06, CP-E07, CP-E08, CP-E10, and CP-E11; O-003, O-004, O-011, and O-013.

26. Identify social and community-service options#

When. Perform when a confirmed social, environmental, or practical need could be addressed by food, transportation, housing, social connection, personal care, home modification, legal/benefits help, adult day, or another community support. Refresh when need, residence, eligibility, availability, cost, language, access, or beneficiary preference changes.

What and how much. Produce a current practical-fit option set for each need, or a truthful no-match/unresolved result. For each candidate, label service, geography, eligibility evidence, current availability status, cost/coverage status, accessibility/language fit, waitlist, and whether a referral or warm introduction is needed. Do not choose on the beneficiary's behalf.

Data.

• The beneficiary's authenticated need and constraints establish the desired practical fit.
• A maintained resource inventory establishes that a program is listed with the recorded attributes and as-of date; it does not prove current availability or acceptance.
• An authenticated program response can establish current eligibility/availability for the exact person and date, subject to stated conditions. It does not prove enrollment or service receipt.
• Claims or payer data may identify prior use or a benefit candidate; they do not prove future community-program availability or beneficiary preference.

Potential Proxi work. Search and rank the maintained inventory; filter by confirmed constraints; request electronic availability/eligibility where supported; explain why each match appears; preserve no-match results; update stale facts; and prepare a referral or requested introduction. Software should do the clean-path matching. L2 enters only for unresolved facts or relationship-sensitive choice.

Human role. No Proxi human is required for a clean, current, explainable match. The beneficiary chooses whether to pursue an option. L1 uses the single shared administrative exception only for manual program verification. L2 handles preference-sensitive discussion, requested warm introduction, distress, disagreement, or no-match problem solving.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 1 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Confirmed need/constraints; dated candidate set or no-match result; source and current status for each key fit fact; explanation of ranking; beneficiary selection/deferment or unresolved disposition; and any required referral dependency.

What does not prove completion. A directory hit, geographic proximity, a high model score, old availability, general program eligibility, website text, referral submission, or a resource link sent to the beneficiary.

Edge cases/open decisions. CP-E01, CP-E02, and CP-E10; O-004 and O-015. Freshness and acceptable evidence for dynamic availability/eligibility should be settled within the content/applicability policy rather than assumed per vendor.

27. Identify caregiver education and support#

When. Perform when a caregiver is involved and a current caregiver assessment or caregiver request identifies an education, skills, coaching, support-call, facilitated-group, peer-support, or other support need. Repeat when caregiver ability, willingness, burden, language, access, or preference changes.

What and how much. Match each assessed caregiver need to one or more approved service/content options with purpose, format, language, accessibility, schedule, human-support level, cost, and participation requirements. Identification of an option is distinct from the caregiver choosing it and from Proxi actually furnishing the education or support.

Data.

• The caregiver's current assessment and authenticated statement directly support that caregiver's reported need and preference.
• The beneficiary's participation choice controls ordinary inclusion in beneficiary planning; it does not convert the caregiver into a representative.
• An approved content and service catalog proves what Proxi or a partner offers, not that it fits the current caregiver or was completed.
• Usage analytics, link views, or chatbot turns prove interaction events only; they do not prove comprehension, skill, support received, or outcome.

Potential Proxi work. Match needs to approved text, video, voice, chatbot, group, coaching, and facilitator options; adapt approved content to language and accessibility without changing meaning; schedule software-deliverable or human services; capture caregiver choice; and route distress or clinical/safeguarding questions. Software can complete clean matching and scheduling; it must not claim that matching equals furnished support.

Human role. No Proxi human is required merely to generate the clean option set or schedule a selected service. The caregiver chooses participation. L2, a trained facilitator, or another qualified Proxi role performs the later human education, coaching, facilitation, or support service when that modality is selected. Clinical/safeguarding content goes to L3/L4.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 0 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Current caregiver need/preferences; dated approved option set; format/accessibility/cost/status; caregiver selection, deferment, or refusal; and a separate downstream service obligation for any selected human or digital service.

What does not prove completion. A generic education library, automated content recommendation, sent link, scheduled session, chatbot availability, caregiver portal access, or old caregiver-assessment result.

Edge cases/open decisions. CP-E03 through CP-E07 and CP-E10; O-011, O-013, and O-015.

28. Identify coordination and medication follow-up#

When. Perform whenever an accepted care-plan element requires care coordination, referral, transition support, community-service connection, medication work, monitoring, or another GUIDE service. Repeat when a source fact, beneficiary choice, clinical decision, destination, or owner changes.

What and how much. Create one atomic downstream obligation for each accepted need. State the exact action or question, source plan element, responsible service/role, destination, required authority, due condition, supporting information, and expected return evidence. Do not hide multiple obligations in one generic “follow up” note.

Data.

• An accepted beneficiary choice or authorized clinical decision establishes that an action belongs in the plan; it does not prove routing, receipt, or performance.
• A medication list, discrepancy, symptom report, or suggested change supplies evidence for the medication service; it does not itself create an authorized medication instruction.
• A work-item acknowledgement proves system or recipient receipt only at the level acknowledged; it does not prove agreement, service performance, or closure.
• AI extraction can propose candidate obligations but cannot decide the destination, urgency, or clinical action.

Potential Proxi work. Detect candidate cross-service needs; require confirmation of the source decision; split them into atomic handoffs; select the destination from an approved service/authority map; deduplicate; prepare the bounded packet; track acknowledgement; and return outcome evidence to the plan. Medication content is routed unchanged to the Medication SOP and authorized clinician.

Human role. L0 performs routine creation/routing. L2 enters only when a sensitive or relationship-based exception requires confirmation that the handoff reflects the beneficiary's choice; that confirmation is counted once in the parent planning interaction rather than as a separate task-28 minute. L1 may pursue a failed administrative endpoint under the shared exception. Clinical staff perform only the downstream clinical work within scope.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 1*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 4.

Completion evidence. One source-linked obligation per accepted action; exact destination, owner, required authority, due condition, packet identity, acknowledgement status, and link back to the plan element. Downstream status remains separate.

What does not prove completion. A generic follow-up paragraph, AI-detected candidate, task creation alone, message send, queue placement, or completion of one obligation when others remain.

Edge cases/open decisions. CP-E01, CP-E02, CP-E08, CP-E09, CP-E17, and CP-E19; O-003, O-004, and O-015. Cross-pillar disclosure and clinical-authority decisions remain with the receiving service.

29. Explain the available options#

When. Perform after the applicable option set is current and authorized, and before the beneficiary is asked to select plan content. Repeat when an option, material burden, cost/coverage fact, availability, clinical recommendation, beneficiary communication need, or source changes.

What and how much. Explain each realistic option's purpose, likely process, material burden, known constraints, known payment status, alternatives, and what remains uncertain. Clearly distinguish approved clinical explanation from navigator explanation of logistics and services. Use the beneficiary's preferred language, format, and accessibility supports. Allow questions and do not pressure the beneficiary toward the system's top-ranked option.

Data.

• The approved option payload and source facts prove what may be explained; they do not prove the beneficiary received or understood it.
• Delivery/open/view logs prove transmission or interaction, not comprehension or voluntariness.
• A beneficiary's authenticated question, correction, or teach-back response directly supports what that person communicated; a yes/no click alone is weak evidence unless a future approved policy accepts it for that decision class.
• AI may render approved content and answer bounded factual questions from the approved knowledge base. It may not improvise individualized clinical rationale, omit material alternatives, or decide what the beneficiary should receive.

Potential Proxi work. Produce plain-language, translated, accessible text, video, and bounded voice/chat explanations from approved content; compare options; disclose data dates and uncertainty; answer routine logistics questions; capture questions and corrections; and summon L2 or the owning clinician when the question exceeds the approved content.

Human role. The prepared option comparison is presented first, and an authenticated beneficiary/representative may obtain the approved explanation and ask bounded questions directly on an approved clean digital or AI path. L2 enters only when a human planning service is required, the person requests help, ambiguity/disagreement/distress remains, the digital route is inaccessible, or relationship work can materially improve the result. The owning L3/L4/L5 clinician receives a prepared exact question and explains individualized clinical rationale, material clinical risks, and clinical alternatives. A qualified interpreter or accessibility support is used when required; AI translation alone does not settle clinically material ambiguity.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 4 / 6; Separate U.S. L3/L4 clinical: 0 / 0 / 5. The zero-L2 low route is the prepared accessible clean path; the typical/high L2 minutes apply only to the human-entry conditions above.

Completion evidence. Exact option set and source identity; explanation date/modality/language/accessibility; material uncertainties and alternatives presented; questions and answers; owning clinician participation when required; and a recorded opportunity for the beneficiary to pause, decline, or ask for a person.

What does not prove completion. Sending a PDF, playing a video, opening a page, a checkbox, a generic chatbot transcript, a model-generated “understood” score, silence, or an explanation that omits known burden, cost uncertainty, or alternatives.

Edge cases/open decisions. CP-E03, CP-E06, CP-E08, CP-E10, and CP-E11; O-003, O-004, O-011, and O-015. Proposed OPEN decision, no ID assigned: define which plan elements may use digital/AI-assisted explanation on the clean path, the evidence required, and the triggers for navigator, interpreter, privacy, or clinician involvement.

30. Let the beneficiary choose plan content#

When. Perform after the beneficiary has received the applicable explanation and before the draft presents a goal, recommendation, service, provider, or support as accepted. Repeat when the beneficiary changes a choice, the option set materially changes, authority changes, or a prior choice was captured from an unverified speaker.

What and how much. Obtain an authenticated disposition for each proposed plan element: accept, decline, defer, request an alternative, or request more information. Preserve the beneficiary's own words and any conditions. A beneficiary can choose among clinically appropriate options and can refuse a recommendation; the choice does not authorize an unavailable, unlawful, or clinically unapproved action.

Data.

• An authenticated beneficiary statement directly proves the beneficiary's choice for the identified element and time.
• A representative statement applies only when decision-specific authority and scope are established.
• A caregiver statement proves the caregiver's view, not the beneficiary's choice or representative authority.
• A digital selection can be direct evidence only when speaker authentication, exact option identity, accessibility, voluntariness, and correction/confirmation requirements are satisfied under an approved policy.
• AI extraction of a conversational preference is a candidate until the beneficiary or authorized representative confirms it.

Potential Proxi work. Present the exact authorized option set; capture and read back selections; preserve refusals and conditions; detect contradictory or stale choices; keep unselected items out of accepted content; and route requests for clinical, legal, coverage, or feasibility review. Proxi cannot choose a default because the beneficiary did not respond.

Human role. The beneficiary leads. An authenticated beneficiary/representative choice may be captured directly when the approved route establishes the exact option, identity, accessibility, voluntariness, and correction opportunity. L2 receives the prepared choice packet only when a human planning service is required, the person requests help, ambiguity/disagreement/distress remains, accessibility fails, or relationship work adds value. L2 supports that human segment and preserves refusal/dissent but does not bless a clear choice. The relevant clinician receives a prepared clinical question and explains clinical limits and alternatives; legal/privacy roles resolve representative authority.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 3 / 4; Separate U.S. L3/L4 clinical: 0 / 0 / 0. The zero-L2 low route is a clear authenticated direct choice; typical/high minutes apply only to the human-entry conditions above.

Completion evidence. Authenticated chooser and authority basis when applicable; exact plan element and option set; accept/decline/defer/alternative status; date and modality; conditions/refusal; unresolved questions; and proof the resulting draft reflects the recorded choice without suppressing required safety or legal follow-up.

What does not prove completion. A preselected default, timeout, silence, caregiver choice without authority, old consent, inferred preference, model confidence, generic “agree to plan,” or a click tied to a changed option set.

Edge cases/open decisions. CP-E03 through CP-E06, CP-E10, and CP-E11; O-003, O-011, and O-015. The proposed digital/AI choice-evidence decision under Task 29 also governs this task.

31. Name recommended providers#

Choice-steering boundary. Filter first on confirmed hard constraints. Keep every serious remaining option and material unknown visible. Do not use a sponsored, undisclosed-composite, or unexplained "best" default; disclose any displayed sort criterion. Whether ranking is permitted only after the beneficiary selects the criterion remains open under O-044.

When. Perform after a service option is accepted and a specific provider/program is needed for execution. Recheck when the service, clinical order, beneficiary constraints, coverage, network, availability, provider location, waitlist, or provider status changes.

What and how much. Identify the specific provider/program or a bounded choice set, not merely a provider category. Record why each candidate fits practical requirements, which fit facts are verified versus unknown, whether clinical selection/order is required, and who chooses. If no suitable provider is found, preserve no-match/unresolved rather than naming the nearest directory result.

Data.

• CMS/payer directory data can establish the payer-listed name, address, phone, specialty, and contracted-network status as of the directory data; it does not prove current availability, acceptance of new patients, treatment relationship, clinical suitability, or beneficiary choice (SRC-CMS-PROVIDER-DIRECTORY-API).
• NPI/NPPES normalizes identity; an NPI does not itself establish licensure, credentialing, coverage, availability, or fit (SRC-CMS-NPI-FACT-SHEET).
• Authenticated provider/program confirmation can establish current operational availability and constraints for the exact inquiry/date.
• A clinician order or recommendation supports clinical selection within scope; it does not prove scheduling or beneficiary acceptance.
• The beneficiary's authenticated selection proves preference among appropriate options, not coverage or clinical suitability.

Potential Proxi work. Search maintained clinical/provider/community directories; normalize identities; filter by confirmed geography, language, accessibility, modality, availability, coverage candidates, and beneficiary constraints; explain ranking; request electronic verification; and identify stale/conflicting facts. Software performs the clean match. L1 handles manual verification only under the shared exception.

Human role. The prepared choice set and unresolved facts are presented directly; the beneficiary selects when choice exists. L2 enters only when a human planning service is required/requested, ambiguity/disagreement/distress remains, accessibility fails, a warm introduction is requested, or relationship-dependent practical fit remains. L3/L4/L5 receives the prepared exact clinical question and selects or orders only when clinical scope requires. L1 verifies dynamic administrative facts but cannot select the provider for the beneficiary.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 3*; Puerto Rico L2 navigator: 0 / 0 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 4. L2 time appears only on the human-entry conditions above.

Completion evidence. Service need; named provider/program and stable identifier when available; dated source; verified and unknown fit facts; beneficiary selection or clinician order as applicable; availability/coverage status; and no-match disposition when no suitable option exists.

What does not prove completion. A directory hit, NPI, network listing, nearest-distance rank, highest model score, historical claim, old availability, appointment request, or L1/navigator selection of a clinically required specialty.

Edge cases/open decisions. CP-E01, CP-E02, CP-E08, CP-E09, and CP-E10; O-003, O-004, and O-015.

32. Assign responsibility for each action#

When. Perform after each action is accepted and before the plan is treated as executable. Repeat whenever the action, owner, team roster, authority, deadline, provider, or beneficiary/caregiver willingness changes.

What and how much. Give every action one accountable owner, any supporting actors, a due condition/date when approved, and a required return artifact. Identify whether the owner is Proxi, the Participant, beneficiary, caregiver, clinician, provider, payer, community program, or another organization. Do not use “care team” as the sole owner when no person or service queue is accountable.

Data.

• An authoritative current role roster and approved service/authority map establish who may receive a task; they do not prove the individual accepted it.
• Beneficiary/caregiver willingness directly supports their own accepted responsibility but does not create professional authority.
• A work-item creation or delivery event proves creation/delivery only.
• An authenticated owner acknowledgement proves receipt and acceptance at the level stated; it does not prove performance.

Potential Proxi work. Propose the owner from the approved role/service map; detect missing, duplicate, ineligible, or overloaded assignments; create the work item; request acknowledgement; reassign only through approved rules; and keep supporting actors distinct from the accountable owner. AI may draft the handoff but cannot create authority.

Human role. Proxi software performs the routine ownership-completeness check and assigns internal work from authoritative rules. L2 enters only for a beneficiary/caregiver refusal or a relationship-sensitive ownership exception; that exception is counted once in the parent planning interaction rather than as a separate task-32 minute. Each named owner accepts their responsibility; external-party time is not Proxi labor.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 1*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Exact action; accountable owner and supporting actors; authority source; due condition/date; required return evidence; owner acknowledgement or truthful pending/refused status; and resulting escalation/reassignment when needed.

What does not prove completion. An assignee field, generic team label, email send, queue placement, calendar invitation, presumed caregiver responsibility, or assignment to someone whose authority/availability is not established.

Edge cases/open decisions. CP-E01, CP-E02, CP-E17, and CP-E19; O-004 and O-015.

33. Identify payment responsibility#

When. Perform for every recommended service provider/program before the beneficiary is asked to rely on the plan's payment statement. Refresh when the service, provider, benefit year, payer, eligibility, authorization, network, price, or coverage condition changes.

What and how much. For each recommended provider/service, identify the expected payment source—Medicare, Medicaid, managed care, another benefit/program, grant, Participant/Proxi commitment, beneficiary self-pay, or unresolved—and label the statement as confirmed for the exact service/date, conditional, estimated, denied, not covered, or unknown. Record known beneficiary cost and prerequisites without implying a guarantee.

Data.

• An authenticated payer/program eligibility or benefit response supports only the exact person, service, provider, date, and conditions queried.
• Prior authorization proves authorization to the extent stated; it does not guarantee payment if other conditions fail.
• Claims/encounters from the Patient Access API prove historical adjudication, not future coverage or price (SRC-CMS-PATIENT-ACCESS-API).
• Provider Directory API network status proves payer-listed contracting status as of the directory data; it does not prove service-specific coverage, availability, or final beneficiary cost.
• A general benefit booklet, estimator, or AI calculation is an estimate unless confirmed by the responsible payer/program.

Potential Proxi work. Query connected benefit and eligibility sources; retrieve service/provider-specific coverage facts; preserve source, date, codes and conditions; compare network and program data; calculate clearly labeled estimates; request missing authorization; and flag conflicts or stale information. Proxi cannot make the payer's coverage determination or guarantee payment.

Human role. L2 uses one minute in the typical or high shared conversation to ensure the beneficiary sees confirmed versus estimated/unknown status and can choose whether to consider uncovered cost; no separate L2 minute is allocated on the low route. L1 may use the one shared 13-minute exception for a manual payer/program inquiry. A payer/program or benefits authority supplies the controlling response; clinical staff supply medical necessity/order content only within scope.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 3*; Puerto Rico L2 navigator: 0 / 1 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Exact service/provider; expected payer/person/program; status label; source and date; known coverage conditions, authorization, network, and beneficiary cost; unresolved facts; and beneficiary decision when self-pay exposure is material.

What does not prove completion. Historical payment, eligibility alone, network listing, benefit booklet, price estimator, prior authorization alone, a quoted charge without coverage, or a model-generated coverage probability.

Edge cases/open decisions. CP-E01, CP-E02, and CP-E10; O-004 and O-015. The content policy must define freshness and acceptable proof for payment statements; the care-plan task must not silently convert “expected payer” into guaranteed coverage.

34. Resolve missing or conflicting information#

When. Run the missing/conflict check before a plan element is accepted, before the draft is finalized, and whenever new, stale, corrected, duplicated, or late-arriving evidence affects an element. The routine no-conflict check is software-only; a resolution episode begins only when a material issue exists.

What and how much. Create one atomic issue per missing or conflicting fact. State the affected plan element, sources and dates, exact conflict or missing field, why it matters, who can supply/decide it, permitted next actions, and whether unaffected elements may continue. Resolve from the correct authority/source or preserve a truthful unresolved status; do not choose the newest or most frequent value by default.

Data.

• Source authority is fact-specific: beneficiary choice comes from the beneficiary/authorized representative, clinical judgment from the authorized clinician, provider availability from the provider, payment from the payer/program, and team role from the authoritative roster.
• A newer record may supersede an older record only when the source and event actually establish correction/supersession.
• Repeated copies of one source are not independent corroboration.
• AI can detect and summarize conflict but cannot select the authoritative fact, decide materiality, assign urgency, or create an exception.

Potential Proxi work. Detect blanks, staleness, duplicated provenance, and contradictions; group copies by origin; ask a targeted question; retrieve the named source; preserve late evidence; show affected elements; and route the issue to the fact/authority owner. Unaffected plan content may continue only when the governing policy permits it.

Human role. No human is needed when the check finds no issue. L1 uses the shared administrative exception for manual record/provider/payer pursuit. L2 resolves beneficiary, caregiver, household, and substantive navigation facts without deciding legal authority or clinical truth. L3/L4/L5, privacy/legal, payer, or provider resolves its own domain.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 0 / 2; Separate U.S. L3/L4 clinical: 0 / 0 / 4.

Completion evidence. Issue inventory; affected elements; source attempts; authoritative returned fact or explicit unresolved disposition; provenance and dates; person/authority that resolved it; preservation of dissent/conflict; and confirmation that no blocked element was presented as settled.

What does not prove completion. Choosing the latest value, majority vote, duplicate records, silence, exhausted outreach, blank-to-false conversion, AI confidence, an ad hoc exception note, or publishing while the affected element remains unresolved.

Edge cases/open decisions. CP-E01, CP-E02, CP-E04, CP-E05, CP-E06, CP-E08, CP-E09, CP-E17, and CP-E19; O-003, O-004, O-011, and O-015.

35. Obtain required clinical decisions#

When. Perform whenever a proposed or changed plan element requires clinical judgment and no current, applicable, scope-valid decision is present. Repeat when underlying facts change, an approval expires or is invalidated, recommendations conflict, a beneficiary requests a clinical alternative, or the returned decision does not answer the exact question.

What and how much. Send each bounded diagnosis, treatment, medication, behavioral-health, safety, capacity-function, equipment/therapy, or other clinical question to the professional whose actual authority covers it. Obtain an exact decision tied to the facts reviewed, effective timing, rationale/limits, required monitoring/follow-up, and communication instructions. Do not use a blanket plan signature to fill multiple unstated clinical gaps.

Each returned decision is a scoped attributable clinical contribution, not a plan-level approval. In a multi-clinician plan, assign and document each clinical question separately. One professional's answer may cover several related elements when the answer genuinely addresses them, but it must not be stretched to unstated questions. If two returned decisions conflict, preserve both positions and obtain a disposition from the professional authorized for the exact disputed issue. The Care Plan owner and navigator may coordinate that resolution; they may not choose which clinical position controls.

Data.

• A clinician-authored decision proves that clinician made the recorded decision for the exact evidence reviewed and within verified scope/location.
• A signature proves authorship/attestation only to the content it clearly covers; it does not cure omitted questions or establish authority outside scope.
• An RN review is not automatically prescribing, diagnostic, capacity, behavioral-treatment, or physician/APP authority.
• An outside recommendation remains attributable external evidence until the appropriate current clinical owner determines its applicability and conflicts.
• AI output, protocol match, risk score, or navigator recommendation is a review candidate only.

Potential Proxi work. Build a concise review packet with the exact unresolved question, current facts, provenance, conflicts, beneficiary goals/preferences, prior decisions, already-attempted actions, allowed response fields, and required return evidence; route it to the authorized role; track nonresponse; and insert only the returned authorized decision into the plan.

Human role. No L2 review is required merely to inspect a structurally complete clinical return. Software checks required fields and keeps unanswered parts open. L2 re-enters only when the beneficiary needs explanation or a new preference-sensitive choice. L3 acts only within RN/LCSW/behavioral scope. L4/L5 makes diagnosis, treatment, prescribing, or other higher-authority decisions within beneficiary-location authority. The final route matrix remains open under O-003.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: shared with tasks 22–34.

Completion evidence. Exact question and affected element; complete source packet; clinician identity, credentials, location authority and scope; dated decision; rationale/limits; effective timing; monitoring/follow-up and communication requirements; conflicts addressed; and every unanswered clinical element still visibly open.

What does not prove completion. A generic “reviewed” note, RN signature without relevant authority, plan-level co-signature, AI recommendation, protocol match, copied outside note, routed task, clinician notification, or nonresponse.

Edge cases/open decisions. CP-E08, CP-E09, CP-E11, and CP-E19; O-003, O-004, and O-PA-001.

36. Draft the written plan#

When. After tasks 1-35 have produced the beneficiary-led goals, strengths, preferences, needs, accepted options, named action owners, payment information, and any required professional decisions. Start only when each included statement has an identified source and unresolved matters remain visibly unresolved. The public anchor is GUIDE RFA Appendix B 2.1-2.2.

What and how much. Produce one coherent written-plan draft for the beneficiary. Include every applicable care-plan category in 2.1: beneficiary goals, strengths, preferences, and needs; relevant comprehensive-assessment findings; accepted clinical, functional, behavioral, psychosocial, safety, coordination, community-service, medication-follow-up, and caregiver-support actions; recommended providers; action owner for every accepted action; and payer, program, or person expected to pay each provider where required. Preserve the beneficiary's own words for preference-sensitive content. Include one explicit unresolved-matters section rather than silently omitting or settling unanswered questions.

Data. Use authenticated beneficiary or authorized-representative choices; permitted caregiver input kept distinct; dated assessment findings; current provider/service information; accepted recommendations; named owners; payment-source evidence; scoped clinical decisions; and unresolved-item records. A populated source field proves only that information was supplied; it does not prove accuracy, currentness, beneficiary acceptance, clinical approval, feasibility, coverage, or completion of the underlying action. AI-generated text remains a draft until the L2 review described below.

Potential Proxi work. Deterministically gather the confirmed inputs, identify missing fields, and build the plan outline. AI may draft plain-language narrative from those inputs, preserve direct statements, cross-reference sources, identify apparent contradictions, and keep unresolved items conspicuous. Proxi must not invent a recommendation, choose an option, remove dissent, convert an estimate into confirmed coverage, or supply a clinical decision.

Human role. The L2 Puerto Rico dementia care navigator checks that the draft accurately represents the beneficiary's direction and does not overstate unresolved or externally owned matters. The specifically authorized L3/L4/L5 actor is involved only for an unresolved or changed element within that actor's scope; drafting alone does not require a blanket clinical rereview.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 0 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. One readable draft containing all applicable categories, source-linked statements, beneficiary-language content, named action owners, payment information where applicable, and an explicit unresolved-item list; plus an L2 review record identifying corrections or confirming the draft is ready for the completeness check.

What does not prove completion. A generated document; a template with filled headings; a copied assessment; a list of recommendations without beneficiary choices; an action with no owner; a provider with no payment-responsibility information where required; or a clinical statement that lacks the required professional decision.

Edge cases/open decisions. CP-E01, CP-E02, CP-E03, CP-E04, CP-E08, CP-E09, CP-E10, CP-E17; O-003, O-004, O-011, O-015, O-PA-001.

37. Check the plan for required content#

When. Immediately after task 36 produces a complete-looking draft and again after any later change that can affect content completeness. The public content anchor is GUIDE RFA Appendix B 2.1; the exact Proxi applicability checklist remains open under O-015.

What and how much. Check every applicable care-plan requirement once, and check every accepted action individually. Confirm that the draft addresses all applicable assessment findings and beneficiary-selected goals; names recommended providers where required; names one responsible owner for each accepted action; identifies expected payment responsibility for each recommended service provider; includes caregiver education/support choices when there is an involved caregiver; and visibly identifies unresolved, declined, not-applicable, or awaiting-answer items without treating them as completed. Return a specific gap list, not a generic "review the plan" task.

Data. Use the draft, the underlying assessment and planning inputs, applicability facts, accepted-action list, provider list, payment information, caregiver status and permissions, clinical-decision records, and unresolved-item list. A checked box proves only that the expected content is present in the draft. It does not prove clinical appropriateness, beneficiary agreement, provider availability, coverage, EHR incorporation, distribution, or action execution.

Potential Proxi work. Software can perform the entire structural and cross-document check, identify blanks, compare action counts, find unassigned actions, detect inconsistent names or payment statements, and route each substantive gap to its fact owner. AI may explain a detected inconsistency but cannot decide that a missing item is inapplicable or that conflicting clinical content is acceptable.

Human role. L2 resolves substantive nonclinical gaps and confirms that a beneficiary-sensitive omission was not intentional. The appropriate L3/L4/L5 actor resolves only the clinical gap placed in that actor's scope. There is no need for a human to recheck fields that pass deterministic checks and do not contain unresolved meaning.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 0 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 4.

Completion evidence. A requirement-by-requirement result tied to the exact draft, with each applicable item passed and each non-applicable item supported by an authorized applicability reason; zero unowned accepted actions; and every remaining gap assigned to a named resolver rather than hidden.

What does not prove completion. Document length; all headings being present; an AI statement that the plan "looks complete"; a checklist run against the wrong draft; a clinical gap relabeled nonclinical; or a missing field marked not applicable without supporting facts and authority.

Edge cases/open decisions. CP-E01, CP-E02, CP-E08, CP-E17, CP-E20; O-003, O-004, O-015, O-017, O-PA-001.

38. Make the plan understandable and accessible#

When. After approved content exists and before beneficiary review or delivery. Repeat when the beneficiary's language, reading, sensory, cognitive, communication, channel, or support needs change, or when task 40/43 changes content that must be rerendered.

What and how much. Produce one usable rendition for each confirmed beneficiary need: preferred spoken/written language, plain-language reading level, large text or other visual accommodation, accessible electronic or print format, and any communication support needed for the review. Render the whole plan that the beneficiary must review; do not simplify only the summary while leaving action instructions inaccessible. Preserve clinical meaning and clearly distinguish instructions, options, unresolved questions, and action owners.

Data. Use confirmed communication preferences and accommodations, the approved source content, approved terminology, delivery-channel capability, and any interpreter/translation requirement. A stated language preference does not prove literacy, comprehension, preferred dialect, or the usability of a particular format. A translated or reformatted file does not prove fidelity or beneficiary understanding.

Potential Proxi work. Software can generate accessible layout, adjust reading complexity without altering meaning, create structured summaries, and produce draft translations or alternative renderings from the approved plan. It can compare the rendition against the source for omitted numbers, names, actions, warnings, and ownership. Unsupported language, ambiguous clinical language, or any fidelity warning must route to a qualified person.

Human role. L2 verifies that the rendition matches the beneficiary's known preferences and remains practically usable. A qualified interpreter, translator, accessibility professional, or relevant clinician is required only when supported automation cannot establish faithful communication or when clinical meaning is genuinely ambiguous. The beneficiary confirms usability during task 39; beneficiary time is not Proxi labor.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 0 / 2; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Accessible rendition(s) tied to the approved content; automated content-fidelity check passed; requested accommodation present; and no unresolved fidelity warning before task 39.

What does not prove completion. Selecting a language from a menu; machine translation alone; a PDF merely being generated; a readability score; or successful delivery to a device that the beneficiary cannot use.

Edge cases/open decisions. CP-E02, CP-E03, CP-E17; O-004, O-015, O-PA-001. Missing decision to consider: a supported-language/accessibility policy that states when automated rendition is sufficient and when a qualified interpreter, translator, or accessibility reviewer is mandatory.

39. Review the draft with the beneficiary#

When. After tasks 36-38 produce a complete and usable draft, before the plan can be treated as beneficiary-led and final. Repeat for the changed portions of a revision. GUIDE RFA Appendix B 2.2 requires the initial plan and future revisions to be led by the beneficiary.

What and how much. Conduct one beneficiary-led review covering every goal and every accepted action, plus the unresolved, declined, provider, payment, caregiver, and coordination sections. Ask what is correct, missing, unacceptable, unclear, or no longer wanted. Capture one disposition for each plan section: accept, correct, decline, ask a question, defer, or require another person's decision. Include a caregiver only at the beneficiary's discretion and within the permitted participation scope. Do not force the conversation into false consensus.

Data. Use the accessible draft, authenticated beneficiary/representative identity, decision-specific authority facts when relevant, permitted participants, communication needs, section-level questions, and any available clinician explanations. Attendance proves presence, not review. Opening a portal proves access, not reading. Silence proves neither acceptance nor refusal. A caregiver response is not automatically the beneficiary's response or a legally controlling decision.

Potential Proxi work. Present the plan section by section, speak or display approved text, capture verbatim answers, create a structured revision list, distinguish speakers, summarize without deleting the original statement, and route clinical/legal questions. Proxi may prompt for missed sections but cannot authenticate a choice without the approved process, decide whose choice controls, answer clinical questions, or mark the plan approved from sentiment or silence.

Human role. The complete accessible draft is prepared before human time begins. Where an approved clean route is sufficient, the authenticated beneficiary/representative may review sections, correct them, and supply dispositions directly. L2 conducts the human review only when the governing requirement requires it, the person requests it, ambiguity/disagreement/distress or accessibility failure remains, or relationship work materially improves the review. L2 does not recreate the draft or bless a clean digital review. An authorized representative acts only within documented scope. L3/L4/L5 receives only the prepared exact clinical question and responds within scope. Beneficiary, caregiver, and external-person time is excluded from Proxi labor.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 4 / 10; Separate U.S. L3/L4 clinical: 0 / 0 / 5. The zero-L2 low route is an approved accessible direct review; the typical/high L2 minutes apply only to the human-entry conditions above.

Completion evidence. Authenticated review event; permitted participant list; section-level beneficiary dispositions; captured corrections/questions/refusals; preserved differing views; and a revision or no-change work list with each restricted issue routed to the correct authority.

What does not prove completion. Sending the draft; portal-view telemetry; a generic "reviewed" checkbox; caregiver approval without beneficiary direction or applicable representative authority; an electronic signature that does not identify what was reviewed; or a navigator deciding that nonresponse means agreement.

Edge cases/open decisions. CP-E03, CP-E04, CP-E05, CP-E06, CP-E07, CP-E09, CP-E10, CP-E11, CP-E15, CP-E19; O-003, O-004, O-011, O-013, O-PA-001.

40. Revise the plan from beneficiary direction#

When. After task 39 produces authenticated corrections, additions, removals, or refusals. This task handles changes within the current planning episode; task 52 handles a later circumstance or request that reopens planning.

What and how much. Apply every beneficiary-directed change once. Separate changes that are purely wording/preference from changes that affect clinical care, safety, legal authority, coverage, provider feasibility, or another person's commitment. Apply only changes already within the beneficiary's authority and route each restricted change to the responsible professional. Return the changed section to the beneficiary when the change could alter meaning; do not silently normalize the beneficiary's words.

Data. Use authenticated section-level change requests, original beneficiary wording, representative scope where applicable, current draft, affected action owners/providers, and required professional responses. A requested change proves beneficiary direction but not clinical appropriateness, provider acceptance, coverage, legal authority, or feasibility. AI-proposed replacement language does not prove beneficiary approval.

Potential Proxi work. Create a precise change list, apply approved nonclinical wording edits, show before/after language, detect downstream sections affected by the change, and route restricted items. Proxi cannot approve the change, decide capacity, substitute a caregiver's preference, or convert a beneficiary request into a clinical order.

Human role. L2 confirms and applies nonclinical beneficiary-directed changes and explains operational constraints from approved information. L3/L4/L5 or privacy/legal/coverage owner decides only the restricted element. The beneficiary decides whether the resulting permitted content reflects their direction.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 0 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 4.

Completion evidence. Every authenticated change request has a disposition; allowed edits appear in the correct section; restricted requests have named owners and remain visibly pending; affected cross-references are consistent; and the beneficiary-facing change summary is ready for confirmation.

What does not prove completion. Accepting a comment; changing the text without preserving the request; labeling an infeasible request "completed"; clinician review of one element being treated as approval of all edits; or a clean document that hides still-pending questions.

Edge cases/open decisions. CP-E03, CP-E04, CP-E05, CP-E06, CP-E08, CP-E09, CP-E10, CP-E17; O-003, O-004, O-010, O-011, O-015, O-PA-001.

41. Record refusals and unresolved matters#

When. Whenever task 39 or 40 produces a declined recommendation, unresolved disagreement, unanswered question, unavailable provider, unconfirmed payment source, or pending clinical/legal/feasibility decision. Recheck before finalization and again before sending any affected action.

What and how much. Create one distinct entry for every refusal and every unresolved matter. Record the speaker, exact choice or issue, affected plan section/action, explanation or alternatives provided from an authorized source, responsible resolver if any, follow-up need, and whether unaffected work may proceed. Preserve partial refusal and differing views; do not collapse them into one generic note.

Data. Use authenticated statements, speaker identity and authority, original recommendation, permitted explanations, offered alternatives, dissenting views, and outstanding requests. A refusal proves that the authenticated person declined the specified item at that time; it does not automatically prove decision-specific capacity, informed understanding, permanent refusal, cancellation of unrelated actions, or safe case closure. An unresolved record proves the issue remains open, not that the plan failed.

Potential Proxi work. Capture verbatim and structured records, keep refusals separate from missing consent, maintain the follow-up list, prevent refused/pending actions from appearing completed, and route clinical-consequence explanations. AI may summarize but may not infer informed refusal or decide whose preference controls.

Human role. L2 authenticates the beneficiary/caregiver/representative statement, preserves dissent, and explains approved nonclinical consequences and alternatives. L3/L4 explains clinical consequences and alternatives when required. Legal/compliance resolves authority ambiguity. No human is needed in the ordinary path when no refusal or unresolved matter exists.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 0 / 1; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. A complete, speaker-attributed entry for every refusal/unresolved matter; accurate effect on affected versus unaffected actions; named follow-up owner where one exists; and no plan statement falsely presenting the item as accepted or completed.

What does not prove completion. "Patient refused" without the specific item; caregiver refusal substituted for beneficiary direction; nonresponse; missed appointment; unresolved clinical disagreement represented as beneficiary refusal; or closing the whole plan because one recommendation was declined.

Edge cases/open decisions. CP-E04, CP-E05, CP-E06, CP-E09, CP-E10, CP-E17, CP-E20; O-003, O-004, O-010, O-011, O-017, O-PA-001.

42. Confirm an unchanged plan#

When. Use only when current assessment/review inputs and the beneficiary-led review identify no needed or requested change. This is an alternative outcome to tasks 40 and 43, not a shortcut around task 39. GUIDE RFA Appendix B 2.3 requires modification as needed or requested; O-010 must settle the evidence for a reviewed-no-change outcome. A brief “still fits” check inside a routine navigator contact is narrower contact-level maintenance; it is not this whole-plan procedure and does not create a monthly whole-plan review requirement.

What and how much. Review every current plan section against new assessment findings, reported circumstances, goals, preferences, needs, open actions, and professional content. Obtain an authenticated beneficiary confirmation that the current written plan still reflects their direction. For any clinician-owned content whose continued appropriateness cannot be established without clinical judgment, obtain a scoped clinician disposition rather than asking the navigator to reaffirm it. Produce one reviewed-no-change record for the whole plan plus explicit dispositions for any questioned elements.

A reviewed-no-change plan does not require a new signature on every clinical section. Clinical content may be carried forward without incremental review only when its source, scope, underlying facts, effective period, and continued applicability remain current and no new symptom, discrepancy, request, conflicting recommendation, or material circumstance creates a fresh judgment question. Neither elapsed time alone nor the absence of newly received data proves continued appropriateness.

Data. Use the current plan, current comprehensive/interim findings, beneficiary review responses, changed-circumstance search, current provider/service facts, open-action outcomes, and scoped professional reviews. A software no-difference comparison proves only that the compared fields did not change. It does not prove that inputs are current, that omitted facts do not exist, that clinical content remains appropriate, or that the beneficiary confirmed the plan.

Potential Proxi work. Compare the current inputs with the plan, identify apparent deltas and stale dependencies, present every section for confirmation, and draft a reviewed-no-change summary. Proxi cannot decide materiality, clinical continued appropriateness, or beneficiary confirmation.

Human role. L2 conducts/authenticates the beneficiary confirmation and resolves nonclinical discrepancies. The appropriately authorized L3/L4/L5 actor reviews only clinical content that requires renewed judgment under the final O-003 taxonomy. Beneficiary time is excluded from Proxi labor.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 1 / 2 on the no-change route only; Separate U.S. L3/L4 clinical: 0 / 0 / 4.

Completion evidence. Authenticated beneficiary confirmation tied to the current plan; comparison against current inputs; explicit reviewed-no-change disposition; no unresolved material delta; and any required scoped clinical reaffirmation. This outcome remains separate from EHR, distribution, action, and service-qualification evidence.

What does not prove completion. No new data arriving; unchanged text; an AI no-delta result; a navigator's unilateral judgment; inability to reach the beneficiary; a caregiver saying "same" without authority; or a prior clinical approval assumed never to expire.

Edge cases/open decisions. CP-E02, CP-E03, CP-E04, CP-E06, CP-E08, CP-E10, CP-E15, CP-E20, CP-E21; O-003, O-004, O-010, O-011, O-015, O-017, O-PA-001.

43. Finalize a changed plan#

When. After tasks 39-41 resolve or explicitly preserve every beneficiary comment and after every changed element requiring professional authority has the correct decision. Use only for a changed plan; an unchanged plan follows task 42.

What and how much. Produce one clean changed plan that incorporates every accepted beneficiary-directed change and every required professional decision, preserves refusals and unresolved matters truthfully, removes superseded draft language, and keeps all action owners, providers, and payment information synchronized. Confirm the final plan has no hidden tracked changes or ambiguous competing instructions. Prepare—but do not yet claim—EHR incorporation, delivery, access, or execution.

Data. Use the reviewed draft, authenticated change requests, professional decisions with actual scope, final action/provider/payment list, refusal/unresolved list, and completed content check. A clinical approval proves only the exact clinical element approved; it does not approve unrelated plan content. A clean compiled document proves final assembly, not EHR incorporation, human release, recipient delivery, access, acknowledgement, action execution, or service qualification.

Potential Proxi work. Apply accepted edits, reconcile duplicated or superseded statements, check internal consistency, rerun completeness/fidelity checks, and prepare the exact final document and recipient/action lists. AI may draft and explain deltas but cannot approve the plan or silently resolve conflicting inputs.

Human role. The Participant-authorized Care Plan owner performs the plan-level completion review. That review confirms beneficiary leadership, complete dispositions, action ownership, correct attribution of every clinical contribution, truthful treatment of unresolved matters, and internal consistency. L2 may hold this responsibility only if the Participant arrangement authorizes it. L3/L4/L5 approves only the changed elements within actual authority; there is no blanket rereview when no clinical element changed. The plan remains incomplete when a required clinical contribution is missing or conflicting clinical instructions are presented as settled.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 0; Puerto Rico L2 navigator: 0 / 0 / 2; Separate U.S. L3/L4 clinical: 0 / 0 / 4.

Completion evidence. One final changed plan; passed completeness and rendition checks; complete disposition of beneficiary comments; scoped professional approvals where required; preserved refusals/unresolved items; and an L2 final-content confirmation. The artifact is ready for task 44, not yet incorporated or distributed.

What does not prove completion. A merged document; disappearance of tracked changes; an electronic signature without scope; clinician approval of one recommendation; beneficiary approval of text that still contains an unresolved clinical instruction; or successful upload/transmission.

Edge cases/open decisions. CP-E01, CP-E02, CP-E04, CP-E05, CP-E06, CP-E08, CP-E09, CP-E10, CP-E12, CP-E17, CP-E20; O-003, O-004, O-010, O-011, O-015, O-017, O-PA-001.

44. Put the plan in the electronic health record#

When. After task 42 confirms the current plan remains applicable or task 43 produces the final changed plan, and before the new/changed content is treated as available to the GUIDE team. GUIDE RFA Appendix B 2.4 expressly requires EHR incorporation.

What and how much. File exactly one current approved plan in the correct beneficiary EHR location with the correct document type, effective/approval date, and relationship to any prior plan. Confirm that the EHR saved and can retrieve the same approved content. Do not distribute a changed plan while incorporation is unresolved under the current CP-E18 fail-safe.

Data. Use correct beneficiary identity, the final approved plan, designated EHR destination, filing metadata, submission result, and matched retrieval/acknowledgement. A successful upload request proves submission only. A generic "saved" banner, queue acceptance, or acknowledgement for another document does not prove that the exact approved plan is incorporated or retrievable.

Potential Proxi work. Deterministically file the plan, prevent duplicate submission, verify correct beneficiary/document type, reconcile the acknowledgement, retrieve the stored copy or exact EHR confirmation, and surface mismatch/outage. No AI judgment is needed for the routine path.

Human role. No human in the clean path. Health-information/clinical-informatics or authorized administrative staff resolves identity mismatch, wrong-document filing, EHR outage, duplicate/conflicting copies, or missing acknowledgement. The prior plan remains the usable record until the current O-016 policy permits another disposition.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 3*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Exact approved plan is associated with the correct beneficiary, stored in the designated EHR location, retrievable by the GUIDE team, and supported by a matched EHR acknowledgement or equivalent objective incorporation evidence.

What does not prove completion. Final-plan approval; an outbound API success response without matched acknowledgement; a file appearing in a local folder; submission to an interface queue; an acknowledgement for the wrong beneficiary/document; beneficiary delivery; or PCP receipt.

Edge cases/open decisions. CP-E12, CP-E18, CP-E20; O-004, O-016, O-017, O-PA-001.

45. Give the beneficiary the plan and supporting resources#

When. After the current plan is approved and, for a changed plan, after successful EHR incorporation under task 44. Share the initial plan and every future revision as required by GUIDE RFA Appendix B 2.2, using the beneficiary's confirmed accessible channel.

What and how much. Deliver one complete current plan plus the specific approved education, supports, and resources needed to pursue each accepted goal. Include action-owner and contact information, what the beneficiary is expected to do, what Proxi/other owners will do, and where unresolved questions remain. If more than one channel is required for accessibility, create and track each required delivery obligation separately.

Data. Use beneficiary identity, current accessible plan, communication preferences, valid delivery endpoint, relevant resource set, any communication permission/authority facts, and exact send/delivery/access results. Transmission proves an attempt. Channel delivery proves arrival to an endpoint, not identity of the reader, meaningful access, understanding, agreement, or action. Task 39 review does not by itself prove the beneficiary received the final changed document.

Potential Proxi work. Prepare the approved package, apply the selected channel, deliver after prerequisites are satisfied, track delivery/access signals, retry only within approved bounds, offer approved explanations, and route questions or failed access. Routine electronic or print-fulfillment delivery should be software work.

Human role. No human is needed merely to press Send or verify a clean delivery result. L2 answers beneficiary questions, addresses relationship-sensitive barriers, or repeats explanation when requested. Under the current V1 release boundary, a separate authorized-human release may be required; O-014 has not settled the exact role. Beneficiary time is outside Proxi labor.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Current plan and relevant resources sent through the approved accessible channel; objective channel result; meaningful-access evidence according to the eventual O-012 standard; and any failed obligation still visible and owned. Completion here does not mean the plan's actions were carried out.

What does not prove completion. Plan finalization; EHR incorporation; a Send button click; an email accepted by a server; a portal notification without usable access; an old plan received; task 39 review of a draft; or beneficiary acknowledgement interpreted as comprehension or action.

Edge cases/open decisions. CP-E02, CP-E11, CP-E13, CP-E14, CP-E15, CP-E17, CP-E20; O-004, O-012, O-014, O-015, O-017, O-PA-001.

46. Share with the caregiver when permitted#

When. After task 44 and only when there is an identified caregiver, the beneficiary has chosen caregiver involvement, and the current permission/legal basis permits the exact content and purpose. Repeat for each revision only after rechecking changed permissions and recipient facts. GUIDE RFA Appendix B 2.2 requires caregiver sharing in the beneficiary-directed, applicable caregiver path.

What and how much. For each permitted caregiver, deliver the exact plan or permitted subset plus the caregiver-relevant education, supports, and resources. Keep one obligation per caregiver because permission scope, endpoint, and access result can differ. Do not expose beneficiary-only content merely because the caregiver attended a prior conversation.

Data. Use caregiver identity, beneficiary choice about involvement, decision-specific authority/permission basis, allowed topics/content, purpose, effective/expiry/revocation facts, accessible channel, and delivery/access result. Being listed as a caregiver, emergency contact, portal proxy, or meeting attendee does not automatically prove authority to receive the whole plan. A prior permission does not prove it remains current after revocation or narrowing.

Potential Proxi work. Build the allowed package, exclude nonpermitted content, deliver after release prerequisites, track each caregiver obligation, stop controllable queued work after revocation, and route ambiguity. Proxi cannot invent permission, decide legal authority, or treat caregiver involvement as beneficiary consent.

Human role. L2 authenticates beneficiary choices and resolves relationship-sensitive questions. Privacy/compliance/legal resolves an unclear disclosure basis or representative scope. No human is needed for routine delivery after permission and release are established. Caregiver and beneficiary time is outside Proxi labor.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. One permission-supported caregiver obligation per recipient; exact allowed package; authorized release when required; objective delivery/access result; and unresolved failures or revoked obligations truthfully recorded.

What does not prove completion. Caregiver existence; caregiver assessment; attendance at task 39; an emergency-contact listing; delivery to one caregiver proving delivery to another; a Send event; or permission to discuss one topic interpreted as permission to share the complete plan.

Edge cases/open decisions. CP-E04, CP-E05, CP-E06, CP-E07, CP-E13, CP-E14, CP-E15, CP-E17; O-004, O-011, O-012, O-013, O-014, O-PA-001.

47. Share with an outside PCP#

When. After task 44 when the confirmed current PCP is not a member of the GUIDE care team. Perform for the initial plan and every updated/revised plan. GUIDE RFA Appendix B 2.4 and 5.1 require the outside PCP to have access to the care plan and revised plans.

What and how much. Create one current-plan access obligation for the confirmed outside PCP/practice endpoint. Deliver or enable access through an approved clinical channel and confirm that the exact current plan is meaningfully available to that PCP. If the PCP changes while work is in flight, stop relying on the old obligation and resolve which current recipient obligations remain applicable.

Data. Use beneficiary-confirmed/current PCP facts, GUIDE-team membership, provider identity, verified clinical endpoint, exact current plan, release basis, channel result, and access evidence. Administrative Medicaid assignment, claims utilization, an NPI, directory listing, or a fax number alone does not prove the beneficiary's current PCP or a usable endpoint. Transmission or fax success does not necessarily prove meaningful PCP access; O-012 must define channel-specific satisfaction.

Potential Proxi work. Reconcile candidate PCP sources, request beneficiary confirmation when needed, verify the endpoint, prepare the approved packet, support human release, transmit, track access, and retry within approved policy. L1 may pursue endpoint/receipt facts but cannot decide PCP identity, permission, content sufficiency, or closure.

Human role. No routine human is needed after PCP identity, endpoint, content, authority, and release are confirmed. L1 handles manual endpoint/receipt pursuit. L2 resolves beneficiary relationship facts or changed-PCP ambiguity. Privacy/compliance or the role settled in O-014 releases the V1 packet. Outside PCP time is not Proxi labor.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 3*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Outside-PCP status established; exact current plan tied to the correct beneficiary and recipient; verified endpoint; required release; and channel-specific evidence that the PCP has meaningful current-plan access. Each revision creates a fresh obligation; prior-plan access does not satisfy revised-plan access.

What does not prove completion. NPI match; provider-directory listing; a historical claim; GUIDE-team assumption; generated fax cover sheet; transmission success without the eventual access standard; office receipt of an older plan; or PCP agreement/implementation, which are separate facts.

Edge cases/open decisions. CP-E02, CP-E12, CP-E13, CP-E14, CP-E15, CP-E17, CP-E18, CP-E20; O-004, O-012, O-014, O-016, O-017, O-020, O-021, O-PA-001.

48. Share relevant information with other providers#

When. After task 44 when a specialist or other provider is involved and care coordination under GUIDE RFA Appendix B 2.4 and 5.2 calls for sharing the plan or relevant information. This is recipient- and purpose-specific, not a default broadcast of the full plan.

What and how much. Create one obligation for each relevant provider recipient. Send the whole approved plan only when appropriate; otherwise send the relevant approved portion and coordination context needed for that provider's role. Keep recipient, purpose, content, channel, release, delivery, and access evidence distinct for every provider.

Data. Use current provider involvement, referral/coordination purpose, exact relevant plan content, provider identity and endpoint, permission/legal basis, specialist sensitivity restrictions if applicable, and channel results. Provider appearance in claims or a directory does not prove current involvement. A treatment-disclosure pathway does not by itself decide that every plan element is relevant, that Part 2/state-law issues are absent, or that the recipient endpoint is correct.

Potential Proxi work. Identify candidate recipients from approved care actions, prepare a purpose-limited packet, verify endpoint facts, support release, transmit, and track each obligation. AI may summarize only approved content; it cannot select recipients, decide disclosure basis, or determine that delivery closes a clinical loop.

Human role. No human in the fully confirmed routine path except the V1 authorized release if required. L1 pursues administrative endpoint/receipt facts. L2 verifies relationship-sensitive recipient facts. Privacy/compliance handles an ambiguous basis or sensitive-data restriction. Relevant providers' time is external.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 3*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. One justified recipient obligation per provider; exact approved content and purpose; verified endpoint and release; and channel-specific delivery/access result. An open failure remains open and does not disappear because another provider received the plan.

What does not prove completion. Provider-name extraction; directory contact data; packet preparation; one provider receiving information; transmission alone; receipt interpreted as agreement; or provider receipt interpreted as implementation or referral-loop closure.

Edge cases/open decisions. CP-E02, CP-E09, CP-E13, CP-E14, CP-E15, CP-E17, CP-E20; O-004, O-011, O-012, O-014, O-017, O-021, O-PA-001.

49. Send each accepted action for execution#

When. After task 42/43 establishes the current plan and task 44 establishes EHR incorporation. Start once for every accepted plan action whose prerequisites are satisfied; do not wait for unrelated actions to resolve.

What and how much. Create exactly one executable work request for each accepted referral, service request, caregiver-support need, medication follow-up, clinical coordination item, beneficiary/caregiver action, or internal GUIDE-team action. Each request states the responsible owner, requested action, required context, due/target timing if approved, prerequisite decisions/orders, and expected return evidence. Keep external treatment, beneficiary action, caregiver action, and Proxi internal work distinct.

Data. Use the accepted-action list, beneficiary choice, responsible owner and acceptance where already known, any clinical order/decision, permission/release prerequisites, verified endpoint/work queue, and required result. A care-plan recommendation proves that an option was included; it does not prove an order exists, a provider accepted, coverage is confirmed, the request was sent, or the action was performed.

Potential Proxi work. Deterministically create internal work items and approved outbound handoffs, attach the minimum necessary approved context, schedule reminders, and prevent an unresolved/refused item from being released. AI may draft a handoff summary from approved content but cannot turn a recommendation into an order, select a provider, or mark work complete.

Human role. No human is needed for routine internal routing after every required fact and authority is present. L2 verifies a sensitive or complex relationship-based handoff, resolves beneficiary/caregiver uncertainty, or makes a requested warm introduction. The authorized clinician/order owner acts for clinical work; the external action owner performs the service. An outside-recipient V1 release may require the role settled in O-014.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 1*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. Every accepted action has one work request tied to the correct owner, prerequisites, approved context, and expected result; blocked actions remain blocked with a named dependency; and no accepted action is silently omitted.

What does not prove completion. The action appearing in prose; an owner name without a request; an AI-created task; placing work in the wrong team's queue; sending a clinical recommendation without an authorized order/decision; or any execution result, which belongs to the action owner and later monitoring.

Edge cases/open decisions. CP-E01, CP-E02, CP-E08, CP-E09, CP-E10, CP-E11, CP-E14, CP-E15, CP-E17, CP-E19; O-003, O-004, O-011, O-014, O-015, O-PA-001.

50. Confirm that action owners received their work#

When. Immediately after each task-49 work request reaches its destination, and again after the approved response interval when no usable acknowledgement has arrived.

What and how much. Confirm each action separately. Establish that the correct person or organization received the exact request, understands what action is being requested, and either can proceed, refuses, needs clarification, or is not the correct owner. Route rejection or reassignment without closing the underlying action. For an internal software work queue, a matched acceptance event may satisfy routine receipt; an outside party may require channel-appropriate confirmation.

Data. Use action identity, owner identity, destination, sent content, delivery result, acknowledgement source/time, stated ability to proceed, refusal/clarification/reassignment, and outstanding dependency. Delivery proves arrival, not comprehension or acceptance. Receipt proves neither performance nor outcome. A generic office acknowledgement does not necessarily prove the responsible clinician or program owns the action.

Potential Proxi work. Match acknowledgements to actions, send routine reminders within approved policy, identify nonresponse, and route exceptions. L1 can make administrative confirmation calls and record verbatim outcomes. AI cannot infer understanding from sentiment, decide reassignment, interpret a clinical dispute, or close the action.

Human role. No human in the clean matched-acknowledgement path. L1 performs manual administrative pursuit when the endpoint is unconnected or acknowledgement is missing. L2 resolves relationship-sensitive refusal or owner ambiguity. A clinical dispute routes to the authorized clinician rather than L1/L2 interpretation.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 3*; Puerto Rico L2 navigator: 0 / 0 / 0; Separate U.S. L3/L4 clinical: 0 / 0 / 0.

Completion evidence. For every action, a matched acknowledgement with owner identity and one of the allowed factual outcomes: can proceed, refuses, needs clarification, or wrong owner/reassignment required. Non-success remains an owned open condition; it is not transformed into completed receipt.

What does not prove completion. Send event; delivery receipt alone; voicemail left; front-desk receipt without ownership; reminder sent; action-owner name in the plan; acknowledgement for a different action; or evidence that the action was performed.

Edge cases/open decisions. CP-E09, CP-E13, CP-E17, CP-E19, CP-E20; O-004, O-012, O-014, O-017, O-PA-001.

51. Follow up on goals and open actions#

When. During the applicable ongoing navigator contact, at an action-specific follow-up date, after a reported outcome, after missed acknowledgement/performance, or when new beneficiary/caregiver/provider information could affect a goal or action. GUIDE RFA Appendix B 4.2 requires ongoing contact to maintain/revise the plan, identify unmet needs, coordinate services, monitor medication management/adherence, and support caregivers.

What and how much. Review every due/open action and each goal affected by new information. Ask what occurred, whether the service or support was actually obtained, whether it helped, what barrier remains, and whether goals, preferences, circumstances, needs, caregiver situation, or clinical status changed. Record one factual outcome for each due action and one beneficiary-reported progress statement for each reviewed goal. Create follow-up, escalation, or revision work without claiming outcomes not evidenced.

Data. Use due-action list, prior acknowledgement, provider/program result, beneficiary/caregiver report, current barriers, goal baseline, clinical measurements where applicable, and new safety/symptom signals. A beneficiary report is valid evidence of lived experience but does not automatically prove a billed external service, clinical effectiveness, or provider performance. Provider documentation proves what it states, not beneficiary understanding or implementation. No response does not mean no change.

Potential Proxi work. Prepare the due-item agenda, ask approved structured questions, capture verbatim updates, compare reported outcomes with the plan, identify missing results, and route safety, symptom, medication, clinical, legal, or authority triggers. Proxi may summarize but cannot determine clinical effectiveness, risk, urgency, service qualification, or closure.

Human role. L2 conducts the required relationship-based portion of the follow-up, authenticates preference and lived-outcome reports, and handles sensitive barriers. L1 may pursue purely administrative status between contacts. L3/L4/L5 addresses clinical outcomes, new symptoms, safety issues, or treatment questions within scope. Beneficiary/caregiver/provider time is excluded.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 2*; Puerto Rico L2 navigator: 2 / 2 / 4; Separate U.S. L3/L4 clinical: 0 / 0 / 10.

Completion evidence. Every due action has a dated factual disposition or remains visibly open with next owner/date; reviewed goals have beneficiary/caregiver outcome information; barriers and new facts are captured; clinical/safety triggers are routed; and any plan-revision trigger is handed to task 52.

What does not prove completion. Reminder sent; action owner acknowledged receipt; appointment scheduled; referral delivered; no complaint; automated claim appearance without matching context; or a navigator note saying "doing well" without action- and goal-specific evidence.

Edge cases/open decisions. CP-E02, CP-E04, CP-E07, CP-E08, CP-E09, CP-E10, CP-E11, CP-E15, CP-E19, CP-E21; O-003, O-004, O-010, O-013, O-017, O-PA-001.

52. Revise the plan when needed or requested#

When. Whenever the beneficiary requests a modification or new evidence suggests changed circumstances, goals, preferences, needs, caregiver situation, provider/service situation, safety, treatment, medication, or action feasibility. This is the re-entry task for a later planning episode; it does not replace task 40's edits within an already active beneficiary review. GUIDE RFA Appendix B 2.2-2.3 requires beneficiary-led future revisions and modification as needed or requested.

What and how much. Authenticate the trigger, identify every potentially affected plan section and downstream action/recipient, and return those sections to the beneficiary-led planning sequence. Review all affected content, not just the sentence that generated the alert. Produce either a changed-plan path through tasks 40, 43, and 44-50 or a compliant reviewed-no-change path through task 42. Do not mutate the prior approved plan silently while review is pending.

Data. Use the beneficiary/caregiver/provider request or new event with provenance, current plan, affected goals/actions, materiality facts, current authority/permission, and any clinical/safety evidence. Detection proves only that a potentially relevant event exists. It does not prove the plan must change, the event is accurate/current, the requester controls the decision, or the old plan is unsafe. A reopened work item does not prove revision completion.

Potential Proxi work. Detect possible change triggers, link them to affected plan sections, request missing facts, prepare a comparison, preserve the prior plan, and reopen the appropriate tasks. AI cannot decide materiality, validate the request, approve revised content, choose a clinical response, or close the revision from a summary.

Human role. L2 authenticates beneficiary/caregiver direction, leads the renewed planning conversation, and resolves nonclinical changes. L3/L4/L5 reviews only affected clinical elements within scope. Legal/privacy resolves authority or disclosure changes. L1 may pursue administrative source documents but does not decide the revision.

Provisional human minutes (low / typical / high). Philippines L1: 0 / 0 / 1*; Puerto Rico L2 navigator: 0 / 0 / 1 for trigger scoping; affected tasks then carry the revision labor; Separate U.S. L3/L4 clinical: 0 / 0 / 10.

Completion evidence. Authenticated trigger and affected-section list; renewed beneficiary-led review; every change or no-change question dispositioned; and either (a) a final changed plan carried through EHR incorporation and required distribution/action updates, or (b) a reviewed-no-change record that meets the future O-010 standard. Opening the revision episode is intermediate evidence only.

What does not prove completion. Alert creation; new data arrival; beneficiary request intake; AI summary; reopening a task; editing one sentence; clinician response without beneficiary-led integration; or a revised document that was not incorporated and distributed as required.

Edge cases/open decisions. CP-E02, CP-E03, CP-E04, CP-E05, CP-E06, CP-E07, CP-E08, CP-E09, CP-E10, CP-E11, CP-E15, CP-E17, CP-E19, CP-E20, CP-E21; O-003, O-004, O-010, O-011, O-012, O-013, O-014, O-015, O-016, O-017, O-PA-001.

Cross-pillar action and result integration#

This SOP participates in cross-pillar episodes under 25_GUIDE_Eight_Pillar_Service_Integration.md. Use the shared no-repeat packet, preparation-before-judgment route, receiving-owner acceptance rule, and count-once labor rule. Care Plan receives service attribution only for a beneficiary-led current plan review, choice, accepted action, reviewed-no-change result, or focused revision with attributable professional content, a named owner, and truthful open dependencies.

One planning conversation may settle several goals and actions and may coincide with another service, but Care Plan attribution requires actual plan review, choice, action, reviewed-no-change, or revision work. Shared human minutes are counted once; downstream labor and results remain with their owning pillars.

Minimum semantic tests applied across the 52 tasks#

Every task-specific review must include the relevant tests below in addition to the examples named on the card. These are semantic acceptance tests for the human SOP; they are not executable automation specifications.

Open Care Plan decisions exposed by the detailed pass#

The cards intentionally leave unresolved authority, timing, release, distribution, applicability, and evidence choices linked to O-003, O-004, and O-010 through O-018. The detailed pass also creates four specific owner decisions in 08_Decision_Log.md: digital or AI-assisted explanation and choice evidence (O-041), caregiver-originated sensitive information (O-042), accessible-rendition quality (O-043), and software option-set completeness, ranking, and freshness (O-044). Until those owners decide, the cards describe permitted preparation and the conservative human route; they do not pretend the disputed clean path is settled.