Caregiver education and support

Status: TASK BREAKDOWN DRAFT

Public service anchor: CMS GUIDE RFA Appendix B §§8.1-8.5 ↗, together with the caregiver assessment and no-caregiver provisions in §§1.2.1-1.2.3. The Proxi column identifies candidate automation, not functionality already deployed. The acquired Participant's current executed agreement still needs to be checked before operations are finalized. Respite is not included.

Detailed caregiver-support procedure — all 34 task rows#

The detailed cards control when a shorter inventory row appears to imply a different human or automation boundary. CG01, CG01A, and CG02-CG09 establish applicability and the caregiver-support foundation; CG10-CG33 below complete the training, diagnosis-information, support-group, one-on-one support, optional-support, counseling-referral, and reassessment work.

Governing boundary and current public sources#

The public requirement basis is the CMS GUIDE Request for Applications ↗, Appendix B §§1.2.1–1.2.3.1 and 8.1–8.5, together with the current CMS GUIDE FAQ ↗. CMS requires a caregiver support program responsive to the caregiver assessment and ongoing needs. It must offer skills training, dementia-diagnosis information, support groups, and ad hoc one-on-one support calls. CMS permits these services virtually or in person, but specifies who may furnish particular services.

For this pillar:

• When no caregiver is confirmed after reasonable efforts, caregiver assessment and caregiver education/support do not apply; additional efforts and safeguards must instead be carried through other GUIDE services.
• Caregiver skills training must be provided by a care-team member or a contracted vendor/community organization reimbursed by the GUIDE Participant, and may be one-on-one or group, virtual or in person.
• The public CMS materials do not prescribe a synchronous instructor for every topic, a minimum number of minutes or sessions, a quiz, passing score, certificate, formal competency examination, or teach-back as a CMS completion test.
• CMS does not expressly say that self-directed text, video, or AI alone satisfies required caregiver skills training. D-019 therefore uses software/AI for the main course and one permitted, competency-verified human application touchpoint; Participant approval of this design and the executed Participation Agreement remain open under O-036.
• Approved text, video, audio, translation, and bounded text/voice AI may present the course, answer in-scope questions, run examples, collect the caregiver's intended use, and prepare the focused human touchpoint. AI-only delivery does not close the required training under the current Proxi design.
• A required one-on-one diagnosis-information conversation and required ad hoc one-on-one caregiver support call are later tasks and must be provided directly by a human care-team member; CG06 must not accidentally claim either service from a menu explanation.

The executed Participation Agreement remains absent under O-PA-001, so participant-specific completion and Partner Organization rules remain provisional.

Relevant registered sources are SRC-CMS-GUIDE-RFA-V1, SRC-CMS-GUIDE-FAQ, SRC-CMS-GUIDE-PMP-3.0, SRC-GUIDE-PA-PY2026, SRC-ALZ-DCN-TRAINING, SRC-ALZ-DCN-OBJECTIVES, and SRC-AHRQ-TEACH-BACK-3E in 09_Source_Register.md. The Alzheimer’s Association training sources support broad nonclinical dementia navigation, caregiver well-being, communication, safety, ADL, planning, and coordination capability; they do not create clinical licensure. The AHRQ source supports a non-shaming plan-back/show-me design; it is not a GUIDE mandate.

Settled Proxi training route#

Clear authenticated offers, choices, and approved self-paced work proceed without mandatory L2 review. L2 enters for the actual required or requested human service or another D-021 relationship condition. Software enforces the approved privacy and disclosure rules. Protected clinical, medication, behavioral-health, and safety questions route directly to the authorized professional; only an unresolved legal-authority question routes to Healthcare Legal Counsel.

The training is not a series of ten human lectures and it is not a quiz. The caregiver moves through approved short video, text, audio, and text/voice-AI learning units at their own pace. The units cover all ten CMS topics, with depth and ordering tailored to the caregiver assessment. A question such as “My mom does not want to bathe—what do I do?” opens a focused mini-series with examples and bounded AI conversation. The caregiver may decline or defer a topic, but the record must say so rather than claiming the full course was completed.

One Puerto Rico GUIDE navigator may perform the focused training-application task on the Proxi internal route only for each topic covered by that navigator's current verified subject-specific competency record. This is a permitted task within the Puerto Rico GUIDE navigator job category, not a separate internal job. Assignment software must block a navigator from any topic outside that verified scope. The assigned navigator reviews the software summary, asks what the caregiver plans to try in the real situation, listens for a usable nonclinical plan and help route, corrects misunderstandings, and records what remains unresolved. This is plan-back or show-me: a check that Proxi explained the material clearly, not an examination of the caregiver. A Puerto Rico GUIDE navigator may facilitate an approved group as the same human touchpoint only when the navigator's competency is verified for that topic and format and participant-level evidence exists. When the required navigator competency is absent, protected clinical content routes to the applicable U.S. clinician and other training work routes to an approved reimbursed contracted training provider or community organization through the permitted outside-Proxi route. If neither permitted route is available, software must prevent assignment and completion, retain a truthful unavailable or blocked status, and follow the approved retry/escalation rule. The approved outside provider is not a Proxi job, and that route remains OPEN where approval is missing under O-036. Software enforces approved privacy and disclosure rules. Physical-technique, individualized medical, medication, behavioral-health, and immediate-safety questions leave the L2 lane for the properly authorized professional; only an unresolved legal-authority question routes to Healthcare Legal Counsel.

If Proxi supplies only the technology, the GUIDE Participant or another permitted organization owns and furnishes the training. If Proxi furnishes the managed training service, the required Participant/Partner Organization approval, arrangement, reimbursement, and documentation access must be in place before service delivery.

Caregiver pulse as an entry route#

The permission-scoped Proxi Caregiver View may offer a brief optional caregiver pulse: how the caregiver is holding up, whether practical help would be useful, and whether the caregiver wants a routed human follow-up request. The answer remains the caregiver's own current report and keeps its source and time. Software may acknowledge it, retrieve related needs/actions, prepare the smallest useful packet, and raise configured stop candidates for routed review; it does not score or diagnose burden, decide urgency, convert caregiver observation into beneficiary fact, promise synchronous availability, or guarantee immediate human help.

A routine request may open CG25 and then the human CG26 support conversation. A material caregiver change may open CG33. A possible clinical, safety, safeguarding, or emergency issue follows CG27 and the authorized route; the pulse is not emergency response or clinical review. The pulse itself completes none of those tasks and never becomes a beneficiary fact, burden diagnosis or score, urgency decision, completed caregiver or clinical assessment, completed one-on-one support call, completed service, or payment evidence. Silence and stale or conflicting information remain unknown and never provide reassurance, including in Caregiver View. D-023 settles these exclusions; O-061 and O-042 govern activation mechanics and information handling only.

Evidence and completion language#

• Direct evidence establishes only the fact its source is entitled to establish. A beneficiary statement establishes that person’s current statement; a potential caregiver’s response establishes that person’s willingness; a completed training record establishes the recorded service only when the permitted provider, content, participation, and learning evidence are present.
• Corroborating evidence supports or challenges a candidate fact but does not establish it alone.
• Identity-only evidence helps match a beneficiary, caregiver, potential caregiver, training provider, vendor, or service. It does not establish caregiver status, permission, willingness, training qualification, participation, understanding, or completion.
• Delivery is not receipt; receipt is not understanding; viewing is not training; a quiz score is not demonstrated ability; scheduling is not attendance; attendance is not completion; and an offer is not acceptance.

Each card’s completion evidence proves only that named task. It does not prove that any later caregiver service was furnished, that every required offer was made, or that GUIDE service/payment evidence is satisfied.

Provisional workload and mutually exclusive routes#

All minute values are Proxi workload-calibration assumptions, not CMS facts, mandated contact lengths, staffing guarantees, or field observations. Automated processing time and beneficiary, caregiver, potential-caregiver, interpreter, approved contracted training-provider/vendor, community organization, and emergency-service time are excluded. Contracted training providers, vendors, and community organizations are outside Proxi.

The caregiver-present foundation route uses 0 / 9 / 18 L2 minutes for clear-direct / ordinary human-review / complex human-review routes across CG03 and CG05. The table below shows the ordinary typical allocation. If the caregiver accepts training, CG09 allocates the first 2 typical L2 minutes inside the single 10-minute application touchpoint, producing an 11-minute typical caregiver-present route through CG09. CG09 is not an extra navigator contact; if it occurs during another scheduled contact, the same active minutes are counted once.

The no-caregiver branch is mutually exclusive. CG01A uses 8 L2 minutes for the reasonable human effort; CG02’s routing is software and adds zero. Do not add the 8-minute no-caregiver branch to a caregiver-present route for one beneficiary episode.

Use one shared 13-minute L1 Philippines administrative exception event only when manual contact cleanup, record retrieval, vendor registration, accessibility logistics, or failed-route recovery is triggered. Routine CG07 offering, CG08 scheduling, and content delivery remain zero-human. Clinical or immediate-safety events are additional and use the appropriate separate U.S. L3/L4/emergency pathway; they are not hidden inside these clean-path minutes.

CG01. Confirm that the caregiver-support pillar applies#

When. At initial caregiver-service setup; after each current caregiver assessment; and whenever caregiver identity, assistance, willingness, beneficiary permission, residence, or relationship changes.

What and how much. Produce one truthful applicability result: one current caregiver; multiple current caregivers; no caregiver confirmed; caregiver status undetermined; or caregiver identity/participation disputed. Identify every person currently reported to provide unpaid ADL/IADL assistance and keep each person separate. Applicability does not establish representative authority or disclosure permission.

Data. Direct: authenticated beneficiary report of who assists; each caregiver’s statement about assistance and participation; and a current caregiver-assessment record attributed to that person. Corroborating: prior assessments, care notes, shared appointments, and observed assistance. Identity-only: emergency-contact record, family relationship, shared address, portal proxy, next-of-kin field, or phone possession does not establish caregiver status, willingness, permission, or authority.

Potential Proxi work. Retrieve caregiver records, normalize duplicate identities, ask structured confirmation questions through accessible text/voice, retain each speaker’s statement, compare with prior status, and present the supported applicability result. AI cannot declare caregiver status from a contact field or choose whose conflicting statement controls.

Human role. Beneficiary and each caregiver supply/confirm the facts; their time is not Proxi labor. No Proxi human is needed on an exact clean path. L2 handles relationship conflict or unclear involvement. Software enforces the approved authority and disclosure rules; if those rules expose an unresolved legal-authority question, hold the affected action and route that exact question to Healthcare Legal Counsel.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 0; L4 0.

Completion evidence. Applicability result; each caregiver identity and reported assistance; source/speaker/date; current willingness/participation; separate beneficiary permission and representative status; and explicit undetermined/conflict status.

What does not prove completion. Emergency contact, shared residence, family relationship, portal access, old caregiver record, caregiver answering the phone, claim/service data, or AI inference.

Edge cases/open decisions. X-001, X-002, X-003, X-004, X-005, X-009, X-015, CP-E04, CP-E05, CP-E06, CP-E07, O-011, and O-013. Test one caregiver; multiple caregivers; no caregiver; status unknown; duplicate identity; caregiver unwilling; beneficiary disputes role; stale assessment; later caregiver change.

CG01A. Make a reasonable effort to identify a caregiver#

When. After CG01 confirms no caregiver or leaves caregiver status undetermined and before CG02 closes caregiver-dependent work. Reopen when the beneficiary proposes a new person or a potential caregiver later responds.

What and how much. Conduct one respectful, beneficiary-centered reasonable-effort episode: explain why a caregiver may help; ask whether the beneficiary wishes to identify relatives or unpaid nonrelatives; obtain permission for each outreach; contact permitted candidates through approved channels; record each response; and stop when the beneficiary declines, no permitted candidates remain, a caregiver accepts, or the future settled policy defines an unsuccessful endpoint. Do not pressure the beneficiary or recruit someone without permission.

Data. Direct: beneficiary’s current choice and permission; potential caregiver’s own response about ability/willingness; verified contact response; and established representative authority when relevant. Corroborating: prior contacts, care notes, community relationships, and beneficiary-suggested candidates. Identity-only: contact lists, social-media/public data, family relationship, shared address, or emergency-contact fields identify possibilities but do not establish permission, suitability, or willingness.

Potential Proxi work. Organize only permitted candidates, prepare approved outreach, schedule callbacks, track attempts/responses, prevent duplicate contact, and surface the unresolved result. Proxi cannot decide that someone is suitable, willing, safe, or authorized.

Human role. Before L2 enters, software/L1 prepares the permitted-candidate and outreach packet: known candidate identity/contact provenance, prior permissions/refusals, duplicate attempts, allowed channels, and exact unanswered choices. L2 conducts only the sensitive beneficiary discussion and relationship-based outreach, calibrated at eight minutes; L2 does not gather or clean the candidate list live. The beneficiary chooses whether to identify/permit contact; each potential caregiver decides participation. L1 may perform bounded contact cleanup under the shared exception but does not recruit or resolve relationship concerns. Prepared clinical/safeguarding questions route directly to the authorized owner.

Provisional clean-path Proxi human minutes by role. L1 0; L2 8; L3 0; L4 0 for the no-caregiver reasonable-effort episode.

Completion evidence. Beneficiary discussion; permission/refusal by candidate; permitted candidate list; dated outreach attempts/channels; each response; accepted caregiver or truthful unsuccessful/undetermined outcome; and any safety/authority issue routed separately.

What does not prove completion. One unanswered call, a family list, finding a relative online, emergency-contact status, beneficiary silence, a candidate’s identity without response, or Proxi deciding someone “should” serve.

Edge cases/open decisions. X-001, X-002, X-003, X-004, X-005, X-006, X-008, X-014, X-015, CP-E05, CP-E06, CP-E07, O-004, O-011, O-013, and O-PA-001. Test beneficiary declines outreach; no candidates; candidate unreachable; candidate declines; multiple accept; suspected coercion/abuse; revoked permission mid-outreach; late acceptance; no response under unsettled attempt policy.

CG02. Route a no-caregiver beneficiary to additional safeguards#

When. After the reasonable-effort episode truthfully concludes that no willing caregiver was identified. Do not use this route while caregiver status is merely unknown or outreach remains open.

What and how much. Stop opening caregiver-assessment and caregiver-education/support work for that beneficiary, preserve the no-caregiver finding, and create owned needs in the appropriate monitoring, care-plan, coordination, clinical, social-work, or other GUIDE service. Do not select the safeguards inside this routing task and do not invent a caregiver substitute.

Data. Direct: completed CG01A evidence and the beneficiary’s current participation/authority facts. Corroborating: assessment findings, function, safety, social needs, current supports, and prior plans that inform later safeguard work. Identity-only: lack of caregiver contact, living alone, diagnosis, tier, or residence does not by itself prove the reasonable-effort obligation was completed.

Potential Proxi work. Recognize the confirmed no-caregiver outcome, stop inapplicable caregiver work, route each identified need to the proper service/owner, preserve links, and monitor whether the receiving work was accepted. Proxi may not decide which individualized safeguards are sufficient.

Human role. No Proxi human is needed for deterministic routing. L2 and the appropriate care-team professionals perform the downstream beneficiary support/safeguard work under O-013; those minutes belong to those tasks, not CG02.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 0; L4 0 in this routing task.

Completion evidence. Confirmed no-caregiver source; caregiver-dependent work stopped/not opened; each added-support need and receiving owner/service; acceptance or unresolved routing status; and beneficiary communication appropriate to the situation.

What does not prove completion. A no-caregiver checkbox, living alone, closing all caregiver tasks without downstream work, generic “monitor more” note, assigned owner without acceptance, or an invented proxy contact.

Edge cases/open decisions. X-003, X-004, X-005, X-008, X-010, X-014, X-015, CP-E07, O-004, and O-013. Test capable beneficiary alone; caregiver status still unknown; downstream owner rejects; safety concern; later caregiver identified; beneficiary declines a safeguard; duplicate route; no-caregiver result later corrected.

CG03. Receive and review the caregiver assessment#

When. For every identified participating caregiver when a current caregiver assessment is available and whenever an updated assessment or caregiver correction arrives.

What and how much. Review each caregiver separately across ability and willingness to help, knowledge, needs, social supports, well-being, stress, and other challenges. Identify what is current, changed, unknown, declined, or concerning without diagnosing the caregiver or choosing services for them.

Data. Direct: that caregiver’s authenticated assessment responses and corrections; a configured instrument’s item responses/score establish only what the instrument recorded; qualified clinical/behavioral findings establish the author’s conclusion within scope. Corroborating: prior assessment, navigator notes, accepted supports, and observed participation. Identity-only: caregiver contact record, relationship, shared address, or beneficiary report alone does not establish the caregiver’s ability, willingness, stress, or needs.

Potential Proxi work. Retrieve the correct assessment, keep caregivers and beneficiary information separate, compare item-level changes, summarize stated needs, preserve original responses, detect missing/stale/conflicting items, and flag configured distress/safety cues for the appropriate human pathway.

Human role. Software first prepares the caregiver-specific assessment delta, original responses, changed/unknown items, and candidate priorities. The caregiver may confirm that prepared summary directly on a clear accessible path; their time is not Proxi labor. L2 conducts a four-minute human review only when required/requested or when ambiguity, distress, comprehension failure, disagreement, accessibility failure, or relationship-sensitive meaning remains. Separate U.S. L3/L4 or behavioral-health/safeguarding professionals receive the prepared exact question only on a named clinical or safety trigger.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0 / 4 / 8 for clear direct / ordinary human-review / complex human-review routes; L3 0; L4 0. Do not add the human route unless its entry condition actually occurs.

Completion evidence. Caregiver identity; assessment date/source; all named domains; changes and unknowns; caregiver-confirmed summary; priorities to be elicited; configured clinical/safety cues and routing; and separate records for multiple caregivers.

What does not prove completion. Score alone, AI summary, old assessment, combined household record, beneficiary-only report about caregiver stress, contact information, or absence of a flagged alert.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-008, X-009, X-015, CP-E04, CP-E05, CP-E07, O-003, O-004, and O-011. Test stale assessment; multiple caregivers; caregiver/beneficiary conflict; missing item; declined answer; high distress; suicidal/abuse cue; late correction; wrong caregiver linkage.

CG04. Identify caregiver priorities#

When. After CG03 presents the current assessment in an understandable form and whenever the caregiver’s needs or preferences change.

What and how much. Ask the caregiver which problems or support needs should be addressed first, what outcome they want, preferred format/timing, and what they do not want now. Preserve ties, uncertainty, deferral, and decline. Priorities control tailoring and sequence; they do not erase the obligation to offer all four required service types.

Data. Direct: the caregiver’s authenticated current priority choices and corrections. Corroborating: caregiver-assessment findings, prior choices, participation history, and requested help. Identity-only: stress score, service match, demographics, or AI ranking does not establish priority.

Potential Proxi work. Conduct an accessible structured text/voice intake, show the assessment-derived candidate needs, let the caregiver rank/select/describe, preserve free text, read back the summary, and flag conflict or unclear comprehension. AI may summarize but may not choose the priorities.

Human role. The caregiver chooses; their time is not Proxi labor. No Proxi human is needed on a clear accessible digital path. L2 enters on caregiver request, distress, comprehension failure, sensitive conflict, or inability to use the interface. The confirmed priorities feed CG05.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 0; L4 0.

Completion evidence. Caregiver-authenticated priority list; desired outcomes; format/timing preferences; deferred/declined items; source/date; confirmation; and unresolved conflict or support request.

What does not prove completion. AI ranking, highest assessment score, navigator assumption, service utilization, caregiver silence, default list order, or beneficiary preference substituted for the caregiver’s own support priority.

Edge cases/open decisions. X-003, X-004, X-005, X-008, X-009, X-015, CP-E04, CP-E05, and O-011. Test multiple equal priorities; no priority chosen; caregiver distress; beneficiary/caregiver conflict; inaccessible format; changed choice; multiple caregivers with different priorities; AI summary correction.

CG05. Build the caregiver-support plan#

When. After a current caregiver assessment and priority list exist, and whenever caregiver needs, willingness, accessibility, availability, or service choices change.

What and how much. Build one support plan per caregiver that uses assessment findings and priorities to tailor sequence, format, provider route, and follow-up while ensuring all four CMS-required services are meaningfully offered: caregiver skills training, dementia-diagnosis information, support groups, and ad hoc one-on-one support calls. Record accepted, declined, deferred, needs-explanation, or unresolved status separately for each. Add optional supports only when responsive and available.

Data. Direct: current caregiver assessment; caregiver-confirmed priorities and choices; current approved service catalog; provider/vendor availability; communication/accessibility needs; and clinician-approved constraints when a clinical issue shapes an option. Corroborating: prior participation, preferred formats, schedule patterns, and care-plan needs. Identity-only: a resource match, service listing, or prior attendance does not establish current fit, availability, acceptance, or completion.

Potential Proxi work. Match confirmed needs to approved options, show all four required offers, explain why options may respond to stated needs, propose accessible delivery formats, check for omissions, and draft follow-up. Proxi may not choose for the caregiver, suppress a required offer, or resolve a clinical concern.

Human role. Software first prepares the caregiver-assessment delta, confirmed priorities, all four required offers, current provider/format facts, and a candidate support plan. The caregiver chooses and may directly confirm the prepared plan on an approved clear path; their time is not Proxi labor. L2 uses five minutes only for a required/requested human review, ambiguity, distress, comprehension failure, disagreement, accessibility failure, or relationship-sensitive tailoring. Separate U.S. L3/L4 receives a prepared exact question only when a clinical, behavioral-health, or safety issue changes the permitted support.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0 / 5 / 10 for clear direct / ordinary human-review / complex human-review routes; L3 0; L4 0. Do not add the human route unless its entry condition actually occurs.

Completion evidence. Caregiver-specific plan; source assessment/priorities; all four required service offers represented; choice/status for each; optional responsive supports; delivery preferences; owners/follow-up; and unresolved clinical/accessibility issues.

What does not prove completion. AI recommendation list, plan with fewer than four required offers, caregiver priority alone, generic brochure, one accepted service, or completed plan treated as furnished training/support.

Edge cases/open decisions. X-003, X-004, X-005, X-009, X-010, X-015, CP-E04, CP-E05, CP-E10, CC-E07, O-003, O-004, O-011, O-036, and O-PA-001. Test one required offer omitted; caregiver declines all; different caregiver plans; inaccessible format; service unavailable; clinical issue; changed priority; late assessment correction.

CG06. Explain available caregiver services#

When. Before asking the caregiver to choose among services, whenever a service description or delivery option changes, and whenever the caregiver requests clarification.

What and how much. Explain the four required services and available optional supports in accessible plain language: what each service is, who provides it, available formats, how to request/use it, and the caregiver’s right to accept, decline, defer, or later change their choice. This is a service-menu explanation, not diagnosis information, a support call, support-group participation, or completed skills training.

Data. Direct: the Participant’s current approved service catalog, actual permitted provider routes, current formats/availability, and caregiver communication/accessibility needs. Corroborating: common questions and prior successful format. Identity-only: marketing page, public course, generic community listing, or AI-generated description does not establish the Participant’s current service offer.

Potential Proxi work. Deliver approved text, video, audio, translated content, and bounded text/voice AI explanations; answer only knowledge-base-supported program questions; compare formats; repeat/rephrase; offer a human; and record questions/choices. Unknown or individualized clinical content is routed rather than improvised.

Human role. No Proxi human is needed when the caregiver can use the approved explanation and confirms the next choice. L2 is available on request or when comprehension, sensitivity, conflict, or scope remains unclear. Clinical questions go to the authorized clinician. Caregiver time is outside Proxi labor.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 0; L4 0.

Completion evidence. Current accessible menu delivered; all four required services explained; provider/format information; caregiver questions and answered/routed status; acknowledgement of how to request a person; and next choices or explicit deferral.

What does not prove completion. Sent link, opened message, video percentage viewed, chatbot response, generic brochure, silence, or menu explanation counted as diagnosis information, support call, group participation, or skills training.

Edge cases/open decisions. X-003, X-004, X-005, X-007, X-008, X-009, X-015, CC-E39, O-003, O-004, O-036, and the proposed content-approval decision. Test wrong language; inaccessible video; stale service description; unsupported AI question; caregiver requests person; clinical question; actual safety disclosure; repeated failed comprehension.

CG07. Offer caregiver skills training#

When. After the caregiver understands the service menu and whenever a caregiver who previously declined or deferred asks again or has changed needs.

What and how much. Make one meaningful, accessible offer of caregiver skills training that identifies required topic coverage, available one-on-one/group and virtual/in-person formats, permitted provider route, how to participate, and the caregiver’s right to accept, decline, defer, or later change their choice. Do not treat a prechecked box or buried link as the offer.

Data. Direct: the approved training offer/catalog; current permitted internal navigator or outside contracted-training-provider route; caregiver assessment and communication needs; and the caregiver’s authenticated acceptance/decline/deferral. Corroborating: prior interest, questions, and participation. Identity-only: email delivery, catalog presence, or a public course listing does not establish a meaningful offer or caregiver choice.

Potential Proxi work. Present approved topics/formats, tailor ordering and examples without suppressing required topics, answer bounded questions, record acceptance/decline/defer, preserve date/source, and reopen after a later change.

Human role. The caregiver chooses; their time is outside Proxi labor. No Proxi human is needed on the clean path. L2 enters only when requested or when comprehension, relationship, or accessibility remains unresolved. The Puerto Rico GUIDE navigator's later training-application task is not needed to make the offer.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 0; L4 0.

Completion evidence. Dated accessible offer; required topic/format information; permitted provider route; caregiver-authenticated acceptance, decline, or defer; questions/routing; and later-change mechanism.

What does not prove completion. Catalog availability, sent message, generic consent, assumed acceptance, training schedule, video link, or completed caregiver assessment.

Edge cases/open decisions. X-003, X-004, X-005, X-007, X-009, X-015, CC-E19, CC-E39, O-004, O-036, and O-PA-001. Test accepts; declines; defers; later accepts; wrong language; no response; caregiver identity mismatch; multiple caregivers choose differently; service temporarily unavailable.

CG08. Schedule and arrange skills training#

When. After the caregiver accepts skills training and chooses a permitted format/provider route. Reschedule routine conflicts without treating them as a new training offer.

What and how much. Create one usable training arrangement: permitted care-team or reimbursed vendor/community-organization provider; one-on-one or group format; virtual or in-person modality; date/time/timezone/location; accessibility/language support; registration; preparation; reminders; and fallback. Do not schedule an unapproved provider as if the required service were arranged.

Data. Direct: approved Puerto Rico GUIDE navigator availability plus current subject-specific competency verification, or an approved outside contracted-training-provider/vendor roster and contractual/reimbursement route; caregiver choice; accessibility needs; and registration confirmation. Corroborating: prior attendance, preferred times, and travel/technology facts. Identity-only: a navigator job title, public course listing, provider biography, calendar entry, website, or community-organization name does not establish competency or permitted GUIDE delivery.

Potential Proxi work. Match approved provider/options, offer times, register, send accessible instructions/reminders, test virtual access, and reschedule routine conflicts. Track preparation and registration separately from attendance and training completion.

Human role. No Proxi human is needed on the connected clean path. L1 uses the shared exception for manual vendor registration, contact, or failed access. L2 handles preference/sensitivity exceptions. A competency-verified Puerto Rico GUIDE navigator's substantive training-application time, or the outside provider's separately measured time, begins at CG09 or later training tasks, not during scheduling.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 0; L4 0.

Completion evidence. Approved provider route; assigned performer; current subject-specific competency or outside-provider approval evidence; registration/booking confirmation; caregiver and format; date/time/timezone/modality; accessibility/language arrangements; preparation/reminders; and unresolved access issue.

What does not prove completion. Offered time, calendar invitation, unconfirmed registration, public webinar, unreimbursed vendor, reminder delivery, attendance, or video access.

Edge cases/open decisions. X-003, X-004, X-005, X-006, X-007, X-013, X-014, X-015, CC-E07, O-004, O-036, and the proposed outside contracted-training-provider/vendor decision. Test no approved provider; vendor unavailable; timezone error; interpreter unavailable; group privacy concern; failed virtual access; cancellation; duplicate registration; caregiver changes format.

CG09. Teach how to use emergency services#

When. When a caregiver accepts skills training and reaches the emergency-services topic, initially and again when the caregiver needs reinforcement, local contact information changes, or teach-back shows the skill is not usable. If a real emergency or safety concern arises, stop training and invoke the approved live pathway.

What and how much. Teach approved general skills for recognizing preapproved emergency situations, contacting emergency services, giving essential location/identity/problem information, following dispatcher instructions, using the beneficiary’s approved emergency plan, and knowing how the GUIDE 24/7 line differs from emergency response. Use scenarios and caregiver practice. Do not make an individualized real-time urgency, diagnosis, treatment, or transport decision inside training.

Data. Direct for training: current approved curriculum; permitted provider identity/route; assigned performer and current subject-specific competency or outside-provider approval evidence; actual content delivered; caregiver participation; scenario responses; teach-back/demonstration; corrections; and unresolved questions. A clinician-approved beneficiary-specific emergency plan is direct for its approved instructions. Corroborating: video viewing, quiz performance, prior calls, and caregiver confidence. Identity-only: navigator job title, content link, attendance record, course title, emergency-contact list, or chatbot transcript does not establish training competency, furnished training, or usable skill.

Potential Proxi work. Deliver approved text/video/audio modules; run bounded voice/text AI scenarios and quizzes; adapt language/format; preserve responses; identify mismatches to the approved answer; prompt repeat practice; verify the assigned Puerto Rico GUIDE navigator's subject-specific competency before preparing that navigator, or prepare the approved outside contracted training provider as applicable; and immediately summon the approved human/emergency pathway on an actual concern. Software must not assign a topic to a navigator whose required competency is absent or expired. AI may not determine actual urgency or close training on its own under current O-036.

Human role. Under the conservative current boundary, a Puerto Rico GUIDE navigator whose current verified competency covers this topic may perform the training-application task, observe practice/teach-back, correct misunderstandings, answer in-scope nonclinical questions, and document the counted topic. Navigator job category alone is insufficient. When competency is absent, protected clinical content routes to the applicable U.S. clinician and other training work routes to an approved reimbursed contracted training provider through the permitted outside-Proxi route; that outside route remains OPEN where approval is missing. Separate U.S. L3/L4/emergency services handle actual clinical/safety situations. Caregiver practice time is outside Proxi labor.

Provisional Proxi human minutes by role (low / typical / high). L1 0 / 0 / 0; L2 1 / 2 / 3; separate U.S. L3 0 / 0 / 0; separate U.S. L4 0 / 0 / 0. This is a topic allocation inside the single caregiver application touchpoint, not an additional navigator contact or a separate emergency-services lecture. Triggered clinical/emergency time is separate under CG27.

Completion evidence. Permitted provider; assigned performer; current subject-specific competency or outside-provider approval evidence; date/modality/format; approved topic/scenarios delivered; caregiver participation; teach-back/demonstration; corrections and final usable response; accessibility; open questions; and any actual concern/handoff recorded separately.

What does not prove completion. Sent content, video viewed, quiz passed alone, AI-only chat/voice session, attendance, course registration, caregiver confidence statement, emergency-plan delivery, or completed emergency handoff.

Edge cases/open decisions. X-003, X-004, X-005, X-007, X-008, X-009, X-014, X-015, CC-E13, CC-E39, O-003, O-004, O-036, O-039, and O-PA-001. Test multimedia plus human teach-back; AI-only module; group training; caregiver cannot demonstrate; language/hearing barrier; outdated local number; hypothetical symptom; actual emergency; failed handoff; caregiver later needs reinforcement.

Remaining detailed procedures — CG10-CG33#

Workload interpretation for the remaining tasks#

The values below are task-local Proxi low / typical / high workload scenarios, not CMS contact-length requirements or field observations. Do not add every high value into one beneficiary episode. A high value applies only when its named exception occurs, and repeated support-group or one-on-one encounters are multiplied only by actual furnished events.

CG09-CG18 are one software-led course with one human-owned application touchpoint. The current calibration is 5 / 10 / 20 L2 minutes total. The allocation is not ten appointments, ten human topic reviews, or an amount added again as ongoing-monitoring contact time. The caregiver's self-paced course time is not Proxi paid-human labor. Approved content and bounded AI perform the main teaching; the human time is concentrated in review, application, correction, and the remaining-needs disposition under D-019.

The caregiver-present foundation uses 0 / 9 / 18 L2 minutes for clear-direct / ordinary human-review / complex human-review routes. When training is accepted, the 5 / 10 / 20-minute aggregate application touchpoint is added once, making the complete foundation-plus-training route 5 / 19 / 38 L2 minutes before any separately triggered clinical, safety, diagnosis-information, support-group, or one-on-one support service. CG09's topic allocation remains inside the aggregate application touchpoint; it is not an additional appointment. When the work occurs inside another qualifying interaction, the shared active minutes are counted once.

The low route fits a caregiver who used the course, has one straightforward question, and can state a usable plan. The typical route includes one practical correction. The high route is remediation, not routine training. If the caregiver cannot use the material after 20 L2 minutes, the result is incomplete with targeted follow-up or qualified-professional referral; it is not silently converted to completion.

L1 is a shared exception lane for manual endpoint cleanup, registration, access, or failed-route recovery—not a default reviewer. L2 supplies nonclinical navigation, teaching, facilitation, and support. Every clinical or medical minute is separate U.S. L3/L4 labor under D-015; L2 never interprets symptoms, determines urgency, diagnoses, treats, or gives individualized medical instruction. External caregiver, beneficiary, interpreter, emergency-service, community-provider, peer-mentor, and contracted-vendor time is not included unless a card explicitly labels an allocated paid-provider share.

CG10. Teach home-safety skills#

When. During the accepted caregiver skills-training event when home safety is applicable, and later when the residence, beneficiary function, equipment, caregiver role, or prior understanding changes. Stop the lesson when a current danger is disclosed.

What and how much. Teach approved general methods for noticing common home hazards, using current care-plan/home-assessment findings appropriately, reducing risk through permitted practical steps, and knowing when to seek help. Use examples and caregiver practice without independently declaring the home safe or prescribing individualized restrictions.

Data/provenance. Direct for training: approved curriculum; current permitted home-safety information; assigned performer and current subject-specific competency or outside-provider approval evidence; exact content delivered; caregiver scenario response/teach-back; and correction by the competency-verified Puerto Rico GUIDE navigator performing the training-application task or the approved outside contracted training provider. Qualified home or clinical assessment findings are direct only for their documented observations and conclusions. Corroborating: caregiver reports, prior incidents, photographs, and checklist responses. Identity-only: address, residence type, equipment list, navigator job title, or generic hazard image does not establish current safety or training competency.

Potential Proxi work. Deliver approved text, video, and checklists; run bounded AI scenarios; tailor examples only from current confirmed facts; collect responses; flag mismatches to approved guidance; repeat content; and summon the appropriate human for an actual or individualized safety concern.

Human role. A Puerto Rico GUIDE navigator whose current verified competency covers home-safety training may confirm application and correct general misunderstandings inside the single application touchpoint. If that competency is absent, assign an approved outside contracted training provider; protected clinical or individualized safety content routes to the applicable U.S. clinician or other authorized U.S. professional. The navigator does not declare individualized safety.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 1 inside the application touchpoint; separate U.S. L3 0 / 0 / 10; separate U.S. L4 0 / 0 / 10. The high clinical values are alternative triggered routes, not routine additions.

Completion evidence. Permitted provider; assigned performer; current subject-specific competency or outside-provider approval evidence; approved topic delivered; relevant source facts; caregiver participation and teach-back; corrections; unresolved individualized questions; accessibility; and any active concern routed separately.

What does not prove completion. Sent checklist, viewed video, completed hazard survey, generic “home safe,” AI-generated recommendation, attendance, or emergency handoff.

Edge cases/open decisions/minimum tests. X-003, X-004, X-005, X-008, X-009, X-015, CC-E39, O-003, O-004, O-036, and O-PA-001. Test generic learning; exposed current hazard; move to a new home; fall/wandering cue; unsafe requested modification; inaccessible content; failed teach-back; and a late home-assessment finding.

CG11. Teach assistance with ADLs and IADLs#

When. During the blended event for the ADL/IADL areas the caregiver currently assists with or expects to assist with, and after a material functional, equipment, professional-instruction, or caregiver-capacity change.

What and how much. Teach approved general assistance principles and topic-specific demonstrations while preserving beneficiary dignity, preferences, independence, and current professional instructions. Separate informational topics from transfer, swallowing, device, wound, medication-administration, or other beneficiary-specific techniques that require an appropriately qualified professional.

Data/provenance. Direct for training: approved curriculum; current assessed ADL/IADL status; clinician or therapy instructions; assigned performer and current subject-specific competency or outside-provider approval evidence; actual caregiver practice; and observation/correction by the competency-verified Puerto Rico GUIDE navigator performing the training-application task or the approved outside contracted training provider. Corroborating: caregiver report, prior assistance, equipment use, and home observations. Identity-only: diagnosis, equipment ownership, service authorization, navigator job title, or caregiver role does not establish current ability, safe technique, or training competency.

Potential Proxi work. Select approved modules tied to confirmed tasks; deliver multimedia demonstrations; run bounded scenarios; collect the caregiver’s explanation; prepare demonstration checklists; record barriers; and stop when pain, weakness, swallowing, falls, a device, medication, or another individualized concern requires a qualified professional.

Human role. A Puerto Rico GUIDE navigator whose current verified competency covers the applicable ADL/IADL topic may teach and correct only general nonclinical principles. If that competency is absent, assign an approved outside contracted training provider; beneficiary-specific or protected clinical instruction routes to the applicable U.S. RN, clinician, OT/PT, speech professional, pharmacist, or other authorized U.S. professional within scope.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 1 / 2 / 3 inside the application touchpoint; separate U.S. L3 0 / 0 / 15; separate U.S. L4 0 / 0 / 10. Triggered rehabilitation/pharmacy time is tracked outside these Proxi role columns when not supplied by Proxi.

Completion evidence. Relevant tasks taught; assigned performer; current subject-specific competency or outside-provider approval evidence; approved source instruction; caregiver practice/teach-back; correction by the competency-verified Puerto Rico GUIDE navigator performing the training-application task or the approved outside contracted training provider; tasks requiring specialist instruction; accessibility; and unresolved safety or functional barriers.

What does not prove completion. Video view, quiz, equipment ownership, caregiver confidence, generic checklist, beneficiary diagnosis, prior assistance, or AI statement that a technique is safe.

Edge cases/open decisions/minimum tests. X-003, X-004, X-005, X-008, X-009, X-015, CC-E39, MED-E41, MED-E43, O-003, O-004, O-036, and the proposed physical-skill scope decision. Test a safe general topic; transfer pain; swallowing concern; inhaler/injection; fall during practice; changed function; conflicting therapist instruction; and inability to demonstrate.

CG12. Teach responses to behavioral and psychosocial symptoms#

When. During the blended training event and later when approved reinforcement is needed. Stop general teaching when the caregiver reports a current severe, escalating, dangerous, suicidal, abusive, or medically concerning event.

What and how much. Teach approved person-centered communication, observation, environmental, routine, and other non-drug approaches for common dementia-related behavioral and psychosocial changes; how to record what occurred; and when to seek qualified help. Do not diagnose cause, grade severity, recommend restraint, change medication, or prescribe treatment.

Data/provenance. Direct for training: approved curriculum; current clinician-approved behavioral plan when applicable; assigned performer and current subject-specific competency or outside-provider approval evidence; exact content delivered; caregiver practice/teach-back; and correction by the competency-verified Puerto Rico GUIDE navigator performing the training-application task or the approved outside contracted training provider. A caregiver report directly establishes what the caregiver reports, not its clinical cause. Corroborating: prior observations, incident patterns, and care notes. Identity-only: diagnosis label, behavior code, navigator job title, or AI sentiment label does not establish cause, severity, safe response, or training competency.

Potential Proxi work. Deliver approved scenarios and non-drug education; conduct bounded roleplay; help capture antecedent, behavior, and context source-faithfully; compare responses with approved guidance; repeat content; and route an actual current concern immediately.

Human role. A Puerto Rico GUIDE navigator whose current verified competency covers the behavioral/psychosocial training topic may rehearse and correct approved nonclinical approaches. If that competency is absent, assign an approved outside contracted training provider; protected clinical content and individualized symptom interpretation, diagnosis, treatment, suicidality, abuse, or danger route to the applicable U.S. clinician, behavioral-health professional, emergency service, or other authorized U.S. professional.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 1 / 2 / 3 inside the application touchpoint; separate U.S. L3 0 / 0 / 15; separate U.S. L4 0 / 0 / 15. A real event ends the training scenario and starts a separate response.

Completion evidence. Approved topic/scenarios; permitted provider; assigned performer; current subject-specific competency or outside-provider approval evidence; caregiver participation and response; correction; help-seeking route; prohibited/advised boundaries; unresolved questions; and actual concerns routed separately.

What does not prove completion. Video, chatbot answer, generic behavior tip sheet, quiz alone, AI behavior classification, attendance, or emergency/clinical handoff.

Edge cases/open decisions/minimum tests. X-003, X-004, X-005, X-008, X-009, X-015, CC-E13, CC-E39, O-003, O-004, O-036, and the proposed behavioral stop-condition decision. Test hypothetical agitation; current escalating aggression; delirium-like change; suicidal statement; abuse disclosure; restraint question; medication question; and failed teach-back.

CG13. Teach how to obtain help across the care continuum#

When. During the blended event and whenever relevant care settings, providers, contacts, beneficiary location, or support routes change.

What and how much. Teach how to use an approved contact guide and seek help from primary, specialty, GUIDE 24/7, urgent, emergency, hospital, post-acute, home, and community services; what administrative information to prepare; and how to use a live clinical/emergency route when directed by approved guidance. This is navigation education, not triage or completed coordination.

Data/provenance. Direct for training: current verified contact guide; approved route descriptions; assigned performer and current subject-specific competency or outside-provider approval evidence; content delivered; caregiver navigation practice; and correction. Corroborating: prior successful contacts and transitions. Identity-only: navigator job title, directory listing, phone number, old plan contact, or website does not establish training competency, a current responsive route, or clinical appropriateness.

Potential Proxi work. Generate and update the guide from verified sources; deliver accessible scenarios; run bounded call/navigation roleplay; collect caregiver choices; detect outdated endpoints; and route an actual current need without deciding urgency.

Human role. A Puerto Rico GUIDE navigator whose current verified competency covers care-continuum navigation training may confirm that the caregiver can use the approved route in general scenarios. If that competency is absent, assign an approved outside contracted training provider. L1 may verify administrative endpoints under the shared exception; protected clinical content or an actual clinical concern routes to the applicable U.S. clinician or emergency service.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 5; L2 0 / 0 / 1 inside the application touchpoint; separate U.S. L3 0 / 0 / 0; separate U.S. L4 0 / 0 / 0. Actual clinical response belongs to CG27 or the linked clinical service, not this teaching task.

Completion evidence. Current guide used; assigned performer; current subject-specific competency or outside-provider approval evidence; settings/routes taught; caregiver scenario response/teach-back; correction; accessibility; outdated-route disposition; and actual need kept separate.

What does not prove completion. Contact guide sent, directory match, memorized number, practice call, real referral, voicemail, or provider receipt.

Edge cases/open decisions/minimum tests. X-003, X-004, X-005, X-006, X-007, X-008, X-015, CC-E07, CC-E13, CC-E39, O-003, O-004, and O-036. Test closed office; wrong number; emergency chosen for a routine example; actual urgent symptom; inaccessible guide; failed teach-back; location change; and late endpoint correction.

CG14. Teach how to work with health and community providers#

When. During the blended event and before or after caregiver-facing provider or community interactions when reinforcement is useful.

What and how much. Teach practical preparation: list concerns, bring permitted information, describe observations accurately, ask questions, request clarification, record next steps, identify the responsible person, and follow up. Preserve the caregiver’s role and beneficiary permission; do not coach misrepresentation, clinical conclusions, or unauthorized disclosure.

Data/provenance. Direct for training: approved interaction guidance; current permission/participation boundaries; assigned performer and current subject-specific competency or outside-provider approval evidence; caregiver practice; and correction by the competency-verified Puerto Rico GUIDE navigator performing the training-application task or the approved outside contracted training provider. Corroborating: upcoming appointments, prior communication barriers, and unresolved questions. Identity-only: appointment, provider name, portal access, navigator job title, or caregiver relationship does not establish disclosure authority, provider understanding, training competency, or follow-up completion.

Potential Proxi work. Generate approved visit-preparation prompts and question lists; conduct bounded roleplay; source-faithfully organize observations; draft reminders; and flag clinical or permission questions for the right person.

Human role. A Puerto Rico GUIDE navigator whose current verified competency covers provider-interaction training may correct roleplay inside the single training-application touchpoint. If that competency is absent, assign an approved outside contracted training provider; protected clinical content routes to the applicable U.S. clinician. Software enforces the approved privacy and disclosure rules; only an unresolved legal-authority question routes to Healthcare Legal Counsel. Actual provider/community interactions are external time.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 1 inside the application touchpoint; separate U.S. L3 0 / 0 / 5; separate U.S. L4 0 / 0 / 5.

Completion evidence. Preparation/follow-up skills taught; assigned performer; current subject-specific competency or outside-provider approval evidence; caregiver roleplay/teach-back; correction; permission boundary; unresolved questions; and accessible tools supplied.

What does not prove completion. Question list sent, appointment attended, portal access, provider response, note drafted, AI-suggested clinical conclusion, or actual loop closure.

Edge cases/open decisions/minimum tests. X-002, X-003, X-004, X-005, X-007, X-008, X-009, X-015, CC-E10, CC-E19, CC-E39, O-003, O-011, and O-036. Test no disclosure authority; a provider gives a new instruction; clinical question; language barrier; hostile interaction; beneficiary disagreement; and failed teach-back.

CG15. Teach recreation, social, and leisure support#

When. During the blended event when applicable and whenever beneficiary preferences, function, setting, support, or available activities change.

What and how much. Teach the caregiver to use beneficiary preferences, current abilities, routines, environment, support, and approved general safety principles to plan meaningful activities; start simply; observe response; and seek qualified help when safety or ability is uncertain. The beneficiary and caregiver choose activities.

Data/provenance. Direct for training: approved curriculum; beneficiary-authenticated preferences; current functional/safety findings; assigned performer and current subject-specific competency or outside-provider approval evidence; content delivered; caregiver planning practice; and correction. Corroborating: prior interests, activity response, community-program information, and caregiver observations. Identity-only: navigator job title, diagnosis, age, ZIP code, listing, or AI preference inference does not establish training competency, suitability, or choice.

Potential Proxi work. Match confirmed preferences to approved activity categories; generate examples; present accessible content; conduct bounded scenario practice; surface missing safety/function facts; and avoid declaring individualized suitability.

Human role. A Puerto Rico GUIDE navigator whose current verified competency covers recreation/social/leisure training may confirm person-centered planning principles. If that competency is absent, assign an approved outside contracted training provider; protected clinical content or an individualized safety/function question routes to the applicable U.S. clinician, OT/PT, or other authorized U.S. professional.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 1 inside the application touchpoint; separate U.S. L3 0 / 0 / 8; separate U.S. L4 0 / 0 / 8.

Completion evidence. Preferences/current facts used; assigned performer; current subject-specific competency or outside-provider approval evidence; planning skill taught; caregiver example/teach-back; correction; safety/help-seeking boundary; and unresolved suitability issue.

What does not prove completion. Activity recommendation, registration, generic dementia activity list, AI “best activity,” prior hobby, attendance, or clinician-free suitability claim.

Edge cases/open decisions/minimum tests. X-003, X-004, X-005, X-008, X-009, X-015, CC-E07, CC-E39, O-003, and O-036. Test changed mobility; wandering/fall concern; caregiver/beneficiary conflict; overstimulation; inaccessible activity; no transport; unsafe AI suggestion; and later preference change.

CG16. Teach future-planning skills#

When. During the blended event and whenever the caregiver requests reinforcement or an upcoming decision makes general planning education timely.

What and how much. Teach general planning skills: identify upcoming decisions, gather existing documents and information, list questions, identify trusted people, schedule the appropriate professional conversation, and use approved resources. Separate education and referral from legal, financial, capacity, clinical, end-of-life, or medical-order advice.

Data/provenance. Direct for training: approved non-advice curriculum and resource routes; assigned performer and current subject-specific competency or outside-provider approval evidence; caregiver participation/teach-back; and correction by the competency-verified Puerto Rico GUIDE navigator performing the training-application task or the approved outside contracted training provider. Existing legal or clinical documents establish only their recorded content and require authorized interpretation. Corroborating: care-plan goals, prior stated wishes, and upcoming events. Identity-only: document title, template, family relationship, navigator job title, or AI interpretation does not establish legal effect, authority, capacity, medical direction, or training competency.

Potential Proxi work. Deliver approved checklists/examples; organize known documents without interpreting them; run bounded question-preparation practice; provide approved jurisdiction-specific resource links; and route advice requests.

Human role. A Puerto Rico GUIDE navigator whose current verified competency covers future-planning training may provide boundary-aware nonclinical planning practice. If that competency is absent, assign an approved outside contracted training provider; protected clinical content routes to the applicable U.S. clinician, while an authorized U.S. social worker, attorney, financial professional, or other qualified professional provides individualized advice within scope.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 1 inside the application touchpoint; separate U.S. L3 0 / 0 / 5; separate U.S. L4 0 / 0 / 5. Legal and financial professional time is tracked outside these clinical columns.

Completion evidence. Approved planning skills taught; assigned performer; current subject-specific competency or outside-provider approval evidence; caregiver-prepared questions/next professional route; teach-back; correction; non-advice boundary; and unresolved professional question.

What does not prove completion. Document template, AI document interpretation, generic checklist, scheduled attorney/clinician, signed form, or navigator legal/medical conclusion.

Edge cases/open decisions/minimum tests. X-002, X-003, X-004, X-005, X-008, X-009, X-015, CP-E05, CP-E06, O-003, O-011, O-036, and O-PA-001. Test capacity question; multiple representatives; legal-advice request; medical-order question; financial question; missing documents; family conflict; and inaccessible checklist.

CG17. Teach caregiver self-care and stress management#

When. During the blended event and later when the caregiver requests reinforcement or reports changed stress, support, health, or coping. Stop general education for active suicidality, abuse, danger, severe distress, or treatment need.

What and how much. Teach approved general skills for recognizing caregiver stress, identifying support, setting realistic routines/boundaries, using nonclinical well-being practices, and seeking professional help. Do not diagnose depression/anxiety, provide psychotherapy, determine suicide risk, or prescribe treatment.

Data/provenance. Direct for training: approved curriculum; the caregiver’s stated stress/needs; assigned performer and current subject-specific competency or outside-provider approval evidence; content delivered; practice/teach-back; and correction by the competency-verified Puerto Rico GUIDE navigator performing the training-application task or the approved outside contracted training provider. A validated screen establishes only its recorded responses and score, not diagnosis. Corroborating: prior support use and assessment trends. Identity-only: burden score alone, caregiving duration, demographics, navigator job title, or AI sentiment does not establish a clinical condition or training competency.

Potential Proxi work. Deliver approved lessons/exercises and reminders; run bounded practice; present approved support routes; preserve caregiver words; detect configured safety cues; and summon the appropriate human pathway without assigning diagnosis or urgency.

Human role. A Puerto Rico GUIDE navigator whose current verified competency covers caregiver self-care and stress-management training may reinforce general help-seeking. If that competency is absent, assign an approved outside contracted training provider; protected clinical content and clinical symptoms, therapy, suicidality, abuse, or danger route to the applicable U.S. clinician, behavioral-health professional, emergency service, or other authorized U.S. professional.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 1 / 2 inside the application touchpoint; separate U.S. L3 0 / 0 / 12; separate U.S. L4 0 / 0 / 10.

Completion evidence. Approved skills taught; assigned performer; current subject-specific competency or outside-provider approval evidence; caregiver participation/teach-back; support route; correction; accessibility; unresolved need; and actual clinical/safety cues routed separately.

What does not prove completion. Wellness article sent, meditation completed, burden score, chatbot reassurance, quiz, “I’m fine,” attendance, or behavioral-health referral.

Edge cases/open decisions/minimum tests. X-003, X-004, X-005, X-008, X-009, X-014, X-015, CC-E13, CC-E39, O-003, O-004, O-036, and O-PA-001. Test routine stress; high distress; suicidal statement; abuse concern; decline; inaccessible exercise; AI overreassurance; and failed live handoff.

CG18. Check understanding and remaining training needs#

When. Throughout and at the end of the blended CG09-CG17 event, and later when the caregiver reports difficulty, an error, or a changed need.

What and how much. Ask the caregiver to explain or demonstrate applicable skills; compare responses with approved teaching; correct misunderstandings; identify topics needing repeat practice, another format, or qualified professional instruction; and record incomplete or declined status truthfully. Do not infer incapacity from difficulty.

Data/provenance. Direct for understanding: caregiver teach-back/demonstration; assigned performer and current subject-specific competency or outside-provider approval evidence for each topic; exact observation/correction by the competency-verified Puerto Rico GUIDE navigator performing the training-application task or the approved outside contracted training provider; accessibility conditions; and remaining-needs statement. Corroborating: quiz, module completion, response latency, prior error, and confidence. Identity-only: attendance, link view, quiz score alone, navigator job title, AI conversation, or certificate does not establish usable understanding or training competency.

Potential Proxi work. Run topic-specific quizzes and teach-back prompts; capture responses; compare with approved expected elements; adapt/repeat content; highlight uncertainty; and prepare focused remediation. AI may not independently certify safe technique or whole-event completion under O-036.

Human role. A Puerto Rico GUIDE navigator whose current verified competency covers each topic being checked may observe, correct, and disposition remaining nonclinical needs. If competency is absent for any topic, assign that topic to an approved outside contracted training provider; protected clinical content routes to the applicable U.S. clinician or other authorized U.S. professional within scope. One navigator may not be inferred competent for all topics from job category alone.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 2 / 3 / 4 inside the application touchpoint; separate U.S. L3 0 / 0 / 10; separate U.S. L4 0 / 0 / 10.

Completion evidence. Applicable-topic teach-back/demonstration; exact response; assigned performer for each topic; current subject-specific competency or outside-provider approval evidence; observation/correction by the competency-verified Puerto Rico GUIDE navigator performing the training-application task or the approved outside contracted training provider; accessibility; remaining needs; repeat/referral plan; topic-level complete/incomplete/declined status; and whole-event disposition under the future settled O-036 rule.

What does not prove completion. “Yes, I understand,” attendance, certificate, quiz score alone, video view, AI certification, no questions, or passing one topic while another remains unsafe.

Edge cases/open decisions/minimum tests. X-003, X-004, X-005, X-008, X-009, X-010, X-014, X-015, CC-E39, MED-E41, O-003, O-004, O-036, and O-039. Test all topics understood; one topic failed; unsafe demonstration; language/hearing barrier; unobserved group participant; declined teach-back; repeat success; and actual concern during demonstration.

Later tried-it/helped check. When the application touchpoint records the caregiver's intended real-world action, carry that action into the brief for the next naturally occurring caregiver interaction: the next qualifying contact with caregiver participation, the next CG26 call, or a permitted accessible asynchronous exchange. Ask whether the caregiver tried it, what happened, and whether it helped, partly helped, did not help, or was not tried, using the caregiver's own words. Do not create a separate outbound contact solely for this check. This later usefulness fact guides quality and follow-up; it does not replace the objective CG18 training-completion evidence. An unhelpful result receives a different approach, repeat practice, another format, or a qualified professional. A failed or unsafe attempt follows CG27 rather than receiving more generic content.

CG19. Ask whether diagnosis information is wanted#

When. At program entry and whenever the beneficiary or caregiver later requests more dementia-diagnosis information or changes a prior decline.

What and how much. Ask the beneficiary and caregiver separately whether more information is wanted; who wants it; language/format; questions/topics; and whether the request is accepted, declined, deferred, or unclear. Keep the request separate from permission to disclose patient-specific information and from the CG21 conversation.

Data/provenance. Direct: each authenticated person’s current request/decline and communication preferences; established representative/permission facts when patient-specific information may later be shared. Corroborating: prior questions and choices. Identity-only: caregiver relationship, portal access, diagnosis code, or prior material delivery does not establish a current request or disclosure basis.

Potential Proxi work. Ask through accessible text/voice; explain the option without providing the diagnosis conversation; preserve each choice/question; read it back; accept corrections; and route an accepted request to CG20/CG21.

Human role. The beneficiary or caregiver chooses; that is human participation but not Proxi labor. No Proxi human is needed on a clear path. L2 enters only for requested help, failed comprehension, or a sensitive relationship issue. Software enforces the approved privacy, disclosure, and authority rules; only an unresolved legal-authority question routes to Healthcare Legal Counsel.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 5; separate U.S. L3 0 / 0 / 0; separate U.S. L4 0 / 0 / 0.

Completion evidence. Requester; authenticated accept/decline/defer; date; requested topics; language/format; permission held separately; and accepted request routed onward.

What does not prove completion. Caregiver status, prior material, diagnosis code, one person’s request substituted for another’s, silence, or prepared CG20 material.

Edge cases/open decisions/minimum tests. X-001, X-002, X-003, X-004, X-005, X-009, X-015, CP-E04, CP-E05, CP-E06, O-011, and O-PA-001. Test beneficiary accepts/caregiver declines; caregiver requests with permission unclear; both decline; later change; multiple caregivers; wrong identity; inaccessible question; and urgent clinical content.

CG20. Prepare understandable diagnosis materials#

When. After CG19 records an accepted request and before the direct CG21 one-on-one conversation. Repeat only when the request, diagnosis context, language, accessibility need, or approved material changes.

What and how much. Select current approved written materials responsive to the requested scope—dementia information, common behavioral changes, functional status, and available resources—in the primary language and a usable format. Use only approved patient-specific content and permission. Do not create new clinical meaning, interpret the diagnosis, or represent preparation as CG21 delivery.

Data/provenance. Direct: accepted request; current approved library and content-to-asset mapping; language/accessibility need; permission; and clinician-approved patient-specific facts when included. Corroborating: prior materials and successful formats. Identity-only: diagnosis code, generic article, auto-translation, caregiver relationship, or file title does not establish fit, approval, permission, or comprehension.

Potential Proxi work. Select approved assets; assemble a request-specific packet; render approved language/accessibility formats; verify readability and links; preserve source/review date; and queue for CG21. AI may draft new translation or explanation only for qualified review.

Human role. No Proxi human is needed when a current approved asset and approved rendition match exactly. A separate U.S. clinical/content owner reviews new clinical meaning; a qualified translator/accessibility specialist reviews a new meaning-bearing rendition; and an authorized human releases a new V1 PHI packet when applicable.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 5 for manual asset/access recovery; L2 0 / 0 / 0; separate U.S. L3 0 / 0 / 10; separate U.S. L4 0 / 0 / 10. Translation/accessibility and release-review minutes are separately measured when triggered.

Completion evidence. Accepted-request link; exact approved assets; source/review date; language/accessibility rendition; permission; patient-specific approved content if any; quality check; and ready-for-CG21 status.

What does not prove completion. Generic article, auto-translation, AI summary, diagnosis-code match, packet without permission, sent or opened material, or prepared packet counted as CG21.

Edge cases/open decisions/minimum tests. X-001, X-002, X-003, X-004, X-005, X-007, X-009, X-015, CC-E39, D-007, O-003, O-011, O-PA-001, and the proposed content-approval decision. Test exact approved asset; stale asset; new translation; wrong diagnosis context; revoked permission; inaccessible PDF; broken link; clinical-interpretation request; and CG21 never completed.

CG21. Provide diagnosis information at program entry#

When. At program entry after CG19 records an accepted request and CG20 prepares an approved accessible packet. Repeat when a caregiver newly enters, a prior decline changes, or materially different approved diagnosis information is requested. Handle each caregiver’s disclosure permission separately.

What and how much. A GUIDE care-team member speaks directly one-on-one with the caregiver, provides the written information, explains approved general information about the beneficiary’s dementia diagnosis, common behavioral changes, functional status, and available resources, invites questions, checks understanding, and routes individualized diagnosis, prognosis, treatment, medication, capacity, or safety questions to the appropriate U.S. clinician. This is a human service; text, video, chatbot, or voice AI may prepare and reinforce it but may not substitute for the conversation.

Data/provenance. Direct for the service: authenticated caregiver identity; current permission/disclosure basis; accepted request; approved packet/content version; care-team member identity; date/modality; content actually discussed; caregiver questions; teach-back; correction; and referrals. Corroborating: material delivery/opening and prior education. Identity-only: caregiver relationship, portal access, course completion, or diagnosis code does not prove permission, direct conversation, or understanding.

Potential Proxi work. Prepare the session brief; deliver the approved packet; schedule or open a live session; render approved content; capture questions and teach-back; draft a source-faithful note; and route out-of-scope questions. Proxi may not generate new patient-specific clinical meaning or count an AI-only exchange as the service.

Human role. L2 may provide approved nonclinical diagnosis information as the direct care-team member and owns the relationship-sensitive explanation. A separate U.S. L3/L4 clinician answers individualized clinical questions within actual authority. L1 may solve only a failed connection or accessibility logistics issue.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 5; L2 8 / 12 / 20; separate U.S. L3 0 / 0 / 12; separate U.S. L4 0 / 0 / 12. The U.S. clinical route is triggered and separately measured; it does not convert the L2 conversation into clinical work.

Completion evidence. Direct one-on-one care-team conversation; caregiver identity; approved packet/content version; permission; date/modality; topics discussed; questions answered or routed; teach-back/correction; and incomplete, declined, or follow-up status.

What does not prove completion. Packet sent/opened, video view, AI text/voice session, group class, note template, voicemail, scheduled call, clinician availability without participation, or a generic education encounter not tied to the caregiver/request.

Edge cases/open decisions/minimum tests. X-001, X-002, X-003, X-004, X-005, X-007, X-008, X-015, CC-E39, O-003, O-011, O-036, and O-PA-001. Test clean one-on-one; multiple caregivers; permission withdrawn; group substituted; AI-only delivery; wrong language; failed teach-back; clinical prognosis question; actual safety disclosure; and interrupted conversation.

CG22. Offer support-group participation#

When. At initial caregiver-support planning, after a prior decline or deferral changes, and when the approved group catalog materially changes. Do not repeatedly pressure a caregiver who declined.

What and how much. Make one accessible offer that explains facilitated group purpose, general topic scope, permitted provider route, format, schedule pattern, privacy/group limits, cost if any, how to join or leave, and the caregiver’s right to accept, decline, defer, attend once, or withdraw later. The offer is distinct from registration and actual participation.

Data/provenance. Direct: approved current group catalog and provider route; actual format/cost/privacy description; caregiver communication needs; and authenticated choice. Corroborating: prior interest, schedule preferences, and similar group attendance. Identity-only: public listing, email delivery, group membership field, or calendar event does not establish a meaningful offer or choice.

Potential Proxi work. Present current approved options; filter on explicit schedule/language/accessibility facts; explain approved participation terms; answer bounded program questions; record choice; and reopen after a later change. It may not infer group suitability from diagnosis or stress score.

Human role. The caregiver chooses; that is not Proxi labor. No Proxi human is needed on a clear path. L2 enters only on request, sensitivity concern, comprehension failure, or a relationship issue. Software enforces the approved privacy and disclosure rules.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 5; separate U.S. L3 0 / 0 / 0; separate U.S. L4 0 / 0 / 0.

Completion evidence. Dated accessible offer; current options and terms; caregiver-authenticated accept/decline/defer; questions and routed status; and later-change/withdrawal mechanism.

What does not prove completion. Group catalog, sent link, assumed acceptance, registration, reminder, attendance, membership, or facilitator note.

Edge cases/open decisions/minimum tests. X-003, X-004, X-005, X-007, X-009, X-015, CC-E07, CC-E39, O-PA-001, and the proposed group-provider/catalog decision. Test accept; decline; defer; later accept; multiple caregivers; no accessible group; privacy concern; temporarily unavailable group; and stale catalog.

CG23. Arrange support-group access#

When. After CG22 records acceptance and the caregiver selects a permitted group. Repeat for a changed group, access failure, cancellation, or caregiver-requested alternative.

What and how much. Register or refer through a permitted route: GUIDE care team; reimbursed contracted vendor/community organization; or a community organization whose support group is free of charge. Confirm date/time/timezone, modality/location, language/accessibility, privacy expectations, technology/transport, reminders, contact, and fallback. Keep registration, attendance, participation, and group-service completion separate.

Data/provenance. Direct: approved group/provider route and reimbursement/free status; current availability; caregiver choice/permission; accessibility needs; registration/referral transaction; and returned confirmation. Corroborating: public listing, prior attendance, and schedule preferences. Identity-only: organization name, calendar invitation, or website does not prove permitted route, successful enrollment, or participation.

Potential Proxi work. Search the approved catalog; deterministically filter exact constraints; present choices; complete a preapproved electronic registration; send access instructions/reminders; test connection; track confirmation; retry; and offer a truthful no-match result.

Human role. No Proxi human is needed on the connected clean path. L1 handles manual registration, endpoint verification, or failed access. L2 handles sensitive fit, preference, conflict, or no-feasible-option discussions; it does not clinically select a group. Software enforces the approved privacy and disclosure rules.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 8; L2 0 / 0 / 5; separate U.S. L3 0 / 0 / 0; separate U.S. L4 0 / 0 / 0.

Completion evidence. Permitted route; caregiver selection; exact group; registration/referral confirmation or truthful failed outcome; date/time/timezone/modality; accessibility; reminders; and unresolved barrier.

What does not prove completion. Search result, directory listing, referral sent, offered time, unconfirmed registration, reminder delivery, waiting list, attendance, or group participation.

Edge cases/open decisions/minimum tests. X-003, X-004, X-005, X-006, X-007, X-013, X-014, X-015, CC-E07, CC-E18, CC-E26, CC-E27, O-PA-001, and the proposed group-provider/catalog decision. Test free group; reimbursed vendor; unapproved paid group; waitlist; manual registration; no confirmation; timezone error; interpreter unavailable; privacy concern; cancellation; and later alternative.

CG24. Facilitate a caregiver support group#

When. At each scheduled group session a caregiver actually joins through a permitted route. End or pause ordinary group work when a participant discloses an immediate safety, clinical, abuse, suicidality, or other out-of-scope concern.

What and how much. A human facilitator leads a bounded group discussion and skills/support activity on approved caregiver topics; establishes group expectations; encourages voluntary participation; avoids unauthorized disclosure; corrects harmful misinformation; prevents one participant’s clinical issue from becoming group treatment; and offers private follow-up when needed. Group facilitation cannot be replaced by a chatbot or an unmoderated forum.

Data/provenance. Direct for the furnished service: permitted facilitator/provider; session date/duration/modality; approved agenda/content; participant-level attendance; facilitation actually performed; caregiver participation or supported nonparticipation; accessibility; incidents; and follow-up. Corroborating: registration, reminder, platform join/leave, and satisfaction survey. Identity-only: group membership, link click, calendar event, or facilitator assignment does not establish attendance or facilitated service.

Potential Proxi work. Prepare approved agenda/materials; manage attendance and access; surface participant questions; provide facilitator-only source notes; draft a de-identified/session note and participant-level service record; and open private follow-up tasks. AI does not autonomously moderate clinical/safety content or determine urgency.

Human role. A dementia- and caregiving-trained human facilitator is mandatory. If Proxi uses L2, L2 facilitates nonclinically. A reimbursed contracted facilitator or permitted free community group may supply the direct service under the approved route. Separate U.S. clinical/emergency staff handle triggered clinical or safety concerns.

Provisional Proxi human minutes (low / typical / high). Per caregiver-session allocated share when Proxi L2 facilitates: L1 0 / 0 / 0; L2 5 / 8 / 15; separate U.S. L3 0 / 0 / 10; separate U.S. L4 0 / 0 / 10. The facilitator still works the full session—for example, a 60-minute session with eight participants consumes 60 facilitator minutes but allocates about 7.5 minutes per participating caregiver. Contracted-provider session time/cost is tracked separately, not relabeled as L2.

Completion evidence. Permitted route/provider; actual facilitated session; participant-level attendance; duration; content; accessibility; participation/support; incidents and private follow-up; and any clinical/safety handoff recorded separately.

What does not prove completion. Registration, link open, unmoderated forum, prerecorded webinar, AI moderator, facilitator assigned but absent, aggregate session note without participant evidence, or clinical handoff.

Edge cases/open decisions/minimum tests. X-001, X-002, X-003, X-007, X-008, X-014, X-015, CC-E39, O-003, O-004, O-PA-001, and the proposed group-service evidence decision. Test normal group; participant listens but does not speak; join/leave early; facilitator absent; prohibited disclosure; misinformation; abuse/suicide disclosure; failed private handoff; and contracted/free-community route.

CG25. Make and initiate ad hoc one-on-one support calls#

When. Whenever the caregiver requests ad hoc support, including through the caregiver pulse; leaves another asynchronous request; or a GUIDE care-team member identifies a reason to initiate a one-on-one caregiver-support call. This task opens and connects the call; CG26 is the substantive support conversation.

What and how much. Maintain an accessible request route; capture the caregiver’s own description and callback facts; acknowledge the request without giving unsupported advice; arrange or initiate a timely direct care-team connection; and preserve whether the call was caregiver-requested or care-team-initiated. An actual clinical/safety signal routes immediately and does not wait for routine scheduling.

Data/provenance. Direct: authenticated requester/caregiver; request time/channel; verbatim concern; callback/location facts when relevant; caregiver availability; care-team initiator and reason; attempts; and live connection. Corroborating: prior support needs and contact preferences. Identity-only: missed call, sentiment score, unread message, or monitoring flag does not establish the concern, identity, urgency, or completed support call.

Potential Proxi work. Provide 24/7 intake; capture structured callback information; preserve verbatim content; detect configured stop phrases; acknowledge receipt; schedule; route; retry; prepare context; and initiate a live connection. AI cannot decide urgency or provide the required human support conversation.

Human role. L1 may answer, capture, and warm-transfer without counseling or triage. L2 may initiate the relationship-sensitive outreach and is the default nonclinical care-team support role. Separate U.S. clinical staff enter only for a clinical/safety trigger. Caregiver-request intake alone is not the furnished CG26 service.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 5; L2 0 / 2 / 5 for proactive or sensitive connection work; separate U.S. L3 0 / 0 / 0; separate U.S. L4 0 / 0 / 0. CG26 conversation and CG27 escalation minutes are separate and counted once.

Completion evidence. Request/initiation source; caregiver identity; verbatim reason; time/channel; callback/contact plan; attempts; direct CG26 connection or truthful pending/failed/withdrawn disposition; and immediate concern routed.

What does not prove completion. Request button, automated acknowledgement, voicemail, scheduled callback, unanswered outreach, warm-transfer attempt, monitoring alert, or later clinical handoff.

Edge cases/open decisions/minimum tests. X-001, X-002, X-003, X-006, X-007, X-008, X-013, X-014, X-015, CC-E13, CC-E39, O-003, O-004, and the proposed contact/service-credit decision. Test caregiver request; care-team initiation; wrong identity; no callback number; repeated no answer; immediate safety phrase; routine distress; failed transfer; withdrawal; and late response.

CG26. Conduct the one-on-one caregiver support call#

When. After CG25 creates a live connection, or during another qualifying direct care-team contact when the caregiver actually raises and receives one-on-one support for a current caregiving issue.

What and how much. A human GUIDE care-team member listens, validates the caregiver experience without making unsupported clinical claims, clarifies the immediate nonclinical issue, provides approved coaching and practical support, agrees next steps with the caregiver, and routes every clinical, safety, unresolved legal-authority, or external-service need. Software enforces the approved privacy and disclosure rules. The service is ad hoc and responsive; no fixed frequency is asserted from the current source packet.

Data/provenance. Direct for the service: authenticated caregiver; live care-team member; date/duration/modality; caregiver’s own concern; discussion; coaching/support actually provided; caregiver response; agreed next steps; referrals/handoffs; and follow-up owner. Corroborating: transcript, prior assessment, and related tasks. Identity-only: call connection, duration, note template, AI summary, or generic resource delivery does not establish substantive one-on-one support.

Potential Proxi work. Prepare a source-faithful brief; offer approved resources; support bounded note capture; summarize only for human correction; create agreed follow-up tasks; and track completion. AI may not replace the human conversation, choose advice, determine urgency, or close the support issue.

Human role. L2 conducts routine nonclinical support and coaching. Separate U.S. L3/L4 or behavioral-health professionals handle clinical interpretation, treatment, suicidality, abuse, or safety. L2 may warm-transfer and remain supportive but does not perform the clinical act.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 5 / 12 / 25; separate U.S. L3 0 / 0 / 15; separate U.S. L4 0 / 0 / 15. High clinical time is a triggered linked service, not a reason to label the whole call clinical. When the conversation enters through the 24/7 line, first-answer intake and connection minutes stay in that episode; the substantive conversation is counted once here, and 24/7 task 20 uses this same band.

Completion evidence. Direct one-on-one human care-team conversation; caregiver identity; duration/modality; concern; substantive support/coaching; caregiver response; next actions/owners; routed concerns; and follow-up disposition.

What does not prove completion. Intake, scheduled call, voicemail, AI/chatbot exchange, generic check-in, resource link, call duration alone, transcript without human support, or safety handoff without returning to the support need.

Edge cases/open decisions/minimum tests. X-001, X-002, X-003, X-004, X-007, X-008, X-014, X-015, CC-E10, CC-E13, CC-E19, CC-E39, O-003, O-004, and the proposed contact/service-credit decision. Test short resolved issue; complex nonclinical issue; caregiver declines coaching; clinical question; suicidal/abuse disclosure; interpreter; beneficiary/caregiver conflict; disconnected call; and follow-up never completes.

CG27. Escalate clinical or immediate safety concerns from caregiver support#

When. Immediately when CG09-CG26 or any caregiver-support channel reveals a possible symptom, medication issue, suicidality, abuse/neglect, danger, severe distress, or other clinical/safety concern. A configured signal may trigger the safe route, but AI/L1/L2 does not determine clinical urgency.

What and how much. Stop the affected routine activity; preserve the caregiver’s exact words and current location/callback facts when safely obtainable; invoke the approved U.S. clinical, behavioral-health, emergency, poison-control, protective-services, or other route; warm-transfer when possible; retry a failed handoff; and preserve the original caregiver-support need for later follow-up. The receiving authorized professional determines urgency and response.

Data/provenance. Direct: exact statement/observation and speaker; event time; current location/callback; known immediate conditions; source context; attempted routes; receiving person/service; connection; and authorized disposition. Corroborating: AI flag, prior history, sentiment, and third-party report. Identity-only: diagnosis, alert label, or prior crisis does not establish current urgency or a completed handoff.

Potential Proxi work. Detect configured concern language; freeze routine content; assemble source-faithful facts; call the approved route; retry; notify the appropriate human; and track connection and returned disposition. It cannot grade risk, reassure, diagnose causality, select treatment, or close from successful dialing.

Human role. L1 or L2 may capture verbatim facts and connect. L2’s work remains nonclinical. A separate U.S. beneficiary-location-authorized L3/L4 or behavioral clinician performs the clinical act; emergency/protective services act within their authority. No Puerto Rico navigator minute is clinical.

Provisional Proxi human minutes per triggered event (low / typical / high). L1 0 / 2 / 4 when first answer is involved; L2 2 / 4 / 8 for nonclinical capture/warm handoff; separate U.S. L3 5 / 12 / 25; separate U.S. L4 0 / 5 / 15. Routes are selected by actual authority and need; do not automatically add every column. External emergency-service time is excluded.

Completion evidence. Source-faithful concern; time/location; stopped routine work; route selected under approved policy; attempts; verified live handoff or truthful failed-handoff escalation; authorized disposition; caregiver communication; and linked follow-up/successor work. CG27 completion does not close the original support issue.

What does not prove completion. AI risk score, keyword match, message sent, attempted call, voicemail, successful transfer without receiving confirmation, emergency referral, clinical note without caregiver follow-up, or L2 judgment that risk is low.

Edge cases/open decisions/minimum tests. X-001, X-003, X-004, X-007, X-008, X-009, X-013, X-014, X-015, CC-E10, CC-E13, CC-E39, MED-E18, MED-E44, D-015, O-003, and O-004. Test hypothetical versus actual concern; false-positive AI flag; unknown or changing location; disconnected handoff; U.S. clinician lacks location authority; current abuse; suicide statement; medication event; emergency-service refusal; and post-handoff support follow-up.

CG28. Offer optional peer support#

When. Only if the Participant offers an approved peer-support program, after the caregiver-support plan identifies interest, and whenever a prior decline changes. Do not present peer mentoring as required GUIDE treatment or professional advice.

What and how much. Explain the optional service; the peer’s lived-experience role and boundaries; format, supervision, privacy limits, and stopping process; collect preferences and consent; match only within approved nonclinical criteria; arrange an introduction; and preserve acceptance, decline, withdrawal, or no-match. The peer is an experienced former caregiver whose own caregiving experience has ended.

Data/provenance. Direct: approved program/peer roster and oversight; peer eligibility and current availability; caregiver choice/consent/preferences; matched criteria; introduction; and participation status. Corroborating: interests and schedule. Identity-only: former-caregiver self-description, public profile, demographic similarity, or AI similarity score does not establish approved peer status or suitability.

Potential Proxi work. Explain approved terms; deterministically filter explicit preferences; prevent prohibited matching; offer choices; obtain consent; schedule/introduction; remind; and track status. AI may not infer personal compatibility or allow peer medical/legal advice.

Human role. No Proxi human is needed to make a clean digital offer and match from an approved roster. L1 handles connection failures; L2 handles sensitive preferences, consent concerns, mismatch, boundary concerns, or no safe match. The peer’s direct mentoring time is human service time tracked separately.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 5; L2 0 / 0 / 5; separate U.S. L3 0 / 0 / 0; separate U.S. L4 0 / 0 / 0.

Completion evidence. Optional-program applicability; approved peer; caregiver consent/choice; match basis; introduction; boundaries; participation/withdrawal/no-match; and issue escalation.

What does not prove completion. Public peer profile, AI match score, suggested introduction, calendar invite, unanswered outreach, peer conversation with no approved program oversight, or peer advice outside role.

Edge cases/open decisions/minimum tests. X-001, X-002, X-003, X-005, X-006, X-007, X-008, X-015, O-PA-001, and the proposed peer-program governance decision. Test clean match; no match; peer inactive; caregiver withdraws; privacy concern; peer gives clinical advice; boundary violation; multiple caregivers; and failed introduction.

CG29. Offer optional self-care and well-being resources#

When. Only if offered by the Participant, when the caregiver requests or the support plan identifies interest in general well-being resources, and after a material preference, access, or need change. Active clinical symptoms or safety concerns use CG27/CG32 rather than a generic wellness offer.

What and how much. Offer approved general resources or referrals for mindfulness, relaxation, arts, music, exercise, nonclinical well-being instruction, or other approved options. Clearly distinguish general programming from psychotherapy, individualized treatment, or clinical cognitive behavioral therapy. Record choice, access support, and whether a separate professional referral is needed.

Data/provenance. Direct: approved current catalog; provider qualification appropriate to the offered activity; availability/cost/accessibility; caregiver preference; and registration/referral outcome. Clinical screening or diagnosis is direct only as the qualified author’s finding. Corroborating: prior participation and stated interests. Identity-only: stress score, AI sentiment, course listing, or “CBT” label does not establish clinical need, treatment, or suitability.

Potential Proxi work. Present approved options; filter exact constraints; explain approved boundaries; register or refer on the clean path; remind; and track barriers. AI cannot diagnose a mental-health condition, prescribe an activity, or present a wellness course as therapy.

Human role. No Proxi human is needed for a caregiver-chosen approved general option on the connected path. L1 handles manual logistics; L2 supports sensitive choice/barriers; a separate U.S. L3/L4 or licensed behavioral-health professional handles clinical symptoms, therapy, or treatment.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 5; L2 0 / 0 / 5; separate U.S. L3 0 / 0 / 10; separate U.S. L4 0 / 0 / 10.

Completion evidence. Optional applicability; current approved option; caregiver-authenticated choice; boundary between general programming and treatment; referral/registration or truthful no-match; access support; and clinical route when triggered.

What does not prove completion. Wellness link, meditation use, app download, AI recommendation, stress score, course registration, or behavioral-health referral counted as completed counseling.

Edge cases/open decisions/minimum tests. X-003, X-004, X-005, X-007, X-008, X-009, X-015, CC-E07, CC-E26, CC-E27, O-003, O-004, and O-PA-001. Test caregiver-chosen class; unavailable option; inaccessible activity; medical exercise concern; “CBT” ambiguity; suicidal disclosure; therapy request; coverage barrier; and stale provider information.

CG30. Explain the caregiver role#

When. Only if offered by the Participant, when a caregiver is newly identified or asks what caregiving commonly involves, and when the beneficiary’s setting, assistance needs, or caregiver participation changes.

What and how much. Explain approved general information about common caregiver activities, voluntary participation, boundaries, self-care, when to seek help, available GUIDE resources, and how the caregiver may accept, decline, or change tasks. Do not assign duties, establish representative authority, infer capacity, or teach a clinical technique within this task.

Data/provenance. Direct: current approved role curriculum; beneficiary/caregiver-authenticated participation facts and choices; applicable current plan/support facts; and content/teach-back. Corroborating: assessment and current assistance patterns. Identity-only: family relationship, co-residence, emergency-contact field, or AI role inference does not establish agreed duties or authority.

Potential Proxi work. Deliver approved accessible content; tailor only from confirmed participation facts; answer bounded questions; capture choices/questions; repeat; and route legal, clinical, safety, or relationship conflict.

Human role. No Proxi human is needed on a clear approved-content path. L2 enters on request, failed comprehension, sensitive expectation conflict, coercion concern, or disagreement. Separate U.S. clinical staff handle clinical questions.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 8; separate U.S. L3 0 / 0 / 5; separate U.S. L4 0 / 0 / 5.

Completion evidence. Optional applicability; approved current content; caregiver identity; participation choices/boundaries; questions; teach-back/correction; and routed conflicts.

What does not prove completion. Family relationship, assigned task list, brochure delivery, AI inference, caregiver silence, portal proxy, or explanation treated as consent or legal authority.

Edge cases/open decisions/minimum tests. X-001, X-002, X-003, X-004, X-005, X-008, X-009, X-015, CP-E04, CP-E05, CP-E06, O-011, and O-PA-001. Test new caregiver; caregiver declines task; beneficiary disagreement; multiple caregivers; coercion; capacity question; clinical-technique request; wrong language; and changed setting.

CG31. Coach access to community resources#

When. Only if offered by the Participant, when a caregiver requests help obtaining transportation, nutrition, homemaker, yard, or another approved caregiving support, and whenever availability, eligibility, access constraints, or caregiver choice changes.

What and how much. Clarify the caregiver’s requested practical need and explicit constraints; present current verified options and a truthful no-match; explain application/intake steps; support the caregiver’s choice; register, refer, or make a requested warm introduction; and track the route through the defined connection outcome. Do not infer clinical suitability, guarantee eligibility, or claim service start from a directory match.

Data/provenance. Direct: caregiver request/choice; current verified resource inventory; eligibility/availability source and date; constraints; permission; referral/intake transactions; returned outcomes; and barriers. Corroborating: prior use and assessment findings. Identity-only: ZIP code, benefit status, directory listing, or AI fit score does not establish actual eligibility, availability, selection, or connection.

Potential Proxi work. Deterministically filter/rank on confirmed facts; explain match reasons and unknowns; prepare applications/referrals; schedule; remind; pursue status; and track connection. Software may complete a clean electronic route. AI may not choose for the caregiver or declare a resource clinically appropriate.

Human role. L1 handles manual calls, forms, and status pursuit. L2 enters for preference-sensitive coaching, relationship barriers, requested warm introduction, conflict, or no-feasible-option planning. If Proxi sells this optional service specifically as human coaching, the future service definition must require an actual L2 interaction; a software-only match cannot be relabeled as human coaching.

Provisional Proxi human minutes (low / typical / high). L1 0 / 3 / 8; L2 0 / 5 / 12; separate U.S. L3 0 / 0 / 5; separate U.S. L4 0 / 0 / 5. Clinical time applies only to a separate suitability/safety question.

Completion evidence. Optional applicability; requested need/constraints; current source; explained options/no-match; caregiver choice; exact referral/intake actions; returned connection or truthful non-success status; barriers; and next owner.

What does not prove completion. Directory list, AI ranking, referral sent, application prepared, acceptance without intake, appointment without service, service claim without caregiver confirmation, or a documented barrier with no disposition.

Edge cases/open decisions/minimum tests. X-003, X-004, X-005, X-006, X-007, X-013, X-014, X-015, CC-E01, CC-E07, CC-E18, CC-E26, CC-E27, O-022, and the proposed optional human-coaching definition. Test connected clean route; phone-only resource; stale availability; false eligibility match; caregiver rejects top result; waitlist; no match; inaccessible service; failed warm introduction; and late service start.

CG32. Refer for psychological counseling#

When. When a caregiver asks for counseling, an authenticated caregiver assessment or human conversation identifies a possible counseling need, or a qualified professional recommends referral. An active suicidal, abuse, danger, or severe clinical concern invokes CG27 immediately rather than waiting for routine referral.

What and how much. Explain the counseling-referral option within role; establish the caregiver’s choice and practical constraints; identify a current appropriately qualified provider; address permission, coverage, language, location, modality, and access; make the referral or help the caregiver self-refer; track receipt/intake; and retain a truthful non-success outcome when connection fails. Referral does not itself provide counseling.

Data/provenance. Direct: caregiver request/consent; assessment responses and qualified findings; current location; verified provider credential/scope and availability; coverage/access facts; referral; receipt/intake; and returned disposition. Corroborating: prior support use, AI cue, and general distress patterns. Identity-only: stress score, sentiment label, diagnosis suggestion, provider directory, or claim does not establish clinical diagnosis, urgency, fit, or counseling start.

Potential Proxi work. Capture the request; preserve exact words; retrieve/filter current providers on explicit facts; present choices; prepare the referral; schedule; remind; pursue status; and route configured danger cues. AI cannot diagnose depression/anxiety, determine suicide risk, choose therapy, or declare a provider clinically suitable from a listing alone.

Human role. L2 provides sensitive nonclinical navigation and caregiver choice support. L1 handles manual referral/scheduling logistics. A separate U.S. L3 LCSW/behavioral clinician or other authorized U.S. professional evaluates clinical symptoms/urgency when required; L4 enters only when medical assessment/treatment authority is needed. The external licensed behavioral-health professional furnishes counseling.

Provisional Proxi human minutes (low / typical / high). L1 0 / 3 / 8; L2 3 / 7 / 15; separate U.S. L3 0 / 10 / 20; separate U.S. L4 0 / 0 / 10. A caregiver-led routine referral may use no clinical minute; a symptom- or risk-driven route does.

Completion evidence. Trigger/request source; caregiver choice; clinical/safety disposition when applicable; verified provider and constraints; referral/self-referral event; receipt/intake or authorized truthful non-success outcome; barriers; follow-up owner; and counseling start kept separate.

What does not prove completion. AI sentiment, burden score, provider list, referral sent, voicemail, appointment scheduled, claim, counseling note, or generic wellness resource. A successful emergency handoff does not close the routine counseling need unless an authorized disposition says so.

Edge cases/open decisions/minimum tests. X-001, X-002, X-003, X-004, X-005, X-006, X-007, X-008, X-009, X-014, X-015, CC-E05, CC-E06, CC-E07, CC-E13, CC-E25, CC-E26, CC-E41, O-003, O-004, and O-PA-001. Test caregiver self-request; positive screen without diagnosis; suicide statement; provider unavailable; wrong license; out-of-state telehealth issue; coverage denial; Part 2/sensitive data; refusal; and referral accepted but intake never occurs.

CG33. Reassess caregiver needs and update support#

When. At each required caregiver-assessment/reassessment cycle established by current CMS/Participant policy, after a material change in caregiver stress, ability, willingness, knowledge, support, beneficiary function/setting, or service preference, and when a caregiver joins, leaves, or corrects prior information. The current packet does not invent an additional fixed interim cadence.

What and how much. Reconfirm the caregiver and participation status; obtain or retrieve the current caregiver assessment; compare each domain with the prior assessment; ask the caregiver what changed; identify new, resolved, declined, or unknown needs; update the support plan; re-offer applicable required services without erasing prior choices; create accepted follow-up; and route clinical/safety issues. Assess each caregiver separately.

Data/provenance. Direct: authenticated caregiver responses/corrections; current assessment items; caregiver choices; actual service use/outcomes; and qualified clinical findings. Corroborating: prior assessment, contact history, and observed changes. Identity-only: elapsed time, caregiver contact field, burden-score delta, missed service, or AI summary does not establish current needs, willingness, or clinical condition.

Potential Proxi work. Schedule/trigger the reassessment; retrieve the right prior record; deliver accessible structured questions; compare item-level changes; preserve source statements; identify missing/conflicting data; prepare a proposed support-plan delta; re-present required offers; and track accepted work. AI cannot choose priorities, infer clinical meaning, or certify the update.

Human role. Software prepares the source-linked reassessment delta and proposed support update before human time begins. The caregiver may directly confirm choices on an approved clear path. L2 resolves only a required/requested human review, ambiguity, distress, comprehension failure, disagreement, accessibility failure, or relationship-sensitive tailoring; L2 does not reconstruct the delta or bless a clean update. L1 may retrieve missing administrative records. Separate U.S. L3/L4 or behavioral staff receive the prepared exact question and address triggered clinical, safeguarding, or treatment issues.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 5; L2 0 / 6 / 12; separate U.S. L3 0 / 0 / 12; separate U.S. L4 0 / 0 / 10. The low route is a clear direct confirmation. Each additional caregiver is a distinct assessment/review workload; do not merge household records.

Completion evidence. Caregiver identity; reassessment trigger/date; current assessment and provenance; item-level changes/unknowns; caregiver-confirmed priorities; updated support plan; all applicable required offers; accepted/declined/deferred services; follow-up owners; and routed clinical/safety concerns.

What does not prove completion. Calendar reminder, unchanged copied assessment, score alone, AI summary, generic check-in, service utilization, “no change” without caregiver confirmation, one caregiver’s update applied to another, or new referral without plan reconciliation.

Edge cases/open decisions/minimum tests. X-001, X-002, X-003, X-004, X-005, X-008, X-009, X-010, X-014, X-015, CP-E04, CP-E05, CP-E07, CC-E19, CC-E39, O-003, O-004, O-011, O-013, and O-PA-001. Test unchanged caregiver-confirmed review; material stress change; caregiver leaves; new caregiver; multiple caregivers; stale assessment; declined answer; urgent cue; failed follow-up; and late correction.

Cross-pillar caregiver-service feeds and learning return#

This SOP participates in cross-pillar episodes under 25_GUIDE_Eight_Pillar_Service_Integration.md. Caregiver Education and Support receives attribution only for actual approved education, application, diagnosis-information, facilitated group, one-to-one support, or other permitted caregiver-service content with the caregiver's response or action and a truthful incomplete or non-success outcome when the intended result is not reached.

A reusable caregiver action, barrier, boundary, or contingency goes to Care Plan for beneficiary-led action review. The exact intended or attempted action and next natural observation go to Ongoing Monitoring; do not create a separate quality-only call. A same-call micro-lesson, the full CG09-CG18 training route, and CG26 one-to-one support remain distinct. A micro-lesson cannot complete the full training, and AI alone cannot complete it under D-019/O-036. Credit only the content actually furnished; count shared interaction minutes once; keep later professional, referral, group, counseling, or clinical work separate.

Contact, labor, and GUIDE-service-credit boundary#

A single live interaction may operationally contain both ongoing monitoring and caregiver support, but the human minutes are counted once. The records must still distinguish the facts, content, and outcome supporting each obligation. A generic monitoring check-in does not prove CG26 one-on-one caregiver support; a CG26 support call does not prove every ongoing-monitoring requirement. Whether and under what evidence one interaction may receive both GUIDE service attributions remains a proposed compliance/billing decision, not an assumption in this SOP.

Pillar completion is not batch completion#

Completing or reviewing these 34 cards means the semantic SOP is specified; it does not mean the caregiver-support pillar was furnished for a beneficiary. Pillar completion requires, for every applicable caregiver: current applicability and assessment evidence; a responsive support plan; meaningful offers of all four required service types; permitted-provider, participation, and completion evidence for each accepted service; truthful decline/defer/unavailable outcomes; clinical/safety follow-through; and no open evidence defect. O-036 still blocks a final machine rule for required skills-training completion, and O-PA-001 still blocks Participant-specific completion policy. A completed training batch is not diagnosis-information completion, support-group completion, one-on-one support completion, or whole-pillar completion.

Proposed decisions exposed by the detailed cards#

1. Confirm D-019 under O-036: approve the software-led/permitted-human-owned route, software-enforced topic-by-topic competency gate before any Puerto Rico GUIDE navigator assignment, ten-topic completion/decline/deferral evidence, group/virtual participant evidence, application correction, incomplete outcomes, and later reinforcement for CG09-CG18 against the executed Participation Agreement.
2. Define curriculum/content ownership, review cadence, translation/accessibility approval, AI knowledge-base limits, the subject-specific navigator competency catalog and verification/expiry evidence, and the physical/behavioral stop conditions for CG09-CG20.
3. Define participant-level evidence, allocated labor, the governance-authored privacy/disclosure and incident rules to be enforced in software, and permitted provider routes for support groups, including paid contractor and free-community routes.
4. Define the exact evidence and billing rule for an interaction that contains both ongoing monitoring and CG26 caregiver support, while counting the human labor once.
5. Define the marketed completion promise for optional “human coaching” in CG31; software-led resource access must not be represented as a human coaching service unless the sold service definition permits it.
6. Define caregiver-support pillar completion and truthful non-success outcomes without allowing a completed batch, offer, handoff, or referral to stand in for every accepted service.
7. Define the tried-it/helped check's evidence and cadence and whether the result is quality-only or ever billing-relevant. Keep offered, delivered, understood, intended-for-use, tried, and helped as separate facts.

What this pillar provides as a service#

Proxi can accelerate assessment review, education preparation, translation, scheduling, reminders, resource matching, intake, documentation, and follow-up. The required human service remains caregiver choice, care-team-delivered diagnosis information and support calls, skills instruction by qualified people, facilitated group support, peer or professional services, and human clinical or safety response.