Comprehensive assessment

The working source is the CMS GUIDE Request for Applications, Appendix B, Section 1 ↗. The Proxi column describes candidate automation, not functionality already deployed. Before these tasks are used in the field, they must be compared with the organization's executed GUIDE Participation Agreement and any current CMS operating instructions.

Detailed assessment procedure — complete inventory#

This procedure expands the full 49-card inventory: tasks 1, 1A, 2-47, and the separately preserved task 33A. The original compact inventory remains below as a cross-check, but the detailed cards control the operating meaning.

Governing boundary and current public sources#

The public requirement basis is the CMS GUIDE Request for Applications ↗, Appendix B §§1.1.1–1.1.4.1.6, together with the current CMS GUIDE Payment Methodology Paper ↗, especially §§2.5, 2.10, and 2.11. The current methodology distinguishes the initial comprehensive assessment, annual comprehensive assessments, and narrower reassessment submissions. It also supplies current submission timing that the original RFA did not fully operationalize.

The main public boundaries for this batch are:

• A previous ordinary assessment completed before the GUIDE-specific voluntary-alignment process does not substitute for the new GUIDE initial comprehensive assessment.
• The initial assessment occurs after the required GUIDE consent discussion and before initial PAAF submission and CMS alignment confirmation. A prior assessment cannot simply be relabeled.
• Annual comprehensive assessments must be conducted and submitted between days 306 and 425 after the prior comprehensive assessment; once started, submission is due at the earlier of 60 days after start or day 425. The day-count convention and timezone remain open under O-018.
• Current reassessment submissions are not governed by one universal 180-day rule: residence-type changes may be submitted every 30 days, while changes in dementia severity or caregiver status/burden may be submitted every 180 days. A reassessment does not satisfy the annual comprehensive-assessment obligation.
• CMS requires the clinical domains in tasks 9–17, but does not provide task-level labor minutes. Appropriate interdisciplinary team members act according to license and scope; medication reconciliation/review requires a clinician with prescribing authority.

The executed Participation Agreement remains absent under O-PA-001. No participant-specific operating choice is treated as settled here.

Evidence and completion language#

• Direct evidence establishes only the fact its source is authorized to establish. A beneficiary report directly establishes what the beneficiary reports; a dispense record establishes a dispense event, not current use; a clinician’s examination establishes that clinician’s findings within scope.
• Corroborating evidence supports, challenges, or dates a candidate fact but cannot establish the target conclusion alone.
• Identity-only evidence helps match the correct beneficiary, caregiver, clinician, product, organization, or record. It does not establish authority, relationship, clinical truth, current medication use, completion, or safety.
• AI may ask approved questions, transcribe, extract, compare, score a configured instrument, and draft. It may not choose the assessment occasion, assign professional authority, perform a clinical examination, determine medical decision-making, interpret dementia stage, or reconcile medications.

Each card’s completion evidence proves only that task. It does not by itself prove the comprehensive assessment, PAAF submission, CMS alignment, tier assignment, care plan, or billable GUIDE service is complete.

Relevant registered sources are SRC-CMS-GUIDE-RFA-V1, SRC-CMS-GUIDE-PMP-3.0, SRC-GUIDE-PA-PY2026, SRC-HHS-TELEHEALTH-LICENSURE, SRC-ALZ-DCN-OBJECTIVES, SRC-AHRQ-MATCH-2022, SRC-HL7-FHIR-MED-R5, and SRC-NLM-RXNORM in 09_Source_Register.md.

Provisional first-year workload envelope#

All minute values are Proxi workload-calibration assumptions, not CMS facts, mandated encounter lengths, staffing guarantees, or field observations. Automated processing time and beneficiary, caregiver, representative, interpreter, external clinician, pharmacy, payer, and Participant-customer time are excluded.

The current workforce anchor treats the annual assessment and synthesis as approximately 35 L3 plus 80 L4 minutes, with first-year additions of 30 L3 plus 30 L4 minutes. Those are not two assessments; they are one initial assessment with incremental first-year work. Tasks 1-17 allocate 18 L3 and 52 L4 minutes. Tasks 18-47, including 33A, allocate the remaining 47 L3 and 58 L4 typical minutes. The complete initial-assessment clinical allocation therefore remains exactly 65 L3 and 110 L4 minutes. The L2 allocation is nonclinical navigator work only. Every L3/L4 minute belongs to the separate U.S. beneficiary-location-authorized clinical workforce under D-015; none is Puerto Rico navigator labor.

The following table shows the typical allocation for the remaining cards. Low and high scenarios appear on each detailed card. Conditional work such as an urgent safety response, POLST completion, no-caregiver safeguards, or additional diagnostic referral is zero in the typical case and appears in the applicable high scenario rather than being charged to every assessment.

Use one shared 13-minute L1 Philippines administrative exception event only when manual identity, record retrieval, scheduling, pharmacy/prescriber record pursuit, or source cleanup is actually triggered. Do not add 13 minutes to each failed task. It does not cover clinical interpretation, reconciliation, beneficiary relationship work, legal/privacy questions, interpreter work, or external-party time.

Assessment lane and spend discipline#

An outside “official dementia diagnosis” is not a universal prerequisite to starting the GUIDE initial assessment. Current CMS guidance expressly allows a person with suspected dementia to receive that assessment, and a dementia-proficient practitioner on the GUIDE Practitioner Roster may attest based on the completed comprehensive assessment or on a qualifying written report from another Medicare-enrolled practitioner. Claims codes alone do not supply the attestation, and mild cognitive impairment alone does not satisfy the dementia criterion.

The economic control is therefore dependency-based, not “wait for every clinical item”:

These figures are workload assumptions, not CMS contact-length requirements. A confirmed eligibility failure stops the GUIDE investment; a merely suspected diagnosis does not. CMS, not Proxi software, makes the final alignment determination from the submitted PAAF. Assessment-specific cases AS-E01 through AS-E08 define the first blocking and continuation examples.

Cross-pillar preparation and routing follow 25_GUIDE_Eight_Pillar_Service_Integration.md. Clean factual and authenticated-choice lanes proceed under the approved route without a routine L2 approval; L2 enters only under D-021; and a protected assessment question routes directly to the beneficiary-location-authorized U.S. professional. This does not change any task-local authority or minute assumption below.

Where Tasks 4, 5, 12, or 13 show L2 time to "confirm" a practical assessment fact, that ordinary assumption is conditional on the approved Assessment route requiring an actual care-team interview or verification. Authenticated self-service may prepare the fact and eliminate repeated collection; no L2 confirmation is added solely to create a human touch. Whether the clean path itself may complete each required Assessment act remains source- and owner-dependent, and the existing minutes remain provisional under O-039.

1. Identify the assessment occasion#

When. At the start of any proposed comprehensive assessment or reassessment work and before scheduling, assigning clinical staff, or representing the work as GUIDE-required. Re-evaluate when a new assessment date, alignment/consent fact, residence change, dementia-stage change, caregiver-status/burden change, or annual deadline arrives.

What and how much. Produce one truthful occasion: initial GUIDE comprehensive assessment; annual comprehensive assessment; narrower reassessment for a current CMS-recognized change; clinical assessment needed without a currently accepted reassessment submission; or unresolved. For an initial case, also produce one explicit spend disposition linked to PO-002 through PO-011: confirmed eligibility exclusion; administrative eligibility fact unresolved; suspected dementia requiring clinician confirmation; or eligibility sufficiently plausible to proceed with the full initial assessment. For an annual assessment, calculate the day-306 to day-425 window and the earlier-of-60-days-after-start/day-425 submission endpoint. For reassessment, keep residence change on the 30-day cadence separate from dementia-severity and caregiver-status/burden changes on the 180-day cadence.

Data. Direct: documented GUIDE consent/alignment-process status; authoritative current and prior comprehensive-assessment dates; PAAF/CMS submission records; authenticated residence/caregiver change facts; clinician-authored dementia-stage findings; and an authorized clinician/program determination that additional clinical assessment is needed. Corroborating: appointments, encounters, caregiver reports, transition records, address changes, and draft assessment records. Identity-only: a calendar label, claim, note title, or copied prior assessment does not establish the occasion or timeliness.

Potential Proxi work. Calculate current public timing windows, compare dates, detect known change events, present the supported occasion and unresolved facts, open the correct work, and keep clinical need separate from CMS data-acceptance cadence. AI may summarize the evidence but may not decide that a new clinical assessment is appropriate.

Human role. No Proxi human is needed when authoritative dates and a recognized occasion match exactly. L1 uses the shared exception for missing administrative dates. The applicable U.S. clinician decides unresolved clinical need within scope. A nonclinical rule ambiguity remains held and is routed to the existing domain owner—Medicaid operations for a Medicaid-program rule or Healthcare Legal Counsel for legal authority—rather than to a generic program/compliance reviewer.

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

Completion evidence. Occasion; supporting dates and source; annual or reassessment timing calculation where applicable; reason; prior-assessment link; and explicit unresolved or missed-deadline status.

What does not prove completion. A due-date alert, a universal “180 days” result, a clinician appointment, a claim, an old assessment, a PAAF draft, or AI classification without the controlling dates.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-012, X-015, CP-E16 as an existing timing cross-reference, O-003, O-004, O-018, and O-PA-001. Test initial; prior assessment; annual days 305/306/425/426; start plus 60 days; residence change day 30; stage/caregiver change days 179/180; duplicate trigger; late corrected date; clinical need before CMS accepts another submission.

1A. Require a new GUIDE assessment when a prior assessment predates alignment#

When. Whenever a prior cognitive, clinical, or comprehensive assessment exists before the person entered the Participant’s GUIDE-specific voluntary-alignment/consent process, and before anyone attempts to count that prior work as the GUIDE initial comprehensive assessment.

What and how much. Make one determination: historical assessment present but a new GUIDE initial comprehensive assessment remains required; no relevant historical assessment found; or dates/status unresolved. The historical assessment remains usable evidence and is not discarded, but it cannot be relabeled as initial GUIDE completion.

Data. Direct: the historical assessment’s actual date and source; documented GUIDE consent date; current initial-assessment record; PAAF submission; and CMS alignment response. Corroborating: claim dates, encounter dates, scheduling history, and copied clinical notes. Identity-only: same practice, same clinician, same diagnosis, or same instrument does not establish that the prior assessment was the required GUIDE initial assessment.

Potential Proxi work. Compare source dates, preserve the earlier assessment as background, prevent completion from being copied forward, and open the new GUIDE assessment work. Proxi should not duplicate source collection that remains current, but it must not bypass the new required assessment.

Human role. No Proxi human is needed for a clean date comparison. L1 pursues a missing or conflicting date under the shared administrative exception; the outside source or Participant-authorized records administrator corrects its own source record. The applicable existing navigator and U.S. clinical categories perform the new assessment tasks later in this SOP.

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

Completion evidence. Prior-assessment identity/date; GUIDE consent/alignment-process dates; determination that the prior work does or does not predate the GUIDE process; new assessment work opened when required; and preserved link to the historical evidence.

What does not prove completion. Same diagnosis, same clinician, CPT 99483 claim, old staging score, copied PAAF fields, current signature on an old note, or an AI statement that the prior assessment is equivalent.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-015, O-018, and O-PA-001. Test one-day-before consent; assessment on consent date with uncertain sequence; same-practice historical assessment; wrong-beneficiary prior note; amended date; duplicate imported record; no historical assessment; late discovery after initial work opened.

2. Confirm the beneficiary#

When. Before collecting, displaying, or sharing person-specific assessment information and before linking prior records, caregivers, clinicians, or medication sources.

What and how much. Establish one uniquely matched beneficiary and a current contact profile sufficient for the planned assessment. Resolve or explicitly hold every material mismatch; do not merge two people because names, dates, or addresses are similar.

Data. Direct: authenticated beneficiary confirmation plus authoritative Participant enrollment/EHR identifiers; or a representative acting under established decision-specific authority. Corroborating: payer demographics, prior authenticated encounters, and prior successful contacts. Identity-only: caller ID, device ownership, emergency-contact fields, email display name, and demographic similarity.

Potential Proxi work. Match identifiers, prefill known details, detect collisions and changed demographics, ask bounded verification questions, and stop person-specific collection when evidence does not uniquely identify the beneficiary. AI confidence cannot waive a conflict.

Human role. No Proxi human is needed for an exact clean match. Software applies the approved identity-match rule and blocks person-specific collection, display, sharing, and record linkage when a material mismatch remains. The beneficiary or authorized representative confirms identity facts; L1 resolves administrative discrepancies. Suspected cross-person record contamination goes to the Participant or EHR operator's authorized records administrator. If the approved rule cannot resolve legal disclosure authority, hold that disclosure and route only that question to healthcare legal counsel.

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

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

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

Edge cases/open decisions. X-001, X-002, X-003, X-004, X-005, X-012, and X-015. Test missing identifier; duplicate names; wrong-beneficiary source; stale address; conflicting contact data; representative without established scope; late demographic correction; death/disenrollment signal.

3. Identify assessment participants#

When. After beneficiary identity is confirmed and before scheduling each required assessment domain. Reassess when caregiver participation, representative authority, beneficiary location, staffing, credential, or clinical scope changes.

What and how much. Produce one participant plan listing the beneficiary; each invited caregiver; any representative; and the exact interdisciplinary team role responsible for each applicable domain. Licensed work must include the beneficiary’s current location and the proposed professional’s actual authority for that act.

Data. Direct: beneficiary caregiver-participation choice; established representative documents and scope; authoritative current staff roster; verified credentials/licensure/privileges; beneficiary location; and approved domain-role policy. Corroborating: EHR care-team labels, prior participation, calendars, and organizational affiliation. Identity-only: job title, degree, NPI, public directory entry, portal proxy, or family relationship does not by itself establish authority for the task.

Potential Proxi work. Compare required domains with the approved role matrix; show missing coverage; match availability; verify current credential and location facts; and prepare the participant list. AI may not assign professional authority or decide a disputed representative role.

Human role. No active Proxi human time is required when the approved role matrix and verified roster produce one valid assignment. HR/workforce administration resolves a real roster-capacity gap; L2 resolves relationship-sensitive participation. Software applies the approved representative-authority rule, with Healthcare Legal Counsel entering only if legal authority remains unresolved. Head of Nursing or the Chief Medical Officer resolves only the nursing- or physician-scope ambiguity actually within that leader's authority under O-003.

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

Completion evidence. Participant list; domain responsibilities; beneficiary location; staff credential/authority source and current status; caregiver permission; representative scope; and unresolved staffing or authority gaps.

What does not prove completion. A name on a calendar, job title, RN or physician credential alone, Puerto Rico or U.S. residence alone, NPI, prior participation, or AI role recommendation.

Edge cases/open decisions. X-002, X-003, X-004, X-005, X-010, X-014, X-015, analogous location issue CC-E43, O-003, O-011, O-040, and O-PA-001. Test missing role; expired credential; clinician licensed elsewhere only; beneficiary travels; representative scope conflict; multiple caregivers; same person assigned outside scope; late staff substitution.

4. Confirm caregiver involvement#

When. During assessment setup and whenever the person providing assistance, caregiver willingness, beneficiary choice, or caregiver participation changes.

What and how much. Identify each person who actually provides unpaid assistance; record the assistance supplied; establish whether that person meets the working caregiver definition; and confirm how they will participate. Keep caregiver status, beneficiary permission, representative authority, and willingness to participate as separate facts. Unknown does not become no caregiver.

Data. Direct: authenticated beneficiary statement about who assists; each caregiver’s statement about assistance and willingness; and established authority documents for any representative role. Corroborating: prior caregiver assessments, care notes, shared appointments, and documented assistance. Identity-only: emergency-contact status, shared address, family relationship, portal access, or phone possession does not establish caregiver status, permission, willingness, or legal authority.

Potential Proxi work. Ask structured questions, maintain one record per caregiver, keep beneficiary and caregiver statements distinct, capture allowed participation, and flag unknown, conflicting, or no-caregiver facts for later tasks. AI may not choose whose statement controls.

Human role. The beneficiary and caregiver state the facts; their time is not Proxi labor. L2 uses two minutes within the combined assessment relationship to confirm the participation profile. Software applies the approved representative-authority and disclosure rules; Healthcare Legal Counsel answers only a remaining legal-authority question. Later tasks handle caregiver assessment or no-caregiver safeguards.

Provisional clean-path Proxi human minutes by role. L1 0; L2 2; L3 0; L4 0.

Completion evidence. Each caregiver’s identity; assistance supplied; involvement/willingness; beneficiary participation choice; source and date; separate representative status; and explicit unknown/no-caregiver/conflict status.

What does not prove completion. A family relationship, emergency-contact designation, shared home, caregiver answering questions, portal proxy, historic involvement, or silence from the beneficiary.

Edge cases/open decisions. 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 no caregiver; unknown caregiver; multiple caregivers; beneficiary declines involvement; caregiver unwilling; conflicting reports; representative overreach; late caregiver change.

5. Capture communication and accessibility needs#

When. Before asking assessment questions or choosing modality, and whenever language, hearing, vision, literacy, cognition, speech, device access, safe contact, or other communication needs change.

What and how much. Create one current communication profile covering preferred language; interpreter need; hearing/vision support; literacy and processing support; usable channel and modality; assistive technology; safe times; and message restrictions. Record unknown rather than defaulting to English, portal, caregiver interpretation, or yes/no comprehension.

Data. Direct: beneficiary or established representative’s current stated needs and preferences; and qualified interpreter/accessibility findings when available. Corroborating: prior successful modality, prior interpreter use, documented accommodation, and observed failed comprehension. Identity-only: demographic language field, diagnosis, device setting, prior delivery, or caregiver presence does not establish current usability.

Potential Proxi work. Offer approved accessible formats and channels, retain original alongside translated content, test the selected connection, arrange qualified support, and flag where software cannot provide meaningful access. AI translation is a draft unless approved policy permits its use.

Human role. The beneficiary/caregiver states what is usable. L2 uses two minutes to confirm the profile in the combined interaction. A qualified interpreter or accessibility professional participates when needed; that external/specialist time is not included. L2 does not substitute as an unqualified interpreter.

Provisional clean-path Proxi human minutes by role. L1 0; L2 2; L3 0; L4 0.

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

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

Edge cases/open decisions. X-003, X-004, X-005, X-007, X-009, X-014, X-015, O-004, and O-011. Test interpreter unavailable; hearing/vision barrier; low literacy; aphasia; inaccessible portal; conflicting channel preferences; failed connection; late accommodation request.

6. Choose the assessment modality#

When. After communication needs and available assessment settings are known, before scheduling the assessment, and again if beneficiary location, technology, preference, or clinical constraints change.

What and how much. Select one usable assessment modality from the permissible options based on beneficiary/caregiver preference and actual availability: telehealth or in person in an allowed setting. Separately record whether the later residence-visit obligation is required, optional, or not yet knowable; a remote initial assessment does not waive a required in-person home visit.

Data. Direct: beneficiary/caregiver modality preference; current beneficiary location; actual technology/accessibility capability; available care settings; and current tier/home-visit obligation when known. Corroborating: prior successful telehealth, transportation facts, caregiver availability, and prior location. Identity-only: device ownership, address, portal activation, or past modality does not establish current usability or clinical authority.

Potential Proxi work. Present permissible options and constraints, check accessibility and technology, record the choice, separate the assessment encounter from the home visit, and prevent licensed telehealth work when current location/authority is unresolved.

Human role. The beneficiary/caregiver chooses among available options; their time is not Proxi labor. No Proxi human is required on a clean rules-based path. L1 handles connection/logistics exceptions; L2 handles preference conflict; clinical/program leadership resolves a genuine modality or scope ambiguity.

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

Completion evidence. Chosen modality/setting; beneficiary preference; current location; accessibility/technology status; applicable clinical constraint; and separate current home-visit-obligation status.

What does not prove completion. Device ownership, portal activation, prior telehealth use, caregiver preference alone, scheduling a remote visit, or assuming a future tier/home-visit result.

Edge cases/open decisions. X-003, X-004, X-005, X-007, X-010, X-012, X-015, analogous location issue CC-E43, O-003, O-004, and O-PA-001. Test unknown tier; remote assessment plus later required home visit; beneficiary traveling; location change mid-visit; no usable technology; caregiver/beneficiary preference conflict; late modality change.

7. Schedule the assessment#

When. After modality and participant requirements are established. Schedule additional domain sessions when one encounter cannot safely or practically complete all required work.

What and how much. Create one coordinated assessment schedule covering all required participants/domains, date/time/timezone, modality/location, accommodations, preparation instructions, reminders, and contingency plan. For an initial assessment, schedule remaining work so the completed PAAF—including required home-visit information—can be submitted within 60 days of the staging date. Apply the separate annual timing when applicable.

Data. Direct: verified participant availability; beneficiary-selected time; modality/location; accommodation resources; staging date once obtained; and authoritative due dates. Corroborating: calendar availability, prior successful contact windows, and travel/transportation information. Identity-only: an email address, calendar identity, or offered appointment does not establish attendance or permission.

Potential Proxi work. Reconcile calendars/timezones, offer eligible times, send accessible preparation instructions and reminders, coordinate multi-session work, reschedule routine conflicts, and warn when remaining work threatens the current CMS deadline.

Human role. No Proxi human is needed on the clean path. L1 uses the shared exception for manual office, interpreter, or failed-contact resolution. L2 enters only when scheduling exposes relationship, distress, authority, or accessibility issues. Clinician availability is an external input, not clinician active minutes.

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

Completion evidence. Confirmed schedule; participant/domain coverage; timezone/modality/location; accommodation status; preparation/reminder delivery; deadline calculation; and contingency or unresolved scheduling gap.

What does not prove completion. An offered time, calendar invitation, unread reminder, one booked clinician without all required domains, or an initial visit date that leaves required work unscheduled.

Edge cases/open decisions. X-003, X-006, X-007, X-012, X-014, X-015, O-004, O-018, and O-PA-001. Test no common time; duplicate booking; timezone error; interpreter unavailable; clinician cancellation; staging date changes deadline; late start; failed reminder; multi-session assessment.

8. Assemble the assessment packet#

When. Before the clinical and functional assessment work and whenever a corrected or later source arrives before final synthesis.

What and how much. Assemble one beneficiary-matched evidence packet organized around the applicable assessment domains: diagnoses, relevant encounters, medication sources, prior assessments, hospital/transition information, providers, services/supports, and beneficiary/caregiver questionnaires. Preserve missing, stale, duplicate, conflicted, and later-arriving material rather than selecting a single “best” source.

Data. Every source is direct only for its source-native assertion: a diagnosis record establishes a recorded diagnosis; a claim establishes billed activity; a dispense record establishes the recorded dispense event; a questionnaire establishes the respondent’s answers. Other sources may corroborate or conflict. Identifiers, filenames, code matches, and directory records are identity-only until matched to the right person, event, and content.

Potential Proxi work. Retrieve authorized connected records, deduplicate exact copies without dropping differing evidence, normalize identities/terminology, organize by domain, retain source/date/author, and flag gaps or conflicts for the correct owner.

Human role. No Proxi human is needed for clean automated assembly. L1 uses the shared exception for existing-record pursuit or identity cleanup. L3/L4 reviews clinical conflicts only when the affected domain requires judgment; the packet itself does not need a universal clinical approval.

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

Completion evidence. Beneficiary-matched packet; source inventory by domain; source/date/author links; exact-duplicate handling; explicit missing/stale/conflicted items; and owners for unresolved retrieval.

What does not prove completion. A document count, copied chart summary, claims-only history, one EHR medication list, unsigned note, old assessment, or AI statement that the packet is complete.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-009, X-011, X-013, and X-015. Test wrong patient; missing domain; stale record; conflicting sources; exact duplicate; nonidentical duplicate; corrupt/illegible file; late correction; outage during retrieval.

9. Obtain the pertinent history#

When. Before and during the cognition-focused evaluation, after the source packet is available, and whenever the beneficiary/caregiver corrects or adds relevant history.

What and how much. Capture cognitive changes, onset/course, symptoms, prior evaluation, medical history, co-occurring conditions, relevant family history when appropriate, and beneficiary/caregiver observations. Attribute every statement to the actual speaker/source and preserve disagreement, uncertainty, and inability to recall.

Data. Direct for the report: authenticated beneficiary and caregiver statements establish what each reports; authenticated clinician records establish that clinician’s recorded history/findings. The dementia-proficient clinician’s verification establishes the pertinent history used for the clinical evaluation. Corroborating: claims, hospital records, prior assessments, pharmacy records, and observed chronology. Identity-only: diagnosis codes, problem-list labels, family relationship, or AI-extracted chronology do not establish current clinical truth.

Potential Proxi work. Present approved structured questions, transcribe/source each answer, build a chronology, compare sources, flag contradictions and gaps, and draft a history for clinician verification. AI may not decide which conflicting account is clinically correct.

Human role. Beneficiary/caregiver time is outside Proxi labor. L3 uses five minutes for structured clinical pre-review, source clarification, and handoff. The L4 dementia-proficient clinician verifies the history as part of Task 10; no separate L4 minutes are added here.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 5; L4 0 here because shared L4 verification is allocated to Task 10.

Completion evidence. Source-attributed history; onset/course and relevant domains; unknowns and conflicts; L3 pre-review; and linked L4 verification during the cognition-focused evaluation.

What does not prove completion. AI summary, problem list, caregiver-only history without attribution, copied old HPI, claims chronology, or completed questionnaire without clinician verification.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-008, X-009, X-015, O-003, and the proposed digital-administration decision. Test beneficiary/caregiver disagreement; unreliable date; missing prior evaluation; wrong-patient history; low-confidence transcription; new urgent symptom; late correction; no caregiver.

10. Perform the cognition-focused examination#

When. During every initial and annual comprehensive assessment and applicable future comprehensive assessment after pertinent history and required accommodations are available. Stop and reroute if beneficiary location or clinician authority is unresolved.

What and how much. A dementia-proficient, beneficiary-location-authorized clinician conducts and clinically interprets the cognition-focused examination appropriate to the person and modality. Record limitations from hearing, language, education, fatigue, acute illness, environment, technology, or inability to complete; do not convert incomplete testing into a normal result.

Data. Direct: the clinician’s actual examination observations/findings, authenticated beneficiary responses, relevant history, and appropriately administered instrument results. Corroborating: prior examinations, functional reports, caregiver observations, and clinical records. Identity/authority-only: license, credential, NPI, roster membership, and location establish who may act but do not establish the examination finding.

Potential Proxi work. Prepare the history and instruments; support accessible delivery; capture responses and clinician dictation; calculate configured scores; show prior results; and draft documentation. Proxi cannot perform the clinical examination, interpret results, diagnose, or decide adequacy.

Human role. L4 dementia-proficient E/M clinician performs the examination and interpretation under actual beneficiary-location authority. L3 may support logistics or observations within scope but has no separate clean-path allocation in this card.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 0; L4 22. This is part of one assessment encounter, not a second standalone review.

Completion evidence. Encounter/date/modality/location; clinician identity, credential, scope, and authority; examination performed; findings and limitations; source inputs; interpretation; and unresolved or incomplete elements.

What does not prove completion. Questionnaire, staging score, history alone, video/audio recording, AI-generated exam note, copied prior finding, clinician signature without performed findings, or diagnosis code.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-008, X-010, X-014, X-015, analogous location issue CC-E43, O-003, O-004, and O-PA-001. Test hearing/language interference; acute delirium concern; telehealth limitation; incomplete exam; clinician out of jurisdiction; location change mid-visit; conflicting prior result; safety signal.

11. Evaluate moderate or high medical decision-making complexity#

When. During the comprehensive clinical evaluation after the relevant problems, data, risks, uncertainty, medications, and examination findings are available.

What and how much. The qualified L4 clinician evaluates and documents the moderate- or high-complexity medical decision-making required by the beneficiary’s presentation and care. This includes the clinician’s actual consideration of problems, evidence, risk, and uncertainty; software does not select complexity or manufacture supporting elements.

Data. Direct: the treating clinician’s documented analysis of the actual problems, data reviewed, risk, uncertainty, and decisions for this encounter. Corroborating: source records, examination findings, medication discrepancies, and consultation information. Identity-only: billing code, problem count, diagnosis list, or risk-alert count does not establish medical decision-making complexity.

Potential Proxi work. Organize source problems/data/risks; show missing or conflicting evidence; draft a source-linked outline; and capture the clinician’s authored reasoning. Proxi cannot choose the complexity level, clinical course, diagnosis, treatment, or billing conclusion.

Human role. The beneficiary-location-authorized L4 clinician with applicable E/M scope performs the judgment. This work occurs in the same assessment encounter as Task 10; the ten minutes are an incremental allocation, not another full visit.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 0; L4 10.

Completion evidence. Encounter-linked clinician-authored MDM analysis; problems/data/risks considered; uncertainty; resulting clinical decisions or truthful no-change; clinician authority; and unresolved dependencies.

What does not prove completion. Auto-selected code, diagnosis count, template completion, copied risk paragraph, AI complexity classification, or signature without encounter-specific reasoning.

Edge cases/open decisions. X-003, X-004, X-005, X-008, X-009, X-010, X-015, analogous location issue CC-E43, O-003, and O-PA-001. Test sparse data; conflicting records; one high-risk problem; copied prior MDM; AI-suggested complexity; out-of-scope clinician; late critical result; incomplete examination.

12. Assess basic activities of daily living#

When. During each comprehensive assessment and whenever reported or observed self-care function changes.

What and how much. Record current performance for each basic ADL addressed by the approved assessment—such as bathing, dressing, eating, toileting, transferring, and mobility—including independent, needs prompting/supervision, needs physical help, unable, declined, or unknown; type/amount of help; source; and usual versus temporary context. Do not collapse all ADLs into one global score.

Data. Direct for the report/observation: beneficiary and caregiver statements establish what each reports; qualified team observation establishes observed performance in that context; a configured instrument establishes its responses/score. Corroborating: home-visit observations, therapy/nursing notes, equipment use, and prior assessments. Identity-only: diagnosis, claim, DME record, or service authorization does not establish current function.

Potential Proxi work. Administer structured questions, retain item-level responses, compare speakers and prior results, calculate configured scores, flag discrepancies/change, and draft the functional record without choosing which report is clinically correct.

Human role. Beneficiary/caregiver response time is outside Proxi labor. L2 uses two minutes to confirm practical item-level facts within the assessment relationship. L3/L4 enters only when discrepancy, clinical interpretation, or action exceeds L2 scope; no routine clinical minute is added here.

Provisional clean-path Proxi human minutes by role. L1 0; L2 2; L3 0; L4 0.

Completion evidence. Item-level ADL status; assistance type/amount; source/speaker/date/context; current versus temporary limitation; score if used; conflicts/unknowns; and beneficiary confirmation where possible.

What does not prove completion. Diagnosis, global “needs help,” caregiver answer without attribution, claim for home care, DME use, old score, or AI-inferred independence.

Edge cases/open decisions. X-003, X-004, X-005, X-008, X-009, X-015, O-003, O-011, and the proposed digital-administration decision. Test beneficiary/caregiver disagreement; fluctuating function; temporary injury; no caregiver; declined item; inaccessible question; fall/safety signal; late home observation.

13. Assess instrumental activities of daily living#

When. During each comprehensive assessment and whenever higher-order daily function or support changes.

What and how much. Record current performance for each IADL addressed by the approved assessment—medication routines, finances, cooking, shopping, transportation, communication, and household tasks—including independent, needs prompting/supervision, needs help, unable, not applicable, declined, or unknown; assistance source/amount; and usual versus temporary context. Medication-routine function is not medication reconciliation, and financial difficulty is not a capacity or exploitation conclusion.

Data. Direct for the report/observation: beneficiary and caregiver statements; qualified observation; and configured instrument responses. Corroborating: missed appointments, home-service notes, transportation records, medication-support use, and prior assessments. Identity-only: diagnosis, residence, bank/contact data, claims, or possession of a pillbox/phone does not establish current IADL ability.

Potential Proxi work. Administer structured questions, capture help level and source, compare speakers/prior results, calculate configured scores, preserve conflict, and flag potential safety or clinical issues without deciding capacity, exploitation, adherence, or treatment.

Human role. Beneficiary/caregiver time is outside Proxi labor. L2 uses two minutes to confirm item-level practical facts. L3/L4 or safeguarding/legal professionals enter only on a named clinical, safety, capacity, or exploitation trigger.

Provisional clean-path Proxi human minutes by role. L1 0; L2 2; L3 0; L4 0.

Completion evidence. Item-level IADL status; assistance type/source/amount; speaker/date/context; score if used; conflicts/unknowns; and separate triggered concerns.

What does not prove completion. Diagnosis, caregiver management of one task, missed refill, late payment, residence type, old score, or AI inference of incapacity/nonadherence.

Edge cases/open decisions. X-002, X-003, X-004, X-005, X-008, X-009, X-015, O-003, O-011, and the proposed digital-administration decision. Test disagreement; no caregiver; fluctuating ability; shared finances; medication reminder support; unsafe driving signal; suspected exploitation; late correction.

14. Screen for hearing loss or arrange screening#

When. During the comprehensive assessment after communication accommodations are known, or whenever existing hearing information is missing, stale, uncertain, or contradicted by current communication difficulty.

What and how much. Produce one truthful outcome: appropriate hearing screen completed with interpretable result; external hearing screening arranged because the GUIDE team cannot complete it; unable/declined; or unresolved. A referral is “arranged” only when an authorized request reached a verified destination with ownership/follow-up—not when a draft exists.

Data. Direct: result from the approved screening method administered under required conditions by the qualified role; or an authenticated hearing-professional report. For the referral route, the authorized release and verified destination/receipt directly establish referral arrangement, not the eventual external result. Corroborating: beneficiary/caregiver report, prior audiology, hearing-aid use, failed communication, and clinician observation. Identity-only: hearing-aid purchase, diagnosis code, directory result, or device setting does not establish current screening result.

Potential Proxi work. Deliver a configured screening flow, record method/conditions/result, identify invalid or incomplete administration, prepare the exact referral and verified destination, and track release/receipt. Proxi cannot clinically interpret an uncertain result or autonomously release V1 PHI.

Human role. L3 uses two minutes to verify clean screening completion/result and choose the approved screen-versus-referral disposition within scope. A hearing professional supplies an external result. An authorized human releases any PHI referral; abnormal/uncertain findings trigger additional qualified review.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 2; L4 0. This assumes an internally completed clean screen; external referral labor is separate.

Completion evidence. Screening method, conditions, date, qualified actor, result/validity; or exact referral content, authorization, verified destination, release/receipt, owner, and follow-up status; plus declined/unresolved status where applicable.

What does not prove completion. Self-report alone, hearing aid, diagnosis code, generated referral, transmission without receipt, directory endpoint, old screening, or AI interpretation.

Edge cases/open decisions. X-003, X-004, X-005, X-006, X-007, X-009, X-014, X-015, D-007, O-003, O-004, and the proposed hearing-screen decision. Test normal screen; invalid noisy test; unable to follow prompts; abnormal result; declined screen; referral wrong endpoint; no receipt; late external result.

15. Stage the dementia with a standardized instrument#

When. During the initial and annual comprehensive assessment and whenever a permitted reassessment requires updated dementia severity. For the initial assessment, the staging-tool administration date becomes the PAAF assessment date and starts the current 60-day completion/submission period.

What and how much. Use the current CMS-required staging option—currently CDR or FAST in the Payment Methodology Paper—selected and administered as allowed by approved clinical policy. Preserve raw responses/observations, configured arithmetic, administration conditions, score, limitations, and the qualified clinician’s interpretation. Do not force an uninterpretable or incomplete administration into a stage.

Data. Direct: instrument responses and observations obtained under the approved method; deterministic score calculation; and the qualified clinician’s interpretation in context. Corroborating: prior staging, cognition exam, functional findings, caregiver report, and clinical history. Identity-only: diagnosis code, prior tier, instrument name, or imported score does not establish current stage.

Potential Proxi work. Present approved prompts, capture raw inputs, calculate the configured score, check completeness/range, compare longitudinal results, record administration date, and flag conflict or invalidity. Proxi cannot select the instrument for the person, interpret stage, or infer CMS tier.

Human role. L3 uses five minutes to administer/support and verify raw inputs and scoring within scope. L4 uses five minutes to interpret the result with the cognition/functional evidence and own the clinical stage. These are linked portions of one assessment, not duplicate reviews.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 5; L4 5.

Completion evidence. Instrument name; administration date; raw inputs/source; administrator and authority; score; completeness/limitations; clinician interpretation and identity/authority; discrepancy disposition; and linked PAAF timing.

What does not prove completion. Imported score, old stage, diagnosis code, current tier, calculator output, incomplete instrument, caregiver-only estimate, or AI stage classification.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-009, X-010, X-015, analogous location issue CC-E43, O-003, O-018, O-039, and O-PA-001. Test CDR/FAST selection; incomplete items; hearing/language barrier; score/clinical conflict; prior score copied; staging date changed; out-of-scope interpreter; late correction.

16. Collect the medication-use information#

When. During every initial, annual, and additional future comprehensive assessment before the prescribing-authority review, and whenever a late medication source arrives before reconciliation is finalized.

What and how much. Gather every prescription, over-the-counter product, supplement, natural treatment, and other substance the beneficiary might use. For each, retain name/ingredients as knowable, strength/form, route, schedule/PRN directions, purpose as reported, prescriber/source, pharmacy, possession, reported use, last use when relevant, and unknowns. Prescribed, ordered, dispensed, possessed, reported-taking, administered, stopped, and reconciled remain separate facts.

Data. Each source is direct only for its native claim: an order proves an order; a dispense record proves its recorded dispense event; a bottle/label proves observed packaging; beneficiary/caregiver statement proves the report; facility MAR proves recorded administration; pharmacy/prescriber records prove their documented events. Claims, fills, lists, bottles, or reports alone do not establish the current intended regimen or actual ingestion. RxNorm supports identity normalization, not therapeutic equivalence or clinical truth.

Potential Proxi work. Retrieve authorized records, ask source-faithful questions, normalize unambiguous identities, preserve unknown/multi-ingredient products, compare sources, identify duplicates/conflicts/gaps, and prepare the prescriber review. AI cannot select the current regimen or dismiss a source.

Human role. Beneficiary/caregiver reporting and external pharmacy/prescriber time are outside Proxi labor. L3 uses six minutes for clinical-support completeness review and source clarification without reconciling. L1 may pursue existing administrative records under the shared exception. L4 reconciles in Task 17.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 6; L4 0 here because prescribing-authority reconciliation is allocated to Task 17.

Completion evidence. Itemized source-faithful medication/substance inventory; source/date for each assertion; unknown identity/strength/schedule; source conflicts; beneficiary/caregiver reports; L3 completeness review; and explicit handoff to Task 17.

What does not prove completion. EHR medication list, claims, fill history, bottle photograph, discharge list, caregiver list, RxNorm match, “medications reviewed” checkbox, or AI-generated best list.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-008, X-009, X-015, MED-E02, MED-E03, MED-E04, MED-E07, MED-E09, MED-E12, MED-E15, MED-E16, MED-E17, MED-E18, MED-E39, MED-E40, O-027, O-028, and O-035. Test cash/sample medication; reversed claim; old bottle; PRN; supplement; multi-ingredient unknown; caregiver conflict; facility MAR; outside order; allergy/reaction conflict; safety symptom; late source.

17. Reconcile and review medications#

When. At the initial comprehensive assessment, every additional future comprehensive assessment, and the annual assessment after Task 16 supplies the source-faithful medication-use information. Do not close from one source or proceed through a clinician lacking actual prescribing authority for the beneficiary/location.

What and how much. The clinician with prescribing authority compares all medication/substance sources, determines the current intended regimen, preserves reported use and possession separately, resolves or explicitly retains discrepancies, and clinically reviews appropriateness, interactions, duplication, adherence/use concerns, allergy/reaction issues, and needed follow-up. A no-change conclusion is an affirmative clinical result, not absence of alerts.

Data. Direct for reconciliation: the prescribing-authority clinician’s documented medication-by-medication disposition grounded in the source assertions and current clinical context. Orders, pharmacy events, possession, reported use, administration, labs, diagnoses, and allergy/reaction reports retain their separate meanings. AHRQ MATCH supports explicit source verification, shared medication documentation, and defined roles as design guidance; it does not replace GUIDE prescribing authority or decide the regimen.

Potential Proxi work. Build the source comparison and discrepancy list, display supporting records, calculate configured interaction/duplicate signals for clinician review, capture the clinician’s disposition for each item, and create linked follow-up work. AI cannot reconcile, determine appropriateness, select a medication action, or declare no change.

Human role. The exact prescribing-authority matrix remains open under O-028; on the provisional clean path, an L4 beneficiary-location-authorized prescribing clinician performs and attests to the reconciliation/review. L3 may prepare but does not substitute. External prescribers/pharmacies clarify their own records; their time is not Proxi labor.

Provisional clean-path Proxi human minutes by role. L1 0; L2 0; L3 0; L4 15. This is the assessment reconciliation allocation, not an additional duplicate Medication-service review.

Completion evidence. Prescribing clinician identity, credentials, location authority, date; every in-scope medication/substance and source; current intended regimen; reported-use/possession distinctions; item-level discrepancy disposition; clinical review/no-change conclusion; unresolved items and safety/follow-up work.

What does not prove completion. Latest list, signed imported list, claims, dispense record, bottle inventory, interaction-alert absence, non-prescribing review, AI comparison, or clinician signature that does not address unresolved discrepancies.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-008, X-010, X-014, X-015, MED-E02, MED-E03, MED-E04, MED-E07, MED-E09, MED-E12, MED-E16, MED-E17, MED-E18, MED-E22, MED-E34, MED-E39, MED-E40, MED-E44, O-003, O-004, O-027, O-028, O-032, O-035, and O-PA-001. Test clean no-change; one unknown medication; conflicting dose; OTC/supplement; outside order; caregiver disagreement; beneficiary refusal/self-change; no valid prescriber; out-of-jurisdiction clinician; allergy conflict; safety signal; late material record.

18. Screen for depression#

When. During each comprehensive assessment after identity, accessibility, and a safe response route are established; repeat only when the approved instrument or a qualified clinician calls for it. A new self-harm response immediately invokes Tasks 21 and 22.

What and how much. Administer the approved depression screen exactly as configured, preserve each response and nonresponse, calculate only the deterministic score, and record administration limitations. The output is a screen result, not a diagnosis, severity determination, treatment recommendation, or proof that symptoms are absent.

Data. Direct: the authenticated respondent's item-level answers, administration conditions, and deterministic score; a qualified clinician's interpretation establishes the clinical disposition. Corroborating: prior screens, reported mood changes, clinical notes, medication changes, and caregiver observations. Identity-only: a diagnosis code, antidepressant fill, or old score does not establish current depression or current safety.

Potential Proxi work. Present approved accessible questions, capture verbatim responses, calculate the configured score, detect missing/invalid items, compare prior results, and route configured positive, unclear, or self-harm responses. AI may not interpret depression, infer a negative answer, or select the follow-up.

Human role. The beneficiary supplies answers. Separate U.S. L3 RN/LCSW/behavioral staff verifies routine completion and evaluates a positive, unclear, or clinically discordant result within scope; L4 enters when diagnosis, medical evaluation, or treatment judgment is needed. Puerto Rico L2 does not perform clinical review.

Provisional Proxi human minutes (low / typical / high). L1 Philippines 0 / 0 / 0; L2 Puerto Rico nonclinical navigator 0 / 0 / 0; separate U.S. L3 0 / 3 / 12; separate U.S. L4 0 / 0 / 8. High assumes a non-imminent positive or unclear result; imminent risk belongs to Task 22 and is not double-counted here.

Completion evidence. Instrument/version; respondent; item-level responses; date, mode, language and accommodations; score and validity; qualified review/disposition when triggered; and links to any safety or follow-up work.

What does not prove completion. Questionnaire delivery, a total score without items, an old score, an AI mood summary, medication history, a checked template box, or a positive score without qualified disposition.

Edge cases/open decisions. X-003, X-004, X-007, X-008, X-009, X-010, X-014, X-015, O-003, O-004, and O-PA-001. The approved instrument, administration qualifications, score thresholds, and digital clean-path policy remain to be settled. Test complete negative; positive; skipped self-harm item; proxy response; language barrier; discordant caregiver report; duplicate administration; delayed response; and positive result while the live response route is unavailable.

19. Screen for anxiety#

When. During each comprehensive assessment after communication supports are active, and when a qualified clinician determines that an updated screen is needed because symptoms or circumstances changed.

What and how much. Administer the approved anxiety screen, preserve item-level responses and context, calculate only the configured score, and distinguish incomplete, declined, invalid, negative, and review-required results. Screening does not diagnose an anxiety disorder or determine treatment.

Data. Direct: authenticated item-level answers and deterministic score; the qualified reviewer's documented interpretation establishes the clinical disposition. Corroborating: reported distress, sleep or behavior changes, prior screens, clinical notes, and caregiver observations. Identity-only: anxiolytic medication, a diagnosis code, or prior behavioral-health visit does not establish current anxiety.

Potential Proxi work. Deliver accessible prompts, calculate the score, identify missing items and material change, and prepare a source-linked review packet. AI may not interpret symptoms, diagnose anxiety, or decide urgency or treatment.

Human role. Separate U.S. L3 RN/LCSW/behavioral staff verifies a typical screen and reviews positive, unclear, or discordant findings. L4 or an external behavioral clinician enters only for medical diagnosis/treatment or another act requiring that authority.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 0; separate U.S. L3 0 / 2 / 10; separate U.S. L4 0 / 0 / 6.

Completion evidence. Instrument/version, item responses, score, conditions/accommodations, validity, reviewer/disposition where triggered, and linked follow-up.

What does not prove completion. A delivered questionnaire, diagnosis code, medication list, caregiver characterization, AI sentiment, score without the items, or referral draft without review.

Edge cases/open decisions. X-003, X-004, X-007, X-008, X-009, X-010, X-015, O-003, and O-004. Test complete negative; positive; incomplete; declined; proxy response; cognitive/language limitation; acute distress; duplicate screen; late correction; and depression/anxiety score conflict.

20. Screen for substance use#

When. During the comprehensive assessment in a private and accessible setting, after the recipient knows who can see the answers and a qualified response route is available.

What and how much. Ask the approved questions about alcohol, prescribed-medication misuse, non-prescribed drugs, tobacco/nicotine if included by policy, and other relevant substances. Preserve substance, pattern, timing, source, declined/unknown status, and any immediate concern. A screen does not establish a substance-use diagnosis, unlawful conduct, intoxication, causation, or treatment need.

Data. Direct: the beneficiary's authenticated responses establish what was reported; a qualified clinician's evaluation establishes the clinical meaning. Corroborating: medication records, prior screens, toxicology or clinical records within permitted use, and caregiver observations attributed to the caregiver. Identity-only: a prescription, claim, diagnosis, or pharmacy event alone does not establish misuse or current consumption.

Potential Proxi work. Present private accessible prompts, preserve source-faithful answers, calculate configured scores, segment specially protected content as required, and route concerning responses. AI may not infer misuse from fills, decide whether a disclosure is truthful, diagnose, or choose a disclosure destination.

Human role. A U.S. L3 behavioral clinician or RN within approved scope reviews the typical result and any concern. The U.S. L4 clinician or an outside addiction/medical professional enters only for diagnosis or treatment within authority. Software segments and restricts specially protected information under the approved rule; Healthcare Legal Counsel enters only if the rule leaves a legal-authority question unresolved. L1/L2 do not clinically interpret or disclose the protected content.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 0; separate U.S. L3 0 / 3 / 12; separate U.S. L4 0 / 0 / 8.

Completion evidence. Instrument/questions/version; privacy conditions; respondent and item responses; score if applicable; declined/unknown items; qualified review/disposition when triggered; permitted access controls; and linked follow-up.

What does not prove completion. Pharmacy fills, a medication list, a diagnosis code, caregiver suspicion, toxicology without context, AI classification, or a screen result without a qualified disposition when required.

Edge cases/open decisions. X-002, X-003, X-004, X-008, X-009, X-015, analogous protected-information issue CC-E41, O-003, O-004, and O-011. Test private versus caregiver-present administration; declined answer; cognitive limitation; prescribed controlled medication; conflicting reports; possible intoxication; specially protected data; positive screen; and immediate safety concern.

21. Screen for suicidal ideation#

When. During the approved behavioral-health screening only when a live, separate U.S. qualified response path can receive an immediate positive response. Never launch this as unattended asynchronous outreach without the approved real-time safety design.

What and how much. Ask the approved questions exactly, preserve each answer and time, and treat any configured positive, ambiguous, interrupted, or technically failed response according to the approved fail-safe. Screening captures responses; it does not determine intent, plan, means, imminence, protective factors, capacity, or safe disposition.

Data. Direct: the beneficiary's actual response and the qualified professional's subsequent risk assessment/disposition. Corroborating: prior risk assessments, caregiver report, clinical observations, and recent events. Identity-only: prior diagnosis, medication, utilization, keyword, sentiment, or model risk score does not establish current suicidal ideation or current safety.

Potential Proxi work. Present approved questions only when the live response precondition passes, capture responses verbatim, immediately summon the designated U.S. clinical path on configured triggers, retain connection attempts, and stop ordinary assessment work. AI may not conduct the risk assessment, downgrade ambiguity, or declare the person safe.

Human role. Separate U.S. L3 LCSW/behavioral clinician or other beneficiary-location-authorized professional verifies the completed negative screen and immediately evaluates any positive/unclear response; L4/external emergency services enter according to actual need and scope. L1 may capture callback/location and warm-transfer under protocol but cannot triage; L2 provides nonclinical relationship support only after the clinical path owns safety.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 5; L2 0 / 0 / 5; separate U.S. L3 0 / 1 / 20; separate U.S. L4 0 / 0 / 10. High includes the initial qualified assessment only; emergency-service and external-party time is excluded.

Completion evidence. Question set/version, live-response readiness, respondent, exact responses/times, administration continuity, qualified reviewer identity/authority, risk disposition, successful handoff evidence, failed-handoff retries, and linked Task 22 case when triggered.

What does not prove completion. Sent questionnaire, keyword alert, negative sentiment analysis, caregiver reassurance, voicemail, a warm-transfer attempt, clinical notification without acceptance, or a positive response without qualified assessment and disposition.

Edge cases/open decisions. X-003, X-007, X-008, X-010, X-013, X-014, X-015, CP-E11, O-003, and O-004. Test negative; passive thought; ambiguous answer; skipped item; connection loss after disclosure; unknown location; beneficiary alone; caregiver intercepts; live clinician unavailable; duplicate alert; and a new disclosure after assessment closure.

22. Respond to an urgent behavioral-health or safety concern#

When. Immediately when any assessment answer, observation, communication, or late-arriving fact may indicate imminent or time-sensitive harm. This interrupt applies from every assessment state and does not wait for the routine assessment to finish.

What and how much. Stop affected routine work, preserve the source verbatim, establish the person's current physical location and callback method without delaying emergency connection, obtain acceptance by the approved U.S. clinical/emergency/protective pathway, and document disposition and safe ownership. The assessment team does not itself determine that the person is safe unless the acting professional has that authority.

Data. Direct: source-faithful disclosure/observation; current location as reported/verified; connection and acceptance events; and the qualified receiver's assessment/disposition. Corroborating: prior risks and caregiver/provider reports. Identity-only: alert priority, model score, message delivery, dialed number, or a referral record does not establish urgency, successful handoff, or safety.

Potential Proxi work. Interrupt routine workflows, display approved immediate instructions, summon the correct U.S. destination, pass the exact minimum necessary packet after the applicable authority gate, retry failed handoffs, preserve all attempts, and suppress unsafe routine messaging. AI may not select urgency, diagnose, or close the safety case.

Human role. L1 may perform first-answer identity/location/callback capture and warm transfer; L2 may remain as nonclinical relationship support. A separate U.S. beneficiary-location-authorized L3/L4 professional or external emergency/protective service performs clinical/safety assessment and disposition. Puerto Rico never becomes the clinical responder.

Provisional Proxi human minutes (low / typical / high). This card is event-triggered: L1 0 / 0 / 10; L2 0 / 0 / 10; separate U.S. L3 0 / 0 / 25; separate U.S. L4 0 / 0 / 15. External emergency/protective-service time is excluded.

Completion evidence. Trigger/source/time; person/location/callback; routine-work stop; destination and authority; connection attempts; receiver acceptance; qualified assessment/disposition; communications; owner and follow-up; and separate status of the interrupted assessment.

What does not prove completion. Alert creation, dialing, voicemail, text delivery, transfer initiation, caregiver promise, emergency-number display, clinician notification without acceptance, or completion of the interrupted assessment.

Edge cases/open decisions. X-001, X-003, X-007, X-008, X-010, X-013, X-014, X-015, CP-E11, CC-E13, CC-E43, O-003, and O-004. Test unknown/wrong location; disconnected call; receiver declines; after-hours event; caregiver is alleged source of danger; abuse concern; false match; duplicate signal; concern after terminal assessment; and safe resume versus linked successor work.

23. Assess home and environmental safety#

When. During the comprehensive assessment through structured report, and again through direct observation when a residence visit occurs. Keep reported safety and observed safety as separate evidence.

What and how much. Address the approved hazard domains, including access/egress, fire and cooking risks, utilities, lighting, stairs, bathroom, mobility paths, medication/storage hazards, weapons or dangerous items when policy permits, emergency readiness, and existing supports. Record present, absent-by-report, observed, unknown, declined, and not assessed separately; do not infer a safe home from missing hazards.

Data. Direct: beneficiary/caregiver reports establish what they report; a qualified visitor's dated observation establishes what was visible in that setting; a clinician's interpretation establishes clinical risk. Corroborating: photographs with permission, prior visit notes, falls, emergency events, and service records. Identity-only: address, residence type, DME claim, or old checklist does not establish current safety.

Potential Proxi work. Administer the checklist, retain item/source/context, compare reports with later observations, organize permitted media, identify missing/conflicting items, and route configured safety signals. AI cannot determine that a home is safe or that a safeguard is sufficient.

Human role. L2 may collect and clarify nonclinical practical facts. Separate U.S. L3 reviews typical clinical/safeguarding implications and any material discrepancy. L4 enters only for a medical/treatment judgment. The residence visitor records direct observations in Tasks 40-42.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 3 / 8; separate U.S. L3 0 / 2 / 10; separate U.S. L4 0 / 0 / 5.

Completion evidence. Domain-level answers; speaker/source/date; observed-versus-reported status; visit context where applicable; unknown/declined items; hazards and existing safeguards; qualified disposition for triggered concerns; and linked actions.

What does not prove completion. Address, residence type, photograph alone, completed checklist without source, absence of reported hazards, old home-health note, or AI-generated risk label.

Edge cases/open decisions. X-003, X-004, X-008, X-009, X-015, O-003, O-004, O-011, and O-013. Test report/observation conflict; no visit; unsafe photo without context; temporary residence; hoarding/clutter; utilities shutoff; weapon disclosure; caregiver minimizes risk; and immediate hazard.

24. Assess driving safety#

When. During each comprehensive assessment when the beneficiary currently drives, may resume driving, or has a relevant recent incident or concern. Record a truthful not-currently-driving result without assuming the issue can never recur.

What and how much. Establish current driving status, frequency/routes, recent crashes/near misses/getting lost/citations, observed difficulties, vehicle access, self-restrictions, who raised concern, and existing evaluation or restrictions. A qualified clinician determines the clinical concern and follow-up; this task does not itself declare fitness, legal incapacity, or license status.

Data. Direct: authenticated reports; official licensing/incident records for their native facts; actual qualified evaluation results; and clinician interpretation. Corroborating: cognition/function findings, medication issues, vision/hearing data, caregiver observations, and prior notes. Identity-only: vehicle ownership, driver's license record, diagnosis, age, or isolated claim does not establish present fitness.

Potential Proxi work. Ask structured questions, build an incident chronology, compare speakers and prior status, retrieve permitted records, and prepare follow-up options. AI may not decide driving fitness, recommend a restriction, report to an authority, or select urgency.

Human role. Separate U.S. L3 gathers/clarifies clinical safety facts; the beneficiary-location-authorized L4 clinician determines clinical concern and next evaluation/action within scope. L2 coordinates only approved transport alternatives and follow-up after the clinical decision.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 8; separate U.S. L3 0 / 3 / 10; separate U.S. L4 0 / 3 / 12.

Completion evidence. Current driving status; itemized events/restrictions; source/date; conflicts/unknowns; clinician identity/authority and disposition; required communication/reporting decision; and transport/follow-up plan.

What does not prove completion. License possession, no recent claim, caregiver opinion alone, dementia stage alone, AI risk score, generic driving brochure, or referral without clinical disposition.

Edge cases/open decisions. X-003, X-004, X-008, X-010, X-015, O-003, O-004, and O-011. Applicable state-law reporting, evaluation, confidentiality, and documentation rules remain open. Test no longer drives but keeps keys; beneficiary/caregiver disagreement; recent crash; getting lost; rural transport constraint; unknown location; clinician outside jurisdiction; and immediate intent to drive after a serious event.

25. Assess wandering or getting lost#

When. During each comprehensive assessment and whenever a new episode, pattern, living-setting change, or safeguard failure is reported.

What and how much. Capture each relevant incident's date, place, duration, circumstances, destination-seeking behavior, supervision, recovery, injury/exposure, recurrence, current access/egress, and existing safeguards. Distinguish wandering, purposeful walking, disorientation, elopement, and unknown rather than forcing one label.

Data. Direct: beneficiary/caregiver/witness reports establish what each reports; official or clinical records establish their recorded events; clinician interpretation establishes individualized risk. Corroborating: location-device records, emergency contacts, residence observations, and prior events. Identity-only: dementia diagnosis, stage, locked door, tracker ownership, or prior wandering code does not establish current risk or adequate safeguards.

Potential Proxi work. Ask incident-specific questions, create a chronology, document current safeguards and failures, compare sources, provide clinician-approved educational content, and route configured urgent conditions. AI cannot determine risk sufficiency, prescribe restriction, or choose a safeguard.

Human role. L2 clarifies nonclinical household routines and current supports. Separate U.S. L3 interprets concerning patterns and safeguarding needs within scope; L4 enters for medical causes or treatment judgment.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 2 / 8; separate U.S. L3 0 / 2 / 10; separate U.S. L4 0 / 0 / 6.

Completion evidence. Incident-level chronology; sources/dates; current context and safeguards; failed safeguards; conflict/unknown status; qualified risk disposition when triggered; and linked safety/support actions.

What does not prove completion. Diagnosis or stage alone, tracker purchase, locked residence, caregiver reassurance, generic safety plan, AI label, or absence of a documented prior episode.

Edge cases/open decisions. X-003, X-004, X-008, X-009, X-015, O-003, O-004, O-011, and O-013. Test purposeful walk versus disorientation; first episode; recurrent events; no caregiver; locked-exit concern; tracker unavailable; temporary residence; late report; and person currently missing.

26. Assess fall risk#

When. During each comprehensive assessment and after any new fall, near fall, mobility change, medication change, acute illness, or home observation that may materially change risk.

What and how much. Capture falls/near falls with dates, circumstances, injury, loss of consciousness if reported, assistance needed, mobility/balance symptoms, devices, footwear, environment, vision, orthostatic or medication concerns as reported, and prior evaluation. Apply only the approved assessment method; a score does not by itself determine cause or treatment.

Data. Direct: respondent reports; observed performance under the approved method; medical records for their documented events/findings; deterministic instrument score; and clinician interpretation. Corroborating: claims, EMS records, therapy notes, DME, home observation, and medication evidence. Identity-only: fall claim, cane, diagnosis, or age does not establish current individualized risk or cause.

Potential Proxi work. Extract candidate fall events, administer structured questions, calculate configured scores, compare chronology and sources, and prepare source-linked risk factors. AI may not diagnose cause, decide urgency, recommend medication/equipment/treatment, or declare low risk.

Human role. Separate U.S. L3 performs structured clinical assessment within scope; L4 determines medical causes, diagnostic work, or treatment when required. L2 may coordinate approved practical supports after the clinical disposition.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 6; separate U.S. L3 0 / 3 / 12; separate U.S. L4 0 / 2 / 10.

Completion evidence. Event history; sources/dates; assessment method/items/score/limitations; observed-versus-reported findings; clinician identity/authority and disposition; and linked urgent, diagnostic, treatment, or support actions.

What does not prove completion. Fall claim, DME, old score, no reported fall, AI risk classification, templated precautions, or referral without qualified review.

Edge cases/open decisions. X-003, X-004, X-008, X-009, X-015, MED-E35 as a stale-data analogue, O-003, and O-004. Test fall with head strike; near falls only; unreliable chronology; beneficiary/caregiver disagreement; inaccessible mobility test; temporary injury; new sedating medication; duplicated claim; and current acute symptoms.

27. Assess abuse, neglect, and exploitation#

When. During the assessment using the approved private/safe method, and immediately whenever a disclosure, observation, coercion signal, unexplained injury, unmet basic need, self-neglect concern, or financial concern arises. Do not ask in a channel controlled by a possible source of harm unless the approved safeguarding path directs it.

What and how much. Ask and observe within the approved protocol for physical, emotional, sexual, and financial abuse; neglect; self-neglect; coercion; and exploitation. Preserve the person's words, visible observations, alleged actor, current access, immediate danger, and reporting preference where lawful. The team does not determine that abuse occurred merely from a signal.

Data. Direct: source-faithful disclosure; qualified observation; authenticated records for their native assertions; and the authorized safeguarding professional's disposition. Corroborating: injuries, financial/service changes, caregiver conflict, home conditions, and prior reports. Identity-only: family role, shared address, unusual transaction, diagnosis, model score, or caregiver burden does not prove abuse or safety.

Potential Proxi work. Verify that the channel is safe, present approved prompts, capture exact words and observations, minimize disclosure, summon the designated U.S. safeguarding path, and preserve reporting/handoff evidence. AI may not decide credibility, perpetrator, mandatory-reporting applicability, or safety.

Human role. Separate U.S. L3 LCSW/behavioral clinician or other trained authorized professional reviews the typical screen and owns safeguarding assessment within scope. L4 enters for medical evaluation. Protective services, law enforcement, legal, or compliance acts under applicable law/policy. L1/L2 may connect and support but cannot investigate or decide reporting.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 5; L2 0 / 0 / 8; separate U.S. L3 0 / 2 / 20; separate U.S. L4 0 / 0 / 10. External protective/legal time is excluded.

Completion evidence. Safe/private administration conditions; exact disclosure/observations and source; current location/access to alleged actor; qualified disposition; applicable reporting decision and authority; successful handoff; beneficiary communication; and follow-up ownership.

What does not prove completion. Automated flag, caregiver burden score, financial anomaly, photograph alone, message to a clinician, report draft, transmission without receipt, or a note stating “no abuse” without a valid private assessment.

Edge cases/open decisions. X-002, X-003, X-007, X-008, X-010, X-014, X-015, analogous privacy case CC-E41, O-003, O-004, and O-011. Jurisdiction-specific mandatory-reporting and confidentiality policy remains open. Test caregiver present; alleged representative; beneficiary recants; communication device monitored; self-neglect; financial exploitation; uncertain location; immediate danger; failed protective-service handoff; and late disclosure.

28. Screen health-related social needs#

When. During each comprehensive assessment using the current Participant-approved HRSN instrument and whenever a major change makes earlier factual needs materially stale. Screening is distinct from referral, service connection, and resolution.

What and how much. Capture the required domains and item-level responses, including food, housing, utilities, transportation, interpersonal safety, and social connection when included by the approved instrument. Record present, absent-by-report, declined, unknown, urgent signal, and already-addressed status separately; do not infer no need from a skipped item.

Data. Direct: authenticated beneficiary/caregiver responses establish what was reported; authoritative program/service records establish their native eligibility, authorization, enrollment, or delivery events. Corroborating: payer/Medicaid data, claims, prior screens, referral outcomes, and care notes. Identity-only: ZIP code, dual eligibility, claim, authorization, resource-list match, or prior referral does not establish a current need or current service receipt.

Potential Proxi work. Deliver the approved screen, preserve item/source/date, calculate configured categories, prefill only clearly sourced facts for confirmation, identify conflicts and stale items, and prepare a need list for L2. AI may not decide eligibility, urgency, provider fit, or that a need is resolved.

Human role. L2 confirms practical need facts, beneficiary priorities, and whether follow-up is wanted. Separate U.S. clinical staff enter only for a named clinical or immediate-safety trigger; ordinary HRSN screening is not clinical work.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 3 / 8; separate U.S. L3 0 / 0 / 8; separate U.S. L4 0 / 0 / 5. Clinical high minutes apply only to a triggered concern.

Completion evidence. Instrument/version; item responses; respondent/source/date; language/accommodations; unknown/declined items; identified needs and beneficiary priorities; triggered safety/clinical route; and explicit handoff to referral/support work.

What does not prove completion. ZIP-code deprivation score, dual-eligibility flag, claim, authorization, resource list, referral sent, scheduled intake, or AI-inferred need.

Edge cases/open decisions. X-003, X-004, X-005, X-008, X-009, X-015, CC-E07, CC-E17, CC-E28, O-003, O-004, and O-023. Test no needs; declined item; urgent food/housing loss; already receiving a service; eligibility without receipt; conflicting beneficiary/caregiver report; inaccessible screen; late service record; and resource inventory unavailable.

29. Ask about advance-care-planning wishes#

When. During each comprehensive assessment after communication and decision-participation supports are established, and before initiating any advance-care-planning or POLST work.

What and how much. Explain the opportunity, voluntary nature, available discussion/document types, and ability to defer or revisit using approved neutral content. Record whether the beneficiary wants to develop, revise, review, defer, or decline. This preference is decision-specific and does not establish capacity, representative authority, medical orders, or completed advance care planning.

Data. Direct: the authenticated beneficiary's choice, or an authorized representative's choice within established scope. Existing documents directly establish their own text/status only after identity and validity checks. Corroborating: prior preferences and clinician notes. Identity-only: age, diagnosis, prior billing code, uploaded form, family relationship, or portal proxy does not establish the current wish or authority.

Potential Proxi work. Present approved accessible education, locate existing candidate documents, record the choice, schedule the requested qualified conversation, and preserve refusal/deferral without repeated coercive prompting. AI may not determine capacity, recommend choices, or interpret a document's legal effect.

Human role. L2 provides the neutral nonclinical opportunity explanation and authenticates the preference. A qualified clinician, legal resource, or trained professional enters only for the requested substantive discussion or authority/capacity issue.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 2 / 5; separate U.S. L3 0 / 0 / 5; separate U.S. L4 0 / 0 / 5.

Completion evidence. Exact approved explanation/version; recipient and accessibility mode; authenticated choice/date; existing-document status; requested next step/owner; and any authority/capacity hold.

What does not prove completion. Brochure delivery, portal click, existing document, billing code, caregiver request, scheduled conversation, or silence.

Edge cases/open decisions. X-002, X-003, X-004, X-007, X-009, X-015, CP-E04, CP-E05, CP-E06, O-011, and O-PA-001. Test beneficiary declines; defers; requests review; caregiver asks but beneficiary objects; representative scope unknown; communication barrier; existing conflicting documents; and later change of preference.

30. Develop, revise, or review the advance care plan#

When. Only after Task 29 records a current request, and whenever the beneficiary requests review or a material preference/document change is identified. Do not treat this as mandatory document completion.

What and how much. Conduct a beneficiary-led conversation about goals, values, important outcomes, unacceptable burdens, surrogate/decision-support preferences, and questions for clinicians; review existing documents; and prepare or revise only the documents permitted by applicable law and policy. Keep values, legal instruments, and medical orders distinct.

Data. Direct: authenticated beneficiary statements; established representative authority for the specific act; the text/signatures/effective status of validated documents; and qualified clinician/legal determinations within scope. Corroborating: prior notes and family reports. Identity-only: uploaded form, witness name, family relationship, diagnosis, or prior code does not establish validity, capacity, current preference, or applicability.

Potential Proxi work. Assemble existing documents, provide approved educational prompts, capture source-attributed values, compare versions, draft allowed text, identify missing signatures/witnesses, and route medical/legal questions. AI may not determine capacity, legal validity, medical recommendations, or create an order.

Human role. L2 facilitates the nonclinical values/preferences conversation. Separate U.S. L3/L4 or another qualified professional handles medical explanation, decision-specific capacity, or document requirements within actual authority. Legal/compliance resolves legal-effect questions.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 4 / 15; separate U.S. L3 0 / 0 / 8; separate U.S. L4 0 / 0 / 12. Typical assumes a short values/document review without a medical decision.

Completion evidence. Triggering request; participants and authority; source-attributed goals/values; documents reviewed; exact draft/final version; required qualified discussion/signatures/witnesses; unresolved questions; and distribution/storage status.

What does not prove completion. Template completion, AI draft, uploaded old document, signature image, caregiver account alone, billing code, or document delivery without beneficiary-led review.

Edge cases/open decisions. X-002, X-003, X-004, X-008, X-010, X-011, X-015, CP-E04, CP-E05, CP-E06, O-003, and O-011. State-specific document, witnessing, remote-execution, revocation, and storage policy remains open. Test capable beneficiary; fluctuating capacity; conflicting documents; multiple representatives; partial revision; interpreter; beneficiary declines to sign; and urgent medical question.

31. Address POLST when available and desired#

When. Only when a legally recognized POLST or analogous medical-order process is available for the beneficiary's current jurisdiction and clinical setting, and the beneficiary or authorized representative wants to consider it. It is not a default output of GUIDE enrollment.

What and how much. Explain the process using approved neutral material, verify the request and decision authority, arrange the required clinician discussion, and complete the jurisdiction-specific form/order/signatures only through an authorized clinician. Distinguish a POLST medical order from an advance directive, goals note, or unsigned draft.

Data. Direct: current beneficiary location; official jurisdiction form/rules; authenticated beneficiary/authorized-representative choices; clinician discussion/order/signature; and form status. Corroborating: prior POLST/advance directives and clinical context. Identity-only: form presence, diagnosis, frailty, code status field, or scanned signature does not establish current validity or desire.

Potential Proxi work. Identify the current jurisdiction/form, prefill verified demographics, present approved process information, compare candidate prior versions, route to the authorized clinician, and track signatures/distribution. AI cannot recommend order selections, determine capacity, create the medical order, or validate legal sufficiency.

Human role. L2 coordinates and confirms preference. A separate U.S. beneficiary-location-authorized L4 clinician conducts the medical discussion and signs/orders as state law permits. Legal/compliance resolves form applicability; Puerto Rico L2 never performs the clinical act.

Provisional Proxi human minutes (low / typical / high). Conditional card: L1 0 / 0 / 5 shared administrative exception; L2 0 / 0 / 8; separate U.S. L3 0 / 0 / 5; separate U.S. L4 0 / 0 / 20. Typical is zero when POLST is not requested/applicable.

Completion evidence. Jurisdiction/applicability; current request; decision authority; qualified discussion; exact form/version; clinician identity/authority; required signatures/date; effective status; distribution/access; and superseded/revoked prior form handling.

What does not prove completion. Blank/prefilled form, advance directive, code-status field, scanned old form, scheduled clinician visit, AI summary, or signature without an authenticated discussion/order.

Edge cases/open decisions. X-002, X-003, X-004, X-005, X-010, X-011, X-015, CP-E06, CC-E43, O-003, and O-011. Test jurisdiction without process; beneficiary traveling; representative conflict; old conflicting POLST; revocation; unsigned form; clinician not authorized; remote discussion; and emergency before completion.

32. Identify current clinical providers#

When. During packet assembly and beneficiary/caregiver confirmation, and whenever claims, attribution, referral, transition, or direct report indicates that a clinical relationship may have started, ended, or changed.

What and how much. Build a current provider list covering PCP, dementia clinician, behavioral-health provider, relevant specialists, and other treating clinicians. For each, retain provider identity, organization/location, role, contact endpoint, relationship status, last/next encounter as known, source, and unknown/disputed status. “Candidate provider” is not “current treating provider.”

Data. Direct: beneficiary/caregiver authenticated report of their relationship; verified provider/practice confirmation; current Participant/EHR care-team data; and authoritative attribution/assignment records for their exact meaning. Corroborating: Medicare/Medicaid/managed-care claims, encounter feeds, referrals, directories, HIE data, and prior notes. Identity-only: NPI, directory listing, claim, attribution candidate, prescription, or one encounter does not by itself establish a current PCP or ongoing relationship.

Potential Proxi work. Pull candidates from EHR, payer/Medicaid/Medicare, claims, HIE, referral, and directory sources; normalize provider identities; rank likely current relationships with transparent source dates; ask the beneficiary to verify; and flag duplicates/conflicts. AI may not promote a candidate to current provider or select a PCP.

Human role. L2 confirms relationship-sensitive provider facts with the beneficiary/caregiver. L1 may use the single shared 13-minute exception to verify office identity/endpoint or pursue a missing administrative fact. No clinician is required for clean provider identification.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / up to 13 shared across the assessment, not additive per task; L2 0 / 2 / 6; separate U.S. L3 0 / 0 / 0; separate U.S. L4 0 / 0 / 0.

Completion evidence. Provider-by-provider identity, role, relationship status, source/date, beneficiary verification, verified endpoint where needed, duplicate/conflict disposition, and explicit no-known-provider/unknown status.

What does not prove completion. Claim, NPI, prescription, directory record, historical note, payer attribution alone, portal connection, or AI-ranked provider.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-006, X-009, X-015, CC-E01, CC-E20, CC-E21, O-020, and O-021. Test no PCP; multiple PCP candidates; concierge/out-of-network clinician; provider left practice; stale claim; new provider without claim; same-name clinicians; beneficiary disputes payer attribution; and deceased/retired provider.

33. Identify current community and home-based services#

When. During each comprehensive assessment and whenever a payer, Medicaid HCBS, agency, community organization, caregiver, or beneficiary reports a service start, stop, waitlist, denial, or material change.

What and how much. Inventory each current or pending service/support: type, provider/program, payer/funder, frequency/authorized amount, actual recent delivery, start/end dates, contact, beneficiary experience, barriers, and status. Keep eligibility, authorization, referral, acceptance, scheduled start, and actual service receipt distinct.

Data. Direct: beneficiary/caregiver report; agency/program confirmation; service logs; current Medicaid/managed-care authorization for its native authorization fact; and claims for billed events. Corroborating: care-plan records, prior assessments, case-manager notes, and referrals. Identity-only: Medicaid eligibility, HCBS waiver enrollment, authorization, claim, resource listing, or agency contact does not alone establish current usable service delivery.

Potential Proxi work. Pull candidates from EHR, Medicaid/payer/HCBS, claims, care-management, referral, and agency sources; deduplicate without collapsing different services; ask for verification; compare authorization with recent delivery; and flag gaps/overlap. AI may not decide eligibility, duplication, responsibility, or service sufficiency.

Human role. L2 verifies practical receipt and beneficiary experience and owns unresolved service-context discussion. L1 may use the shared exception to confirm administrative service status. Clinical staff enters only when a service gap creates a named clinical/safety issue.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / up to 13 shared across the assessment, not additive per task; L2 0 / 2 / 8; separate U.S. L3 0 / 0 / 6; separate U.S. L4 0 / 0 / 4.

Completion evidence. Service-by-service type/provider/payer; eligibility/authorization/referral/receipt states; actual frequency/date; source; beneficiary confirmation; overlap/gap candidates; and unresolved ownership.

What does not prove completion. Dual eligibility, waiver status, authorization, claim, referral, scheduled intake, resource-list entry, or old care plan.

Edge cases/open decisions. X-003, X-004, X-005, X-006, X-009, X-015, CC-E17, CC-E26, CC-E27, CC-E28, CC-E29, CC-E30, O-022, and O-023. Test eligible-not-receiving; authorization but missed visits; cash/private service; duplicate payer records; paused service; agency closed; caregiver-provided support; waitlist; and conflicting case-manager report.

33A. Identify needed clinical coordination#

When. After Tasks 9-33 establish clinical findings, current providers, medications, transitions, and services, and before treating the assessment as clinically synthesized or handing it to care planning.

What and how much. Determine which clinical issues require information exchange, consultation, shared management, diagnostic follow-up, medication work, pending-result ownership, transition follow-up, or clarification among the GUIDE clinician and other treating providers. Produce issue-by-issue need, proposed recipient/question, urgency as determined by the qualified clinician, and accountable owner. This card identifies coordination need; it does not prove outreach or closed-loop completion.

Data. Direct: clinician-authored assessment findings and coordination determination; authenticated provider relationships; active orders/recommendations; and transition/pending-result records. Corroborating: claims, referrals, medication discrepancies, caregiver reports, and prior coordination. Identity-only: provider list, diagnosis, risk alert, shared EHR, or portal access does not establish that coordination is needed or complete.

Potential Proxi work. Compare assessed issues with current relationships, assemble source-linked candidate questions, identify missing recipients/records, and prepare separate coordination handoffs. AI may not decide clinical need, recipient suitability, urgency, or that existing access resolves the issue.

Human role. L2 clarifies nonclinical relationship/logistics and supports beneficiary choices. Separate U.S. L3 evaluates nursing/psychosocial coordination needs; L4 dementia-proficient clinician determines medical coordination needs and questions. Execution belongs to the Care Coordination SOP and is not duplicated here.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 2 / 5; separate U.S. L3 0 / 4 / 12; separate U.S. L4 0 / 3 / 12.

Completion evidence. Each issue/need; source findings; clinical determiner/authority; recipient/question; urgency/disposition; beneficiary preference when applicable; owner; and linked Care Coordination case or explicit no-coordination-needed conclusion.

What does not prove completion. Provider list, shared EHR, referral draft, inbox message, risk alert, appointment, packet sent, or AI-generated coordination suggestion.

Edge cases/open decisions. X-003, X-004, X-008, X-010, X-015, CC-E01, CC-E10, CC-E12, CC-E21, CC-E37, O-003, O-004, O-019, O-020, and O-021. Test no coordination needed; unknown PCP; multiple specialists; conflicting recommendations; urgent issue; pending result without owner; provider already on GUIDE team; beneficiary refusal; and outside record arriving after synthesis.

34. Assess caregiver ability and willingness#

When. When one or more caregivers are identified, during each comprehensive assessment and whenever assistance, health, availability, relationship, or willingness changes. Complete one source-attributed record per caregiver.

What and how much. For each relevant support task, capture what the caregiver currently does, can do, is willing to do, cannot do, will not do, or is unsure about; frequency/limits; training/support needed; and anticipated changes. Keep ability, willingness, availability, legal authority, beneficiary permission, and actual performance separate.

Data. Direct: each caregiver's own authenticated statements; beneficiary statements separately; and observed demonstration when applicable. Corroborating: service notes, prior assessments, schedules, and documented tasks. Identity-only: family relationship, co-residence, emergency-contact status, portal access, employment, or prior help does not establish current ability/willingness or authority.

Potential Proxi work. Ask task-specific questions, maintain separate caregiver records, compare reports and prior status, identify unsupported functions, and prepare training/support needs. AI may not choose whose account controls, infer willingness, assign duties, or determine capacity.

Human role. L2 conducts the relationship-sensitive nonclinical clarification. Caregivers decide what they are able and willing to provide. Separate U.S. clinical staff enters only when a caregiver's health, safety, clinical skill, or beneficiary risk requires qualified assessment.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 3 / 8; separate U.S. L3 0 / 0 / 8; separate U.S. L4 0 / 0 / 5.

Completion evidence. Caregiver identity; task-by-task ability, willingness, availability and limits; source/date; beneficiary perspective; conflicts/unknowns; training/support needs; and triggered clinical/safety follow-up.

What does not prove completion. Family role, shared home, prior caregiving, signed portal proxy, one global “willing,” beneficiary assumption, or AI inference.

Edge cases/open decisions. X-002, X-003, X-004, X-005, X-008, X-009, X-015, CP-E04, CP-E05, CP-E07, O-011, and O-013. Test multiple caregivers; unwilling caregiver; limited availability; caregiver/beneficiary disagreement; representative overreach; caregiver health issue; paid worker misclassified as caregiver; and later withdrawal.

35. Assess caregiver knowledge, needs, and supports#

When. For each identified caregiver after Task 34, and whenever the care tasks, dementia stage, setting, caregiver role, or available supports materially change.

What and how much. Capture what the caregiver understands and wants help with across dementia, communication/behavior, daily care, safety, medication routines, transitions, services, self-care, and future planning as applicable; preferred learning mode/language; existing family/social/community supports; and priority needs. Assessment of a need is not delivery or demonstrated mastery of education.

Data. Direct: caregiver statements and any observed teach-back/demonstration under an approved method. Corroborating: prior training records, service participation, documented questions, and beneficiary needs. Identity-only: course completion, video view, relationship, years caregiving, credential, or material delivery does not establish current knowledge, need, support, or skill.

Potential Proxi work. Administer structured questions, map stated needs to approved content/support categories, identify accessibility needs, preserve preferences, and prepare an education/support plan. AI may not certify competence, select individualized clinical instruction, or claim education was furnished.

Human role. L2 clarifies needs and priorities and later provides/facilitates approved nonclinical education/support. Separate U.S. clinical staff enters only for individualized clinical instruction or a clinical/safety concern.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 3 / 10; separate U.S. L3 0 / 0 / 8; separate U.S. L4 0 / 0 / 5.

Completion evidence. Caregiver-specific domain responses; stated priorities; current supports; preferred language/mode; source/date; gaps/unknowns; proposed education/support handoff; and any qualified clinical route.

What does not prove completion. Content sent/viewed, course certificate, prior caregiving duration, navigator assumption, generic needs list, or AI-generated curriculum.

Edge cases/open decisions. X-003, X-004, X-005, X-007, X-009, X-015, O-036, and O-043. Test multiple caregivers with different needs; low literacy; language barrier; caregiver declines; prior training but failed teach-back; high clinical-skill need; caregiver has professional credential; and changing setting.

36. Assess caregiver well-being and strain#

When. For each identified caregiver during the comprehensive assessment, in a setting that permits honest responses, and whenever caregiver role, health, burden, availability, or safety materially changes.

What and how much. Administer the approved strain/well-being questions or instrument; capture physical and emotional health, stress/burden, sleep, isolation, work/financial impact, conflict, safety, and desired help as included by policy. Preserve item responses and limitations. A score does not diagnose a caregiver or prove inability/unwillingness.

Data. Direct: caregiver item responses and deterministic score; the qualified professional's review establishes a clinical/safety disposition. Corroborating: prior screens, service use, observed distress, and beneficiary/caregiver reports kept separate. Identity-only: hours of care, relationship, diagnosis, leave claim, or model sentiment does not establish current strain or safety.

Potential Proxi work. Present private accessible questions, calculate configured scores, compare prior results, surface concerning items, and prepare approved support options. AI may not diagnose, infer caregiver capacity, or choose clinical/safety disposition.

Human role. L2 conducts the nonclinical needs conversation and support preference clarification. Separate U.S. L3 LCSW/behavioral/RN staff reviews the typical screen and any clinical or safety concern; L4 enters only when medical evaluation/treatment is needed.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 3 / 10; separate U.S. L3 0 / 2 / 12; separate U.S. L4 0 / 0 / 8.

Completion evidence. Caregiver/respondent; instrument/version and items; score/limitations; privacy/accommodations; stated support needs; qualified review/disposition when triggered; and linked caregiver-service or safety work.

What does not prove completion. Hours of care, global burden label, content delivery, score without items, AI sentiment, beneficiary report alone, or referral without review.

Edge cases/open decisions. X-002, X-003, X-004, X-007, X-008, X-009, X-015, O-003, O-004, and O-042. Test caregiver cannot answer privately; multiple caregivers; positive distress/self-harm response; beneficiary access to sensitive caregiver answers; caregiver declines; conflicting reports; language barrier; and late severe disclosure.

37. Try to identify a caregiver when none is present#

When. Only after Task 4 establishes that no caregiver is currently identified, not merely that caregiver information is missing or one caregiver is unavailable.

What and how much. With the beneficiary, identify permitted potential supporters; establish permission for each outreach; make the policy-defined reasonable attempts; explain the voluntary role; and record whether anyone is able and willing. Do not invent a caregiver, coerce participation, disclose PHI without authority, or let an unbounded search delay beneficiary services.

Data. Direct: beneficiary-identified candidates and outreach permission; potential caregiver's own willingness; and established representative authority if relevant. Corroborating: permitted contact records and prior involvement. Identity-only: emergency contact, family relationship, shared address, portal proxy, or past assistance does not establish current caregiver status, willingness, or disclosure permission.

Potential Proxi work. Present known permitted contacts for beneficiary confirmation, prepare approved outreach, track attempts/channels/outcomes, and stop at the approved boundary. AI may not select a caregiver, infer permission, or treat nonresponse as refusal or acceptance.

Human role. L2 conducts the reasonable relationship-based effort. L1 may use the single shared exception for beneficiary-authorized administrative outreach. The potential caregiver decides. Software enforces the approved authority and disclosure boundary; Healthcare Legal Counsel enters only for a remaining legal-authority question.

Provisional Proxi human minutes (low / typical / high). Conditional card: L1 0 / 0 / up to 13 shared across the assessment, not additive per task; L2 0 / 0 / 12; separate U.S. L3 0 / 0 / 0; separate U.S. L4 0 / 0 / 0.

Completion evidence. Confirmed no-current-caregiver trigger; beneficiary capability/participation; each candidate and permission basis; exact attempts/outcomes; willing caregiver confirmation or truthful exhaustion; and handoff to Task 38 if none is identified.

What does not prove completion. Emergency-contact list, one unanswered call, family relationship, caregiver portal account, AI-selected candidate, outreach draft, or administrative closure without the defined reasonable-effort evidence.

Edge cases/open decisions. X-002, X-003, X-006, X-007, X-010, X-014, X-015, CP-E07, O-004, O-011, and O-013. The reasonable-effort denominator, channel/attempt limits, and beneficiary-capability route remain open. Test capable beneficiary declines search; candidate refuses; no contacts; multiple candidates; wrong number; late acceptance; unsafe family contact; and unknown versus confirmed no caregiver.

38. Follow the no-caregiver branch and plan added safeguards#

When. After Task 37 truthfully completes without an able and willing caregiver, or when a previously identified caregiver withdraws and no replacement is established. Do not apply this branch while caregiver status is merely unknown.

What and how much. Mark caregiver-dependent assessment/education/support work not applicable for the current state, identify each function that lacks support, preserve beneficiary-led work, and create an individualized package of additional assistance/safeguards sufficient for qualified review. The package may address contact, daily tasks, medication support, safety, services, emergency access, residence, and monitoring, but no software or navigator determines sufficiency.

Data. Direct: completed caregiver-search evidence; beneficiary preferences; assessment findings; current services; and authorized clinical/safeguarding determinations. Corroborating: home observations, utilization, prior supports, and service availability. Identity-only: living alone, no emergency contact, residence type, technology, service referral, or monitoring-device enrollment does not prove no caregiver or adequate safeguards.

Potential Proxi work. Compile unsupported functions, match approved service/safeguard options to confirmed constraints, disclose inventory gaps, prepare tasks, track execution, and monitor for missing evidence. AI may not decide adequacy, placement, risk acceptance, or beneficiary capacity.

Human role. L2 leads nonclinical planning with the beneficiary. Separate U.S. L3/L4 staff evaluates clinical, functional, behavioral, or safety sufficiency within scope. External service providers furnish actual support. Puerto Rico does not perform the clinical review.

Provisional Proxi human minutes (low / typical / high). Conditional card: L1 0 / 0 / up to 13 shared across the assessment, not additive per task; L2 0 / 0 / 18; separate U.S. L3 0 / 0 / 15; separate U.S. L4 0 / 0 / 12. Ongoing service time belongs to the corresponding service SOP, not this assessment card.

Completion evidence. No-caregiver/search result; unsupported-function inventory; beneficiary preferences/authority; proposed and approved safeguards; clinical/safety reviewers and scope; implementation/connection evidence; unresolved gaps; and monitoring/reevaluation owner.

What does not prove completion. Living-alone flag, device order, referral sent, resource list, generic safety plan, one check-in scheduled, clinician signature without function-by-function review, or AI statement that safeguards are adequate.

Edge cases/open decisions. X-002, X-003, X-004, X-006, X-008, X-010, X-014, X-015, CP-E07, CC-E07, CC-E26, O-003, O-004, O-011, and O-013. Test capable beneficiary with no caregiver; uncertain capacity; services unavailable; technology failure; new caregiver later; beneficiary refuses a safeguard; temporary caregiver absence; emerging risk; and safeguard service accepted but never starts.

39. Determine the home-visit obligation#

When. After the initial assessment has enough validated tier facts to apply the current public home-visit rule, and before the assessment/PAAF deadline can expire. Re-evaluate if tier, residence, or visit completion evidence changes.

What and how much. Determine one status: in-person current-residence visit required for the applicable moderate/high tier; visit optional and in-person or remote for the applicable low tier; already satisfied by a qualifying visit; not yet determinable; or missed/at risk. Calculate the current two-month/60-day operational target without treating an unverified tier or scheduled visit as satisfaction.

Data. Direct: qualified stage/assessment findings; current caregiver status/burden and residence facts; the current CMS tier result or authoritative inputs; prior qualifying visit evidence; and current assessment/staging date. Corroborating: scheduling, claims, address, prior tier, and visit notes. Identity-only: diagnosis, old tier, residence address, calendar event, telehealth connection, or claim does not establish current obligation or completion.

Potential Proxi work. Apply the published rule to verified facts, calculate the deadline, detect missing/conflicting inputs, separate required/optional modality, open scheduling work, and flag risk/miss. AI may not infer tier, waive a required visit, or approve an exception.

Human role. L2 confirms practical residence and beneficiary preference and reviews the typical obligation output. L1 may use the shared exception for scheduling facts. The applicable U.S. clinician resolves only the clinical or tier fact within authority. Software holds an unsupported policy pattern as a rule gap; Healthcare Legal Counsel enters only for an unresolved legal-authority question, and a Participant-specific requirement remains design-blocked under O-PA-001 until the controlling agreement names the authorized act. No generic program/compliance reviewer is added.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / up to 13 shared across the assessment, not additive per task; L2 0 / 1 / 5; separate U.S. L3 0 / 0 / 5; separate U.S. L4 0 / 0 / 5.

Completion evidence. Assessment/staging date; current tier inputs/result/source; residence; rule/requirement outcome; required modality; due date/calculation; prior qualifying visit if any; scheduling owner; and unresolved/missed status.

What does not prove completion. Prior tier, diagnosis, calendar booking, claim, address, remote visit link, home-care encounter, or AI classification.

Edge cases/open decisions. X-003, X-004, X-005, X-010, X-012, X-015, CP-E16, O-004, O-018, O-024, and O-PA-001. Test low dyad; low individual; moderate/high dyad; moderate/high individual; tier unknown; staging date corrected; move before visit; visit already occurred; day-60 boundary; and missed deadline.

40. Conduct the residence visit#

When. When Task 39 establishes that the visit is required or the beneficiary elects an optional visit. Use the current residence and required in-person modality; use remote only when the current public rule permits it.

What and how much. Complete one structured residence encounter with the beneficiary, enabling caregiver or other care-team participation as useful and permitted. Confirm actual location, modality, participants, start/end, accessibility, and the observation domains assigned to Tasks 41-42. This card proves the encounter occurred; it does not by itself prove every observation, clinical conclusion, or downstream action.

Data. Direct: authenticated encounter record, actual current-residence location, participant attendance, contemporaneous visitor observations, and connection evidence for a permitted remote visit. Corroborating: scheduling, travel/arrival, caregiver report, and claim. Identity-only: calendar event, GPS point, telehealth link, claim, photo, or unsigned note does not establish a completed qualifying visit.

Potential Proxi work. Schedule, confirm location/modality, support remote connection, present the structured guide, time-stamp and source notes, capture consented media, and identify incomplete domains. AI cannot substitute for the human visit or attest that it occurred.

Human role. L2 may conduct the human visit and collect structured nonclinical observations. If the assigned visit includes licensed assessment, a separate U.S. L3/L4 professional performs only those acts within beneficiary-location authority. These roles are alternatives/additions based on scope, not interchangeable; Puerto Rico never performs clinical assessment.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 10 for a connection/scheduling exception; L2 20 / 30 / 60 encounter minutes; separate U.S. L3 0 / 0 / 30; separate U.S. L4 0 / 0 / 15. Travel and external participant time are excluded. Do not add L3/L4 unless the visit actually includes an authorized clinical act.

Completion evidence. Obligation/election; date/start/end; current residence; modality; participants and performing organization; accommodations; human visitor identity; completed/declined observation domains; interruptions; and links to Tasks 41-42.

What does not prove completion. Booking, reminder, check-in ping, phone call when in-person was required, claim, geolocation alone, photo upload, AI interview, or visitor signature without encounter facts.

Edge cases/open decisions. X-001, X-003, X-007, X-010, X-012, X-013, X-014, X-015, CC-E43, O-003, O-004, and O-PA-001. Test wrong address; beneficiary moved; no answer; required visit attempted remotely; connection failure; unsafe entry; caregiver attends remotely; interrupted visit; visitor lacks clinical authority; and partial completion before deadline.

41. Assess home safety, navigation, and function in the residence#

When. During the completed Task 40 encounter while the beneficiary is in the actual current residence, with limitations recorded if a permitted remote modality prevents direct observation.

What and how much. Observe the approved home-safety domains, how the beneficiary navigates key spaces, and relevant ADL/IADL performance naturally or through safe approved observation. Record exactly what was observed, what was reported, what could not be assessed, assistance/cueing, environmental context, and discrepancies with Tasks 12, 13, and 23. Do not stage a test or ask the person to perform an unsafe act.

Data. Direct: contemporaneous visitor observation in its actual context; beneficiary/caregiver reports kept separately; configured instrument responses; and qualified clinical interpretation. Corroborating: consented media, prior visits, therapy/nursing notes, and equipment use. Identity-only: home type, DME, service claim, photograph, diagnosis, or old visit does not establish current function or safety.

Potential Proxi work. Prompt the structured observation sequence, time/source each finding, compare observed and reported function, identify unassessed items/conflicts, and prepare clinical review. AI may not infer unobserved independence, diagnose impairment, or determine safeguard sufficiency.

Human role. L2 records nonclinical direct observations and practical context. Separate U.S. L3 interprets the typical functional/safety significance within scope and routes clinical concerns; L4 enters for medical diagnosis/treatment. L2 observations remain evidence, not clinical conclusions.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 5 / 15 incremental observation documentation; separate U.S. L3 0 / 5 / 15; separate U.S. L4 0 / 0 / 10. Task 40 encounter time is separate; avoid counting the same minutes twice when roles document concurrently.

Completion evidence. Visit/date/context; domain-level observed/reported/unassessed status; source/observer; assistance/cueing; limitations; discrepancies; qualified interpretation where triggered; and linked safety/service/care-plan actions.

What does not prove completion. Checklist without observations, photograph, beneficiary/caregiver report alone, DME, diagnosis, old note, AI inference from video, or visitor statement that the home is “safe.”

Edge cases/open decisions. X-003, X-004, X-008, X-009, X-010, X-015, O-003, O-004, and O-013. Test remote visual limitation; beneficiary refuses a room/task; temporary clutter; unsafe request to demonstrate; observation/report conflict; caregiver cues throughout; fluctuating function; immediate hazard; and late-corrected visit note.

42. Assess other home-context factors#

When. During Task 40 after the core safety/navigation/function observations, and whenever the residence context changes materially.

What and how much. Capture environmental, social, and behavioral factors affecting beneficiary or caregiver function: household composition, routines, noise/privacy, cultural practices, pets, landlord/building constraints, food and utilities as visible/reported, social support/isolation, caregiver dynamics, technology/connectivity, emergency access, and other relevant context. Attribute observations and reports; avoid diagnostic, legal, or moral conclusions.

Data. Direct: contemporaneous visitor observation and each participant's source-attributed report. Corroborating: HRSN results, service records, prior visits, and permitted media. Identity-only: ZIP code, residence type, household roster, device ownership, claim, or photograph does not establish the factor's current effect or risk.

Potential Proxi work. Prompt relevant domains, organize observation versus report, compare with prior/HRSN facts, flag conflict or missing context, and prepare follow-up categories. AI may not infer relationships, abuse, capacity, clinical cause, or adequacy of the environment.

Human role. L2 conducts the nonclinical contextual conversation and observation. Separate U.S. L3 reviews behavioral, psychosocial, functional, or safety significance when triggered; L4 enters only for medical interpretation.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 5 / 12; separate U.S. L3 0 / 2 / 8; separate U.S. L4 0 / 0 / 5.

Completion evidence. Domain/source-level context; observer/speaker/date; relevance stated without unsupported conclusion; unknown/declined/unobserved items; material conflicts; qualified review for triggered concern; and follow-up links.

What does not prove completion. ZIP-code inference, household roster, photo, generic social-risk score, residence label, device presence, or AI-generated household narrative.

Edge cases/open decisions. X-002, X-003, X-004, X-008, X-009, X-015, O-003, O-011, O-013, and O-042. Test crowded home; caregiver conflict; monitored communications; temporary guest; cultural practice misread as hazard; unsafe pet; no connectivity; social isolation; possible abuse; and beneficiary asks to restrict caregiver-facing content.

43. Synthesize findings across domains#

When. After every applicable assessment domain is completed, explicitly unavailable/declined, or assigned an unresolved owner; repeat when a material late correction arrives before finalization. Do not synthesize from silently missing domains.

What and how much. Integrate history, examination, MDM, cognition/stage, ADL/IADL, medication reconciliation, behavioral-health screens, safety, HRSN, advance-planning status, providers/services, caregiver assessment, and home findings into one clinically coherent and service-usable assessment. Resolve or explicitly preserve conflicts, limitations, uncertainty, clinical findings, nonclinical needs, risks, strengths, priorities, and further work. Also identify, for Ongoing Monitoring and Support, the current attributable baseline, explicit unknowns, watch items, promised results, and next natural observation; this is a context handoff until Ongoing furnishes its own service. Synthesis is not arithmetic aggregation or AI summarization.

Data. Direct: the qualified clinicians' authored interpretations within scope; source-attributed domain results; prescribing-authority medication reconciliation; visitor observations; caregiver/beneficiary reports; and authoritative external findings. Corroborating: claims, prior assessments, AI drafts, and timelines. Identity-only: completion count, score dashboard, diagnosis list, signed template, or AI narrative does not establish a clinically synthesized assessment.

Potential Proxi work. Compile a domain manifest, generate source-linked draft sections, display conflicts/missing/stale facts, compare prior findings, prevent unsupported carry-forward, and capture clinician edits/attestation. AI may not choose which evidence controls, diagnose, assign risk/tier, or attest completeness.

Human role. Separate U.S. L3 RN/LCSW/behavioral staff synthesizes nursing/psychosocial/function/safety findings within scope. The separate U.S. L4 dementia-proficient E/M clinician performs and owns the integrated clinical synthesis, diagnoses/medical judgments, and attestation. Puerto Rico L2 supplies nonclinical facts but does not synthesize clinically.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 0 / 0 / 0 incremental; separate U.S. L3 5 / 10 / 20; separate U.S. L4 25 / 35 / 60. These are the main synthesis minutes in the assessment envelope, not a duplicate full re-review of each earlier card.

Completion evidence. Applicable-domain manifest; source-linked findings; explicit missing/declined/invalid/conflicted facts; L3 and L4 authorship/authority; integrated clinical and nonclinical conclusions; uncertainty; actions/referrals/coordination; and signed/attested final assessment identity.

What does not prove completion. All checkboxes filled, score dashboard, AI summary, copied prior assessment, clinician signature without integrated reasoning, diagnosis list, PAAF draft, or care-plan handoff.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-008, X-009, X-010, X-011, X-014, X-015, MED-E34, O-003, O-004, O-018, O-028, and O-PA-001. Test missing domain; declined item; conflicting reports; invalid stage; medication incomplete; positive safety screen; no caregiver; late correction; duplicate source; and clinician outside beneficiary-location authority.

44. Arrange additional diagnostic evaluation when needed#

When. As soon as the authorized clinician concludes that the available evidence does not support a sufficiently clear or safe clinical conclusion or that additional evaluation is otherwise clinically indicated; do not wait for final Task 43 synthesis merely to open the route. This is an enhanced clinical route, not a universal public GUIDE requirement.

What and how much. Record the unresolved clinical question, explain the uncertainty and purpose, identify the appropriate evaluation category/recipient, obtain beneficiary choice and required authorization, release the approved packet, and track receipt/appointment/result ownership. When dementia attestation remains unsupported, hold the PAAF, alignment, and ongoing GUIDE-service activation while diagnosis-supporting, safety, and otherwise-authorized work continues under AS-E02. Arrangement does not equal completed evaluation or diagnostic resolution.

Data. Direct: clinician-authored uncertainty/need and referral question; beneficiary choice; verified recipient; authorized release; delivery/receipt; encounter/result events. Corroborating: assessment conflicts, prior evaluation, claims, and scheduling. Identity-only: abnormal score, AI suggestion, directory match, referral draft, appointment, or claim does not establish clinical need, recipient fit, or completion.

Potential Proxi work. Assemble unresolved questions and supporting evidence, identify candidate destinations from current facts, prepare the exact referral, support beneficiary choice, verify endpoints, and track each step. AI may not decide that diagnostic evaluation is needed, select clinical specialty/urgency, or release PHI.

Human role. Separate U.S. L4 dementia-proficient clinician determines need/question/urgency. L2 coordinates beneficiary choice and practical follow-through. L1 may use the shared administrative exception for verified endpoint, scheduling, or records pursuit. External specialists perform the evaluation.

Provisional Proxi human minutes (low / typical / high). Conditional card: L1 0 / 0 / up to 13 shared across the assessment, not additive per task; L2 0 / 0 / 8; separate U.S. L3 0 / 0 / 8; separate U.S. L4 0 / 0 / 15. Downstream specialist time and Care Coordination follow-through are excluded.

Completion evidence. Clinical need/question and clinician authority; uncertainty explained; beneficiary choice; verified recipient; approved packet/release; receipt; appointment/status; result-review owner; and linked Care Coordination case.

What does not prove completion. Abnormal screen, AI recommendation, directory result, referral draft, transmission, scheduled appointment, claim, or specialist encounter without returned/reviewed findings.

Edge cases/open decisions. X-003, X-004, X-006, X-007, X-010, X-014, X-015, CC-E01, CC-E05, CC-E06, CC-E09, CC-E37, O-003, O-004, O-019, and O-021. Test uncertain diagnosis; beneficiary declines; no available specialist; wrong endpoint; referral rejected; waitlist; no result owner; urgent deterioration; and late result after assessment completion.

45. Discuss findings and immediate next steps#

When. After Task 43 produces clinician-approved findings and before the assessment is treated as understood and ready for care planning; provide an earlier limited discussion when immediate safety/action cannot wait.

What and how much. Explain confirmed findings, material uncertainty, limitations, medication conclusions, caregiver/home/social findings, additional evaluation, immediate safety/support actions, and next ownership in accessible language. Check understanding through meaningful response/teach-back appropriate to cognition and preference. Keep clinical explanation, nonclinical service navigation, and beneficiary/caregiver choices distinct. When the discussion creates an agreed observation, promised result, or unanswered current question, send that exact item and its return condition to Ongoing Monitoring and Support; do not turn it into a generic check-in.

Data. Direct: exact clinician-approved findings; the acting professional's explanation; recipient responses/teach-back; and authenticated choices/questions. Corroborating: content delivery/access records and caregiver observations. Identity-only: message delivery, portal view, signature, “verbalized understanding,” yes/no assent, or AI summary does not establish comprehension or informed choice.

Potential Proxi work. Generate an accessible source-linked draft from approved findings, translate/render approved content, organize questions, capture teach-back verbatim, and create follow-up work. AI may not decide what clinical content to communicate, answer new clinical questions, or declare understanding.

Human role. Separate U.S. L4 explains cognition, diagnosis, medical, medication, and clinical uncertainty; separate U.S. L3 explains nursing/behavioral/psychosocial findings within scope. L2 explains approved services, logistics, supports, and next steps and authenticates preference-sensitive questions. Beneficiary/caregiver time is excluded.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 0; L2 3 / 5 / 12; separate U.S. L3 0 / 3 / 10; separate U.S. L4 10 / 15 / 30.

Completion evidence. Approved findings/version; speakers and scope; recipients/authority; language/accommodations; exact materials; questions/answers; teach-back or supported alternative; choices/refusals; unresolved misunderstanding; and next owners/times.

What does not prove completion. Summary generated/sent/viewed, signature, yes/no “understands,” caregiver presence, clinician note without recipient response, or AI chatbot interaction without approved clinical ownership.

Edge cases/open decisions. X-002, X-003, X-004, X-007, X-008, X-009, X-014, X-015, CC-E39, MED-E41, O-003, O-004, O-011, and O-043. Test limited cognition; interpreter; hearing/vision barrier; beneficiary/caregiver disagreement; sensitive caregiver information; new clinical question; failed teach-back; refusal; and urgent finding before full synthesis.

46. Deliver assessment findings to care planning#

When. After Task 43 is attested and Task 45 has recorded the applicable discussion, while preserving unresolved/urgent items and any additional evaluation rather than waiting indefinitely for perfect information.

What and how much. Transfer the applicable confirmed findings, strengths, goals/preferences, caregiver needs, services, home context, medication status, safety issues, clinical restrictions, unresolved facts, and owned follow-up into the Care Plan intake. Preserve source/author/date and clinical versus reported status. Before final CMS alignment, this handoff supports only CMS-permitted internal care planning; it does not prove that an ongoing GUIDE Care Plan service was furnished. Handoff does not authorize the navigator to reinterpret clinical findings or prove a care plan was built.

Data. Direct: the exact attested assessment version and its source-linked findings; authenticated beneficiary/caregiver choices; and active follow-up obligations. Corroborating: generated summary and delivery/access evidence. Identity-only: file attachment, API success, inbox notification, document count, or copied text does not establish correct, complete, usable handoff.

Potential Proxi work. Compile the required care-planning input, bind it to the exact assessment version, identify missing/applicability states, preserve restrictions and owners, transmit to the assigned internal destination, and reconcile acknowledgement. AI may not reinterpret findings, remove conflicts, select plan actions, or mark care planning complete.

Human role. L2 verifies that the nonclinical care-planning handoff is understandable and routed to the assigned navigator. Clinical authors remain responsible for their findings; no routine repeat L3/L4 review is added. A source correction routes back to the author rather than being edited by L2.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / 5 for an internal routing exception; L2 1 / 3 / 8; separate U.S. L3 0 / 0 / 5; separate U.S. L4 0 / 0 / 5 only for a material source correction.

Completion evidence. Exact assessment/version; applicable input manifest; source/author links; unresolved and urgent items; receiving navigator/destination; matched acknowledgement; correction restrictions; and linked Care Plan case.

What does not prove completion. Generated summary, file upload, API success, inbox delivery, navigator assignment, copied clinical text without provenance, or existence of a later care-plan draft.

Edge cases/open decisions. X-001, X-003, X-004, X-007, X-011, X-013, X-015, CP-E01, CP-E02, CP-E08, CP-E12, O-003, O-010, and O-015. Test wrong beneficiary; wrong assessment version; missing domain; unresolved urgent item; corrected finding after handoff; duplicate handoff; failed acknowledgement; and navigator attempts to alter clinical text.

47. Document completion and plan the next assessment#

When. After every applicable detailed card has objective evidence or an explicit unresolved/declined/not-applicable disposition and Task 43 is attested. Re-evaluate the schedule when CMS accepts a correction or a qualifying residence, dementia-severity, caregiver-status/burden, or clinical-change signal arrives.

What and how much. Record the final assessment identity, performed dates/actors, applicable-domain manifest, unresolved work, actual PAAF/submission status, real-time preliminary result, final Beneficiary Alignment Report result, and next annual window. Keep preliminary planning permission separate from final service activation and billing. For annual work, preserve the day-306 to day-425 window and the earlier of 60 days after annual-assessment start or day 425. Preserve current separate reassessment cadences: residence-type changes may be submitted every 30 days; dementia-severity or caregiver-status/burden changes every 180 days. A reassessment never satisfies the annual assessment, and clinical care is not delayed to fit a CMS submission cadence. Send Ongoing Monitoring and Support the accepted baseline, explicit unknowns, watch items, open return conditions, and next Assessment horizon. Keep the Assessment-owned annual/reassessment due list distinct from the Ongoing contact calendar.

Data. Direct: exact final attested assessment/version; domain completion evidence; staging/assessment dates; PAAF/submission/acknowledgement events; prior qualifying comprehensive-assessment date; and qualified determination of additional clinical need. Corroborating: calendar, claim, task status, and alerts. Identity-only: signed note, claim, due-date reminder, PAAF draft, or task closure does not establish complete/timely assessment or CMS receipt.

Potential Proxi work. Validate the evidence manifest, calculate the current annual and reassessment windows, prepare/store the record, submit only through the authorized pathway, reconcile the exact acknowledgement, schedule reminders, and surface change signals for qualified review. AI may not declare completion, choose earlier clinical reassessment, infer CMS acceptance, or convert a missed deadline to timely.

Human role. L2 verifies the nonclinical record and next-contact explanation. L1 may use the shared exception for administrative submission or acknowledgement pursuit. The applicable U.S. clinician determines clinical reassessment need. A nonclinical rule conflict is held and routed to the existing domain owner—Medicaid operations for a Medicaid-program rule or Healthcare Legal Counsel for unresolved legal authority—rather than to a generic program/compliance role.

Provisional Proxi human minutes (low / typical / high). L1 0 / 0 / up to 13 shared across the assessment, not additive per task; L2 0 / 1 / 4; separate U.S. L3 0 / 0 / 5; separate U.S. L4 0 / 0 / 5.

Completion evidence. Final assessment/version and attestation; domain manifest; performed actors/dates/authority; unresolved items/owners; exact PAAF submission and receipt; matched preliminary notification; matched final BAR with aligned/ineligible status, tier, and effective date; activation or non-alignment disposition; prior assessment anchor; next annual start/end/submission dates; reassessment-cadence state; and compliance disposition for any miss.

What does not prove completion. Signed note alone, claim, closed tasks, calendar reminder, PAAF draft, transmission without matched acknowledgement, preliminary alignment without the final BAR, one reassessment, or AI statement that all domains are complete.

Edge cases/open decisions. X-001, X-003, X-004, X-005, X-010, X-011, X-012, X-013, X-014, X-015, CP-E16, O-003, O-004, O-018, and O-PA-001. Test annual days 305/306/425/426; start plus 60 days; residence day 30; stage/caregiver day 179/180; reassessment before annual; duplicate/corrected submission; rejected/mismatched acknowledgement; late clinical finding; move/death; and missed deadline with corrective work.

Cross-pillar baseline and return loop#

This SOP participates in cross-pillar episodes under 25_GUIDE_Eight_Pillar_Service_Integration.md. Use the shared no-repeat packet, preparation-before-judgment route, receiving-owner acceptance rule, and count-once labor rule. Assessment receives service attribution only for a current attributable comprehensive assessment or focused reassessment in which every applicable item has a truthful disposition and every protected contribution is made by the authorized professional. Assessment and Care Plan are independently closable services joined by a source-preserving handoff; neither closes the other.

Assessment covers the initial comprehensive assessment, required annual comprehensive assessment, and bounded reassessment of affected domains. It is the baseline source, not the permanent owner of every downstream action. A later service result returns through Ongoing Monitoring and Care Plan; only a defined material change, correction, or stale/conflicting baseline question reopens the affected Assessment domains.