Skip to content

Benefits Discovery

GiveCare maintains its caregiver benefits registry in gc-benefits, then projects reviewed subsets into public wiki pages, the web benefits browser, and the SMS runtime bundle.1 Programs remain the caregiver-facing unit, while authorities, agencies, fulfillment channels, and claim evidence explain who funds a benefit, who runs it, how a caregiver applies, and how current each claim is. Mira can retrieve cautious candidates when the caregiver's current need and stated location support a lookup.

The problem

Over $60 billion in benefits goes unclaimed annually2. Participation in safety net programs ranges 40-60%, with eligible non-participation rates of 16-72% depending on the program.

The gap is not a lack of programs. It is a lack of discovery, navigation, and follow-through. Most caregivers do not know they qualify. The application processes are fragmented, documentation requirements are unclear, and deadlines are easy to miss. See Benefits Landscape for the full landscape analysis.

How it works

flowchart TD
    A["Current need + caregiver-stated state"] --> B["Search the reviewed runtime bundle"]
    B --> C["Filter by jurisdiction and relevance"]
    C --> D["Return candidates worth checking"]
    D --> E["Cite the source and preserve uncertainty"]
    E --> F["Offer one useful next move"]
  1. Need and location: Mira uses the current need, bounded known facts, and a caregiver-stated state. Without a state, it asks for that missing fact.
  2. Candidate retrieval: The runtime ranks reviewed records and filters out state-specific programs that do not cover the stated jurisdiction.
  3. Uncertainty: A result is a program worth checking, not an eligibility, approval, payment, or legal determination.
  4. Grounded reply: Any named program or application detail must come from the tool evidence for that turn and carry its source reference.
  5. Next move: Mira offers one proportionate next action. A later check-in is scheduled only with explicit caregiver consent; the runtime does not claim to track an external application's status.

Program coverage

Federal programs

Program Zone(s) Focus
NFCSP (National Family Caregiver Support Program) GC1, GC4 Respite, counseling, training, supplemental services
VA PCAFC (Program of Comprehensive Assistance for Family Caregivers) GC4, GC2 Stipend, health insurance, respite for veteran caregivers
VA PGCSS (Program of General Caregiver Support Services) GC1, GC5 Peer support, education, navigation for veteran caregivers
FMLA Military GC4, GC5 26 weeks unpaid leave for military caregiver employees
Medicaid HCBS (Home and Community-Based Services) GC3, GC4 Home modifications, personal care, adult day services
SSI/SSDI GC4 Income support for disabled care recipients
Lifespan Respite GC1, GC2 Respite care coordination and funding
Medicare GUIDE GC5, GC1 Dementia care navigation and support
SNAP GC4 Nutrition assistance
LIHEAP GC4, GC3 Energy bill assistance
Medicaid GC4, GC2 Health coverage for low-income caregivers

State-level programs

Category States/Jurisdictions Zone(s)
Mandatory paid family leave 14 jurisdictions GC4
Structured Family Caregiving 11 states GC4, GC1
Caregiver tax credits 8 states GC4
7 state-specific implementations Varies Multiple

Technical architecture

Registry and eligibility rules

The canonical benefit data lives in ../gc-benefits/data/programs/*.json and validates through Pydantic models in ../gc-benefits/src/benefit_engine/models.py.1 Program records carry machine-readable eligibility, coverage, verification, and freshness fields. The registry strategy adds optional edges from a program to:

  • an authority such as a statute, waiver, grant, tax provision, or insurance line
  • an agency that funds, administers, or serves as the front door
  • a fulfillment channel for the program-by-jurisdiction transaction
  • claim-level evidence for load-bearing facts such as current official name, family-payment eligibility, front door, and application URL

ETL pipeline

flowchart LR
    A["Canonical gc-benefits JSON"] --> B["Validate + evidence audit"]
    B --> C["Generate wiki benefit pages"]
    C --> D["Validate wiki docs in memory"]
    D --> E["Publish wiki runtime contracts"]
    B --> F["Export benefits runtime bundle"]
    F --> G["Mira benefits screening"]
    E --> G
  • Discover / ingest: source candidates are queued, scraped when allowed, and loaded into staged records
  • Validate / audit: program JSON is checked for schema validity, source coverage, evidence gaps, freshness, and readiness
  • Generate: validated corpus records refresh wiki-owned benefit pages
  • Publish: wiki frontmatter is validated in memory and projected into the owner-side resource, strategy, and guidance-card contracts; the benefits runtime bundle is exported directly from gc-benefits

Impact evidence

  • 74% of caregivers report that services enabled them to provide care longer3
  • 62% indicated that without services, the care recipient would be in a nursing home3
  • Average caregiver out-of-pocket cost: $7,242/year4 — benefits discovery directly reduces this burden

Code reference

Current SMS benefits lookup lives in ../gc-sms/knowledge/catalog.ts plus the ../gc-sms/knowledge/toolAdapters.ts and ../gc-sms/agent/tools.ts in-process tool surface, as a cautious adapter over the generated benefits bundle owned by gc-benefits. The adjacent resources lookup reads a runtime copy validated from this wiki's data/exports/resources.json rather than a hand-maintained file. See also the Benefits Landscape for the broader policy context and Market Gap for the unclaimed benefits analysis.

  1. GiveCare. "GiveCare Benefits Corpus." Source → ↩↩

  2. Code for America. "Benefits Enrollment Field Guide 2024." Source → ↩

  3. ACL. "2024 Report to Congress on the 2022 National Strategy to Support Family Caregivers." Source → ↩↩

  4. AARP/NAC. "Caregiving in the United States 2025." Source → ↩