Phase 0 Findings — GBMT-7 (Elder Care)

Date: 2026-08-03 Scope: First pass per §12's phase gate — verify the ★ Anchor Table rows, build the §2 population baseline as a committed pipeline, scan WA Cares (via the anchor-6 work) + the CLASS Act repeal + one HCBS waiting-list state case. Go/no-go input for full execution.

Verdict: GO — with one kill condition fired as a headline, one anchor broken, and one federal-status finding that changes the whole domain's framing

No kill condition fires as a stop. §12's first named kill condition fires exactly as specified: "if §2 shows the unpaid-caregiving baseline cannot be reconciled to within a defensible range across sources, that is the headline." It is not reconcilable, and that is now the strongest single finding this pass produced — it mirrors housing's shortage-estimate result almost exactly. The second named kill condition (§3: is direct-care wage vs. waiver-funding the binding constraint on HCBS capacity) was not testable within this pass's scope — it needs the kind of multi-pass workstream depth housing's §3 required, not a Phase 0 anchor check. Left open rather than force-verdicted.

Separately, one anchor (row 7, the CMS staffing rule) surfaced a finding bigger than the protocol anticipated: it doesn't just have "litigation and rescission risk" — it is dead, by both courts and Congress, as of February 2026. And one anchor (row 11) broke outright.

1. The unpaid-caregiving valuation is parameter-driven, not a fact (KILL CONDITION → HEADLINE)

AARP/NAC's own headline figure moved from $350B (2006) → $600B (2021) → $1.01T (2024), partly from a stated methodology revision, not just underlying growth. Two independent, non-root-sharing estimates for a comparable era diverge from each other and from AARP by more than 2×:

Three independent knobs explain the spread: population scope (all-condition caregiving vs. elder-care-only), wage-rate assumption (replacement labor cost vs. forgone personal earnings — these move in opposite directions depending on the caregiver's own income), and hours source (self-report survey vs. time-diary vs. panel data). None of the three studies is wrong; they are answering different questions that get reported under the same headline framing.

Effect on the protocol: §8 cannot cite a single caregiver-value figure as the number. It must report a conditional range by scope and method, the way §2 of the housing protocol had to report a conditional shortage surface rather than a point estimate. H8.1 survives as "the imputed value is large and load-bearing" but the specific-dollar-figure framing needs rebuilding.

2. Prior broken and reframed — the 2024 CMS nursing-home staffing rule is not "contested," it is dead (anchor 7)

The protocol's §1.4 setting taxonomy and §9 both flagged this as a live-status unknown requiring verification, not assumption. It has since resolved, decisively, in the two years since the rule was finalized:

• Original rule (April 2024): 3.48 total HPRD, 0.55 RN HPRD, 2.45 NA HPRD, 24/7 on-site RN, phased 2026–2029 compliance.
• Vacated by two federal courts — American Health Care Association v. Kennedy, Case Nos. 24-144 and 24-171, 777 F. Supp. 3d 691 (N.D. Tex. 2025), and Kansas v. Kennedy, No. C24-110-LTS-KEM (N.D. Iowa, June 18, 2025). Corrected on Phase 1 verification (2026-08-10): CMS's repeal rule (90 FR 55687) attributes a major-questions-doctrine holding to the Texas case only and states no ground for the Iowa case; "statutory-authority grounds in both" was this record's inference. Neither opinion has been read by this project (see method/sources/UNOBTAINED.md §3), so all characterisation of their reasoning comes from the repealing agency.
• HHS withdrew its own appeals of both losses in fall 2025. Unverified and withdrawn on Phase 1 verification: 90 FR 55687 records the Texas case as "(appealed June 2, 2025 to the Fifth Circuit)" and says nothing about withdrawal. No docket was reached.
• Congress imposed an implementation moratorium via H.R. 1 (2025) — Pub. L. 119-21 §71111, running from enactment (2025-07-04) to September 30, 2034, i.e. 9 years 3 months, not the "10-year" shorthand this record previously used. A belt-and-suspenders legislative kill shot on top of the court losses.
• CMS formally rescinded the rule, effective February 2, 2026, citing both the litigation and the moratorium.
• The facility-assessment and Medicaid staffing-spend-transparency provisions from the same 2024 rule were not rescinded and remain live.

Independently confirmed (AHA, Federal Register, CANHR, Center for Medicare Advocacy, Duane Morris, Skilled Nursing News) beyond the subagent that first surfaced it.

Effect on the protocol: §6, §9, and §11's candidate-architecture scoring all assumed a framing of "assess an emerging, contested federal standard." The actual research question for full execution is "assess the aftermath of a reversed one" — a different question, with a different evidence base (pre-rescission compliance data, if any exists; state-level minimum-staffing laws that predate and now outlive the federal rule; the political economy of why a rule modeled to save 13,000 lives/year was reversed). §11's federalism hypothesis (H9.1: is the binding lever federal or state?) gets a strong empirical answer for the staffing-standard question specifically: federal rule-making was reversed at the federal level twice over (court + Congress), which argues for state-level minimum-staffing laws as the more durable lever going forward — worth testing against WA Cares' own durability profile.

3. Prior broken — assisted living does not out-census skilled nursing (anchor 11)

§7's own seed anchor claimed AL/residential care serves a "comparable or larger" number of paid LTSS recipients than SNF. Point-in-time census comparison:

• AL/residential care: ~1.016M residents (CDC/NCHS NPALS, 2022 wave).
• SNF: ~1.24M residents (KFF analysis of CMS CASPER data, July 2025).

SNF is still ~20% larger by census. The claim is right in order of magnitude (both settings are on the same million-person scale, a genuinely under-appreciated fact relative to nursing-home-only framing) but wrong on direction. The vintages differ by ~3 years (2022 vs. 2025) and a fully independent AL resident count from NIC MAP specifically (rather than unit inventory) was paywalled and unobtainable in this pass — flagged, not treated as resolved.

Effect: §7's H7.1/H7.2 (undercounted incidents; growing Medicaid AL footprint) are untouched by this correction — they don't depend on AL being larger than SNF, only on it being large and thinly regulated. The research/regulatory-attention-asymmetry framing survives; the "larger" framing does not and should not recur in the whitepaper.

4. CLASS Act: a clean design-stage failure, and a legible transferability lesson

Confirmed as a pure adverse-selection failure, not a funding or political one. HHS's own 19-month actuarial review (2010–2011) found that CLASS's voluntary, guaranteed-issue enrollment with only a 5-year vesting period could not be priced into solvency — every design fix HHS tried (employer auto-enrollment, tiered/phased benefits, an 80% benefit cutback after 5 years) still left "high uncertainty about long-run solvency," and Secretary Sebelius stated in October 2011: "I do not see a viable path forward for CLASS implementation." It was formally repealed via the American Taxpayer Relief Act of 2012 (the Jan. 2013 "fiscal cliff" deal), which simultaneously created a Commission on Long-Term Care to study alternatives. No enrollment period ever opened, no premium was ever collected, no benefit was ever paid — this is a first-principles design failure, independently corroborated by CBO's own projection of the same structural problem (growing deficits post-2030).

Transferability lesson, testable against WA Cares: CLASS died specifically because voluntary participation let only the highest-risk people opt in. WA Cares's mandatory, universal payroll tax removes exactly that channel by construction. This is the single cleanest natural experiment available to §11's architecture-scoring stage — not "does a public LTC benefit work," but "does removing the adverse-selection channel that killed the last attempt actually fix it," and WA Cares' first-ever benefit payout (July 1, 2026, per anchor 6) means this pass catches it at the earliest possible moment, with essentially zero months of real payout data yet to evaluate.

5. HCBS waiting-list case: Texas, and a second "the data doesn't exist" finding

Texas was chosen over Florida (non-responsive to KFF's 2025 national survey, currently less documented) and Illinois. Corrected on Phase 1 verification (2026-08-10): Illinois was disqualified here as having "no waitlist at all." KFF's own brief reports Illinois waiting lists growing from 14,444 to 15,905 between 2023 and 2024 and lists Illinois among the states that do not establish eligibility until selection. The case-selection premise was wrong; the Texas and Indiana findings themselves are unaffected. Findings:

• Texas is one of the states that don't screen for eligibility before adding someone to an interest list. Corrected on Phase 1 verification: KFF reports "the eight states that do not screen people for eligibility on any lists, six have only waiting lists, one (Texas) has only interest lists, and one (Washington) uses both" — not "six states (with FL, IA, OK, OR, SC)" as this record had it — the state's own 2015 data (the last fully-documented baseline) shows 198,538 raw list entries collapsing to 101,948 unduplicated people, a ~95% duplicate-listing inflation, plus 55,009 people on the list who were already receiving other Medicaid LTSS services while nominally "waiting."
• A current (2025) STAR+PLUS HCBS-specific headcount (~15,850 against ~24,000 slots) could only be sourced from secondary aggregators; Texas HHSC's own primary reporting pages returned HTTP 403 to automated fetch and were not independently confirmed.
• Texas's institutional-vs-HCBS spending mix — the variable §5's H5.1 needs to test whether Texas's list size is an architecture story — could not be determined. Texas did not report FY2020 LTSS expenditures to CMS's Mathematica rebalancing study and was excluded from AARP's 2023 LTSS State Scorecard's spending-balance indicator specifically for data-accuracy reasons.

Effect: this is a second, independent instance of the "the US doesn't collect the data needed to answer its own question" finding-type that housing's LIHTC anchor produced — here for the state with the country's largest and most-discussed HCBS interest list. §5 should treat this not as a Texas-specific gap to route around, but as a candidate finding in its own right: the state most commonly cited as evidence for the institutional-bias thesis is also the state where that thesis is least verifiable.

6. Anchor verifications

Rows completed in §3 of the protocol.

7. Pipeline: proven for population, queued for spending

The ACS bulk-file route (proven in housing/baseline) works end to end for elder-care's population leg, committed in elder-care/baseline/:

Two real bugs were caught by inline identity checks during this session and are recorded in the incompatibility log: an off-by-one that pulled a margin-of-error jam value (-555555555) instead of the estimate for national population, and a SUMLEVEL-only geography join that silently admitted ~550 metro/non-metro sub-state breakdowns that B01001 publishes but housing's tables didn't — both caught because the numbers were implausible or the row count was obviously wrong, not because the pipeline assumed correctness.

LTSS spending by payer vs. utilization by setting — the other half of §2 — was not pulled this pass. No single bulk-file source publishes it the way ACS publishes population; it needs CMS National Health Expenditure Accounts combined with CMS-64 Medicaid expenditure data, a heavier lift queued for full execution.

8. Re-ranking for full execution

§12's prior was that §2, §3, §4, and §10 carry the most information per hour. Phase 0 partially confirms and partially reorders:

1. §8 (family caregiving) rises. The valuation-reconciliation finding is a headline in its own right and a direct analogue of housing's shortage-estimate result — worth promoting from background context to a primary deliverable finding.
2. §6/§9 (institutional quality / federalism) change shape, not rank. The staffing-rule rescission doesn't reduce their importance, but it changes the research question from "assess a live standard" to "assess why a standard modeled to save 13,000 lives/year got reversed, and whether state-level rules are the more durable substitute."
3. §5 (HCBS waiting lists) gets a sharper opening. The Texas case shows the state most commonly cited for the institutional-bias thesis is also the one where that thesis is least independently verifiable — a finding with a constituency, in the same shape as housing's LIHTC-cost result.
4. §11 (candidate architectures) gains a genuinely clean test case. CLASS's adverse-selection failure and WA Cares' mandatory design are as close to a controlled contrast as this domain offers, and WA Cares' first payout (July 2026) is caught at its earliest possible moment.
5. §3 (workforce) holds at high priority but is not yet tested against its own kill condition. Whether wages or waiver funding is the binding constraint on HCBS capacity needs workstream-level depth (the kind housing's §3 took four passes to resolve), not an anchor check — this is the single largest piece of unstarted work carried into full execution.

One time-sensitivity to carry: WA Cares' benefit payouts are brand new (weeks old at this writing). Any analysis of program performance, claims experience, or actual vs. projected utilization will be working with essentially no real outcome data yet — a constraint to state explicitly rather than paper over with projections dressed as results.