Phase 2 Steelman — GBMT-1 Childcare — Target: the regulatory-constraint diagnosis

Verification Protocol method/verification-protocol.md, Phase 2 (S1–S5). Built independently; no prior red-team or fact-check view of this filing's weaknesses was consulted before S2 was written.

1. S2 — The tilt, derived from the filing's own text

(Written before any steelman research was begun.)

Direction the filing leans. GBMT-1 leans, throughout and without hedging, toward a factor-price diagnosis of scarcity: childcare is scarce because the people who provide it are paid too little to stay, and the fix is public money routed to compensation. The abstract states it outright — "The binding constraint on universal childcare in the United States is not appropriations but labor" — and the plain-talk gloss reduces it to four words: "Fix pay first." Part 2 is stamped Binding constraint. Part 7 orders the whole program around it: "pay, build, then promise." The scorecard's top-weighted dimension is relief_workforce, whose 5-anchor is "Funds compensation at parity + pipeline, DC-PEF-style," and every one of the four surviving architectures (a3 Head Start scaled, a6 supply-first, a8 public option, a9 fallback ladder) scores 4 or 5 on it. The constituency framing adopted is the ECE workforce coalition's own: NAEYC's Unifying Framework is cited approvingly (ws13) and the fault line it names — "credentials without compensation" — is read as an argument for more compensation, never as an argument for fewer credentials.

Central causal claim. Stated precisely, it is a two-step chain: (i) universal 0–12 coverage requires ~2.8M educator FTEs against ~1.05M today, so the required labour flow (~798k hires/yr) exceeds any pipeline; and (ii) because that flow is unattainable, money spent on demand before labour arrives converts into queues (Quebec), litigation (Germany), or price (Australia) rather than care. Therefore compensation is the lever with the highest derivative on coverage. The five-country comparative record in Part 5 is marshalled entirely as confirmation of step (ii).

The position argued against, scored down, or never entertained. The filing has a genuine opposing case on file — but it is steelman-feasibility.md, which argues the program is more feasible than assumed. That is a steelman along the optimism axis, not the diagnosis axis. It concedes the workforce constraint explicitly ("What this case does not claim: that the 2.8M-FTE workforce flow has a precedent (it does not — red-team attack 1 stands)"). All five red-team attacks likewise accept the labour diagnosis and contest sourcing, segmentation, or political sequencing within it.

The un-entertained position is the regulatory one, and the record shows this by construction:

1. Zero hits. deregulat*, relax* (of a standard), loosen*, regulatory burden appear nowhere in childcare/research/ or site/childcare/. ratio appears in the research record only in fte_model.py's docstring and deviations-log entry #4 — never as a policy variable, only as a modelling input.
2. Ratios enter as physics, not policy. fte_model.py treats staff-child ratios as an exogenous technological constant. The 2.8M figure — the filing's single most load-bearing number, and its Anchor 5 correction — is defined by RATIO["central"]. Nothing in the record asks whether that parameter is a fact about children or a fact about state administrative code.
3. No deregulatory architecture exists among the twelve. a1–a11 plus the cash comparator are all financing and delivery instruments. Not one alters the regulatory production function. A design space that contains "Tri-Share federalized" but not "harmonise or relax ratio and credential floors" has excluded the alternative diagnosis before scoring begins.
4. No scorecard dimension can register it. The 14 dimensions score money, coverage, procedure, federalism, durability, pass-through, participation, preference fit, integrity, distribution, evaluability. There is no dimension for cost-per-slot efficiency or regulatory burden. Even if a deregulatory architecture had been listed, the instrument could not have scored it.
5. The filing's own inquiry protocol asked, and the research pass did not answer. childcare/docs/research-inquiry.md §161–162 explicitly commissioned work on "square footage per child, outdoor space, egress and fire code for infants… zoning and land use as an obstacle." ws05-facilities.md is twelve lines about deserts and CDFI financing and does not mention a single facility code. The regulatory workstream was scoped and then dropped.
6. Where regulation does surface, it is assumed benign. ws11 notes that "ratio/credential mandates on the existing market are near-certain [Byrd] strikes" — treating the inability to impose them federally as a loss, never noticing that the same sentence concedes ratios are set by states and are therefore a live policy margin.

Steelman target, stated as the filing would have to refute it: the binding constraint on childcare supply is not the wage but the regulatory production function — mandated staff-child ratios, group sizes, credential floors, licensing and facilities codes — which fixes the number of adults per child by law, and thereby manufactures both the 2.8M requirement and the price that makes the wage look unpayable. On this diagnosis, compensation funding is not the first-best lever; it is the lever you are forced to reach for after the ratio has already determined how many bodies you must buy.

2. The steelman: the binding constraint is regulatory, not compensatory

Seven claims, each sourced to a document I fetched and read. Where I have only a search summary or a second-hand report, it is marked [UNVERIFIED AT SOURCE].

S-1. The 2.8M requirement is not a demographic fact. It is a legal parameter times a coverage guess.

The filing's own model makes this inarguable. fte_model.py computes FTE = population × coverage ÷ RATIO × OVERHEAD. Population is measured; overhead is assumed; coverage is assumed; and RATIO is a transcription of what state licensing law requires. Change no fact about children and no fact about the labour market — change only the statute — and the headline moves by more than half. The numbers are in §3 below.

The filing's docstring calls its ratios "licensing-typical central values." They are not. Its r02 = 4.5 cannot be produced by any combination of infant and toddler ratios at or above the median state's — holding infant at 4:1, it requires a toddler ratio near 4.8:1, which only Connecticut and DC (both 4:1) currently beat. The median US state requires infant 4:1 and toddler 7:1 (Prenatal-to-3 Policy Impact Center, Evidence Review ER 0920.012A, Table 7 — compiled from state licensing regulations), which blends to 5.60 children per FTE across the 0–2 band. NAEYC's own recommended standard (infant 4:1, toddler 6:1) blends to 5.14. The filing's central ratio for infants and toddlers is tighter than the median state requires and tighter than NAEYC recommends. The sensitivity band it does report (tight/loose, 4.0–5.5) is centred on that same too-tight point and never reaches the median state, let alone the legal range.

For 3–4-year-olds the filing uses 10:1 — exactly NAEYC's recommendation, not law. Florida requires 15:1 at age three and 20:1 at age four (Fla. Stat. § 402.305(4)(a)4–5, fetched from flsenate.gov). For 5–12 it uses 14:1; Florida's statutory minimum for age five and over is 25:1 (§ 402.305(4)(a)6), Mississippi's is 20:1 to age nine and 25:1 for 10–12 (MSDH, Regulations Governing Licensure of Child Care Facilities, ratio table, retrieved from the ACF national licensing-regulation repository).

This matters beyond arithmetic. Part 5 of the whitepaper holds Florida up as one of the durable American successes — "GA/OK/FL · Universal pre-K · Decades durable in red states." The filing praises a programme built under ratios roughly 1.7× looser at 3–4 and 1.8× looser at 5–12 than the ratios its own model uses to prove the programme is impossible. The success case and the impossibility proof are running on incompatible parameters.

S-2. Tightening ratios and credentials measurably destroys supply — and the best-identified estimates are large.

Hotz & Xiao, "The Impact of Regulations on the Supply and Quality of Care in Child Care Markets," AER 101(5): 1775–1805 (2011). Read at NBER w11873 (Dec 2005, revised Aug 2010), fetched and extracted. Panel data on establishments and local markets, with state, time, and establishment fixed effects, using the Census of Services and NAEYC accreditation records. Verbatim:

"such an increase in the stringency of the staff-child ratio for infants would reduce the number of child care centers in the average market by between 9.2% and 10.8% depending on the year analyzed."

The increase in question is one fewer infant per staff member — from the sample-mean maximum of 4.425 infants per staffer to 3.425. Reducing the director education requirement by one year runs the other way at 3.2–3.8% per year of education. And the damage is not evenly spread; the paper's abstract states the effect "reduces the number of center-based child care establishments, especially in lower income markets," and the body reports that in high-income markets the sign actually flips positive. This is the distributional core of the steelman: ratio stringency is a regressive tax on the supply of care, and the filing's own §5 finding that deserts are rural and worsening sits squarely in the markets where Hotz & Xiao say the damage lands.

Ali, Herbst & Makridis, "Minimum Quality Regulations and the Demand for Child Care Labor," IZA DP 14684 (August 2021). Fetched and extracted. This is the strongest recent quasi-experimental design on the question, and it is the one the filing most needed and does not have: it exploits the pandemic-era wave of state regulatory changes, identifying off variation within states over time and across children's age groups, which permits state × time fixed effects and therefore purges state-level pandemic policy. Verbatim:

"the introduction and increased stringency of group sizes and child-to-staff ratios reduces the demand for child care labor… reduces the number of child care job postings by 5.5 to 8.4 percent." "increasing the stringency of child-to-ratios has strong effects—leading to fewer job postings, fewer lead teacher and bachelor's degree postings, and more postings in violation of the state education regulations." "introducing a group size regulation has implications for the broader labor market, leading to lower labor force participation rates among mothers with children ages 0 to 10."

The maternal LFP effect is a 2.0 percentage point decline.

This result deserves emphasis because it inverts the filing's causal model. GBMT-1 treats the ratio as fixed technology and the workforce as the scarce input: more staff, more slots. Ali–Herbst–Makridis show the arrow runs the other way — tightening the ratio reduces total childcare employment, because the market-shrinking effect dominates the staffing-intensity effect. On this evidence, ratio regulation is not a claim on the workforce; it is a determinant of how large the workforce gets to be. Note also the authors' provenance: Chris Herbst is a mainstream early-childhood economist who publishes extensively on subsidy expansion, not a deregulation advocate.

Gorry & Thomas, "Regulation and the cost of childcare," Applied Economics 49(41): 4138–47 (2017). Identification is within-state across age groups, differencing out state-level unobservables. Reported estimates: allowing one additional child per staffer reduces the cost of care by 9–20%; requiring a high-school diploma of lead teachers raises infant care prices by 25–46%. [UNVERIFIED AT SOURCE — I could not retrieve the paper itself; both figures are quoted from the AEI report PDF I did fetch (Calder, Childcare Regulation and Affordability, October 2025, p. 4) and from the Mercatus working-paper landing page. Treat as second-hand.]

S-3. The regulation buys very little of what it is sold as buying. This is the ECE research community's own conclusion, not a deregulationist's.

This is the point most likely to move someone holding the filing's view, because none of the sources are hostile witnesses.

Perlman et al., "Child-Staff Ratios in Early Childhood Education and Care Settings and Child Outcomes: A Systematic Review and Meta-Analysis," PLOS ONE (2017). 29 eligible studies, 31 samples. Verbatim:

"the available literature reveal few, if any, relationships between child-staff ratios in preschool ECEC programs and children's developmental outcomes." "within the range of permissible child-staff ratios, variations in ratios have small, if any, associations with concurrent and subsequent child outcomes."

Pooled association with receptive language: r = 0.03, 95% CI 0.00 to 0.05. The authors also tested for the threshold story explicitly and rejected it: "A systematic comparison of the results from those studies whose mean child-staff ratios fell within the lowest, highest, and mid-range of values in our systematic review sample did not produce evidence to support a curvilinear trend between ratios and child outcomes."

Dalgaard, Bondebjerg, Klokker, Viinholt & Dietrichson, "Adult/child ratio and group size in early childhood education or care to promote the development of children aged 0–5 years: A systematic review," Campbell Systematic Reviews 18(2): e1239 (2022). 31 studies met inclusion; only 12, covering 8 populations (N = 4,300), were usable in synthesis. Process quality: ES = 0.10, 95% CI [−0.07, 0.27] — not significant. Language and literacy: ES = −0.04, 95% CI [−0.61, 0.53]. No socio-emotional outcome study could be included at all. "No study had an overall low risk of bias." The authors' conclusion:

"the research literature to date provides little guidance on what the appropriate adult/child ratios and group sizes are."

and they note "The existing studies on the topic are on average almost 30 years old."

Prenatal-to-3 Policy Impact Center, Vanderbilt, Evidence Review ER 0920.012A, "Child Care Ratios." This is a research centre whose institutional purpose is advancing prenatal-to-three policy. Its verdict on ratios is "Needs Further Study," with "The few causal studies identified for this review suggest mixed impacts." And, flatly:

"No US studies since then [the 1979 National Day Care Study] have employed true experimental designs to examine the effects of ratios or group size."

Its own summary of the strong causal evidence base is two studies, lettered A and B. Study A is Blau (1999), which finds null effects on cognitive and behavioural outcomes for infants and toddlers once mother fixed effects are included.

Early et al., "Teachers' Education, Classroom Quality, and Young Children's Academic Skills: Results From Seven Studies of Preschool Programs," Child Development 78(2): 558–80 (2007). Seven major datasets, and note the author list — Burchinal, Pianta, Peisner-Feinberg, Howes, Clifford, plus NICHD and IES staff. This is the field's establishment. Verbatim from the abstract:

"The findings indicate largely null or contradictory associations, indicating that policies focused solely on increasing teachers' education will not suffice for improving classroom quality or maximizing children's academic gains."

Drange & Rønning, "Child care center quality and early child development," Journal of Public Economics 188 (2020). Norwegian administrative data with quasi-random allocation of children to centres. The reported findings are that centres with a higher share of male staff produce better early language scores, and that centres with high staff sickness absence produce lower language and mathematics scores. Staff qualifications are examined and no qualification effect is reported. [Abstract read via the RePEc record; the article itself is paywalled — treat the qualification null as an absence of a reported finding, not a published null.]

The synthesis, and it comes from the filing's own §7. ws07-quality.md concludes that "quality parameters (curriculum, coaching, workforce stability) are not garnish — they plausibly determine the sign of child effects," and identifies what separated Boston (positive) from Tennessee (negative) as "Boston: curriculum + coaching + pay near parity; TN: contested implementation quality." Every item on that list is process quality. Not one is a staff-child ratio or a group-size cap. The filing's own adjudication of the central quality question names the inputs that matter, and they are not the inputs its model treats as physics. Structural regulation regulates the things that are easy to count and, on the best available evidence, is close to orthogonal to the things that work.

S-4. Cross-state variation: the raw correlations are contested and confounded — and I found the confound.

The filing has no cross-state analysis of this at all, so I built one, using its own data.

Method: BLS OEWS May 2024 state files (fetched from bls.gov; childcare workers 39-9011 plus preschool teachers 25-2011), divided by children 0–12 from the filing's own committed Census PEP v2024 state pull (childcare/baseline/data/pep_v2024_age0-12_by_state.csv), against 0–2 licensing permissiveness computed from the Prenatal-to-3 state ratio table. n = 50. Full output in §3.

The headline raw result goes against this steelman: looser-ratio states have fewer paid ECE workers per child (r = −0.49, t = −3.90) and lower childcare wages (r = −0.54). Taken at face value that is CAP's argument, and it is the honest first answer.

But the confound is decisive and measurable: corr(permissiveness, state all-occupation median wage) = −0.537. Strict-ratio states are rich states — DC, Connecticut, Vermont, Maryland, New York at one end; Louisiana, Mississippi, Arkansas at the other. Regulation is proxying for state income, which independently drives female labour-force participation, market childcare demand, and every wage in the state. Two things follow.

First, the childcare wage penalty is not a regulatory artifact and it is not fixed by regulation. Childcare workers' pay relative to their own state's all-occupation median is essentially uncorrelated with ratio stringency (r = −0.089, t = −0.62); it sits between 0.50 and 0.75 everywhere, in Mississippi and in Massachusetts alike. This is a point against a naive version of the steelman that blames low wages on regulatory cost burden, and it is a point for the filing's structural reading of the wage problem.

Second, headcount is the wrong supply measure, and mechanically so: a state with a 12:1 toddler maximum needs half the staff of a 6:1 state to serve the same children. The question is whether employment falls more than one-for-one with permissiveness. It does not. Regressing ln(ECE workers per 1,000 children) on ln(permissiveness) and ln(state all-occupation median wage) gives an employment elasticity of −0.551 (se 0.224, t = −2.45), 95% CI [−0.990, −0.111] — above −1, so capacity-equivalent supply rises with permissiveness, with a partial correlation of +0.318 (t = +2.30). That CI's upper bound sits close to −1 and this is cross-sectional with no causal identification; I state it as suggestive and nothing more. It is Ali–Herbst–Makridis, not this correlation, that carries the weight, and their within-state, across-age design finds the sign the raw cross-section hides.

Full output of the cross-state test (xstate2.py, this scratchpad):

n = 50 states+DC

RAW (bivariate, no controls) — the design CAP and AEI both use:
  corr(0-2 permissiveness, ECE workers per 1k children) = -0.490  t=-3.90
  corr(0-2 permissiveness, CAPACITY-equivalent slots/1k) = +0.221  t=+1.57
  corr(0-2 permissiveness, childcare median wage)        = -0.536  t=-4.40
  corr(0-2 permissiveness, ALL-OCCUPATION median wage)   = -0.537  t=-4.41   <- the confound

PARTIAL, controlling for state all-occupation median wage (state wage/price level):
  corr(permissiveness, ECE workers per 1k children     | state wage level) = -0.308  t=-2.22
  corr(permissiveness, CAPACITY-equivalent slots per 1k| state wage level) = +0.318  t=+2.30
  corr(permissiveness, childcare worker median wage    | state wage level) = -0.183  t=-1.28

RATIO of childcare wage to the state's all-occupation median wage (relative pay position):
  corr(0-2 permissiveness, childcare wage / all-occ wage) = -0.089  t=-0.62

                       0-2 ratio  wkrs/1k  CAPACITY/1k   ccwage  allwage  cc/all
Louisiana                   9.00    15.75        141.7   22,100   43,770    0.50
Georgia                     8.18    20.89        170.9   27,940   47,020    0.59
Mississippi                 8.18    17.21        140.8   21,760   39,070    0.56
Nevada                      8.18    11.92         97.5   29,140   46,440    0.63
New Mexico                  8.18    19.31        158.0   34,240   45,870    0.75
Arkansas                    7.71    17.79        137.3   27,180   41,020    0.66
New York                    4.62    24.09        111.2   36,630   58,560    0.63
Oregon                      4.62    19.21         88.7   36,250   53,390    0.68
Vermont                     4.62    33.54        154.8   37,830   52,410    0.72
Maryland                    4.50    17.52         78.8   35,150   58,050    0.61
Connecticut                 4.00    31.42        125.7   35,290   58,400    0.60
District of Columbia        4.00    38.43        153.7   45,400   88,000    0.52

====================================================================================================
ELASTICITY TEST (the honest version of the capacity claim)
====================================================================================================
cap = workers x ratio, so cap is positively related to ratio BY CONSTRUCTION unless
workers fall more than 1-for-1. The real question is the elasticity of ECE employment
w.r.t. legal permissiveness. If d ln(workers)/d ln(ratio) > -1, capacity rises.

  ln(workers/1k) = a + b*ln(ratio)                     b = -0.733 (se 0.183, t -4.01)
  ln(workers/1k) = a + b*ln(ratio) + c*ln(state wage)   b = -0.551 (se 0.224, t -2.45)

  uncontrolled                  : elasticity -0.733  ->  implied capacity elasticity = 1-0.733 = +0.267
                                  95% CI on elasticity [-1.092,-0.375]  -> is it above -1 (capacity rises)? not conclusively
  controlling state wage level  : elasticity -0.551  ->  implied capacity elasticity = 1-0.551 = +0.449
                                  95% CI on elasticity [-0.990,-0.111]  -> is it above -1 (capacity rises)? YES

On CAP's counter-claim. CAP's A Path Forward on Child Care Regulation (April 2025, PDF fetched) asserts "no correlation between the strictness of state regulations and state levels of child care supply." Root-tracing it: CAP endnotes 25 and 26 both point to a single source — a NAEYC blog post of 5 February 2020. That is the entire evidentiary basis. I attempted to retrieve it (naeyc.org 403s; Wayback is blocked from this environment) and could not, so I cannot check its method. What CAP itself describes is an unweighted four-component scoring index overlaid on a supply measure, with no mention of any control — i.e. precisely the bivariate cross-sectional design whose confound I just measured at r = −0.54. Under the Verification Protocol's own source hierarchy (V2), a think-tank blog post reporting an uncontrolled correlation does not outrank a fixed-effects panel study in the American Economic Review and a triple-difference design on a natural experiment. AEI's competing cross-state numbers (Calder 2025, PDF fetched: infant care $22,362 in 3:1 states vs $10,837 in 6:1 states; toddler care $20,196 in 4:1 states vs $7,254 in Mississippi's 12:1; R² = 0.69, infant-ratio coefficient p = 0.026 with a state-income control) have the opposite sign and barely better identification. Neither advocacy cross-section should decide this. The identified literature should, and it does.

S-5. The wage cost of relaxing ratios is small — smaller than one year of the programme the filing recommends.

The obvious objection is that loosening ratios is a wage cut by other means: fewer required bodies, slacker labour demand, lower pay. There is a direct estimate.

Garcia-Vazquez, "The equilibrium effects of state-mandated minimum staff-to-child ratios" (February 2024), Cowles Foundation / Federal Reserve Bank of Minneapolis. Fetched and extracted. An estimated equilibrium model of the childcare market with endogenous quality, endogenous childcare-worker and lead-teacher wages, and rich family heterogeneity. Verbatim:

"Going from the average regulation to the most stringent increases childcare worker wages by up to 2-3%. Moreover, going from the average regulation to the least stringent decreases wages by around 2% in all regions."

So moving every state from its current rules all the way to the loosest regime in the United States costs childcare workers about 2% of pay — on the filing's own 32, 050median, roughly * *640 a year**, against a wage-parity programme the filing prices at $45.9B/yr. The exchange rate is extraordinary: the entire wage cost of national ratio liberalisation is a rounding error against the compensation bill, while the capacity released is on the order of half the workforce gap. If wages are what you care about, you can buy back the 2% ninety times over out of the savings and still be ahead.

S-6. The regulatory path clears the procedural obstacle that dominates the filing's Part 4.

The filing spends its longest and most carefully corrected section on the Byrd rule, and concludes that quality standards are "safest housed inside federal spending programs" and that "ratio/credential mandates on the existing market are near-certain strikes" (ws11-legislative.md). Read that finding forward instead of backward. Ratios are state law. They require no federal vehicle, no reconciliation instruction, no parliamentarian, no 60 votes, no CBO score, and no appropriation. The filing's own ws16 sequencing already concedes a version of this — its Year 0–1 list of things needing "no new legislation" includes "zoning preemption, school co-location authority," and calls them "Cheap, fast, small." Ratio and credential reform belongs on that list, and on the arithmetic in §3 it is not small: it is the largest single lever in the model.

Against a filing whose central political finding is that the durable American precedents are state-level and bipartisan (Georgia, Oklahoma, Florida), the steelman's instrument is the one that matches that finding. It is also the one already moving: Iowa (2022), Wisconsin (2023, failed), Utah (2022 and 2024), South Dakota (2023), Idaho (2025) all legislated or attempted ratio, group-size, or licensure-threshold changes — enumerated, with citations to session law, at pp. 8–9 of the CAP report I fetched.

S-7. What the steelman does not claim

Stated explicitly, because a steelman that overclaims is refuted at its weakest point:

• Not that deregulation is free. §S-8 below is the counter-case, and part of it lands.
• Not that ratios never matter. They plausibly matter for infants under one, where the evidence is thinnest and the physical constraints most real, and where every state clusters at 3:1 or 4:1 anyway.
• Not that compensation funding is worthless. DC's Pay Equity Fund result is real; the claim is that it is the second lever, not the first, and that it is being asked to solve a shortage that regulation manufactured.
• Not that the 2.8M figure is fabricated. It is honestly computed from stated parameters. The claim is that the filing calls it a finding when it is substantially a choice, and that the deviations log's assurance of robustness "to ±1 ratio point" tests a band far narrower than the legal variation that actually exists.

S-8. The counter-case: where the regulatory diagnosis genuinely loses

(a) Credential requirements buy safety, and this is well identified. Currie & Hotz, "Accidents Will Happen? Unintentional Injury, Maternal Employment, and Child Care Policy," NBER w8090 (2001), published J. Health Economics 23(1): 25–59 (2004). Fetched and extracted. Vital Statistics panel with state fixed effects:

"requiring caregivers to have training beyond high school has large and significantly negative effect on accident rates… the point estimate suggests that such a requirement could lower the incidence of 'other' accidental deaths by 18%."

An 18% reduction in non-vehicular accidental child deaths is not a rounding error, and it attaches specifically to training beyond high school — a credential requirement. Any honest version of this steelman must separate ratios from training: the ratio evidence is weak-to-null on outcomes; the training evidence is not. Currie & Hotz also note this regulation, uniquely, "does not appear to reduce the use of regulated care" — it buys safety without crowd-out.

(b) Loosening ratios in home-based settings does harm. Same paper: "allowing higher ratios of children to caregivers in family homes is associated with slight increases in the rates of other accidents," at roughly 3.7% per additional child. Since the filing's own §8 finds that under-3s' modal arrangement is individual/home-based care, this is the segment where deregulation is most dangerous and where the steelman should not go.

(c) The crowd-out that helps supply may hurt children. Hotz & Xiao raise this against their own result: regulation pushes children from centres into family day care homes, and "a number of recent studies… have found that children in center-based care, especially between the ages of 3 and 4, have higher levels of cognitive and language skill development … than do children who spend an equivalent amount of time in family day care homes," with "the developmental benefits of center-based care… typically greater for children in poor and/or minority families." Run in reverse, a deregulated centre sector pulls children in from home care, which is developmentally good. But their second point cuts the other way: looser ratios mean "additional children cared for in these homes are cared for by the same number of staff," and they concede "there does not appear to be any evidence on the effects of differences in the number of children per provider within family day care homes." The evidence base runs out exactly where the policy would bite.

(d) The distribution of effects is not uniform, and some poor children lose. Garcia-Vazquez's counterfactuals show tighter ratios producing "big skill gains for some children and large drops for others," with gains concentrated among poor families least able to substitute away from paid care. Symmetrically, loosening ratios would take something from exactly that group. A regulatory diagnosis that ignores this is not honest.

(e) Boston is a bundle, and the bundle included credentials and parity pay. The filing's positive case — Gray-Lobe, Pathak & Walters on Boston pre-K — is a programme with BA-credentialed, near-parity-paid teachers plus curriculum plus coaching. Nothing in that design isolates the coaching from the credential. The steelman's claim that credentials are inert cannot be established from Boston, and Boston is the strongest positive attainment result in the US literature.

(f) The direct cross-state correlation, on its face, favours the filing. Restated so it is not buried: looser-ratio states employ fewer ECE workers per child and pay them less. The confound is real and measured, but a reader who declines to accept my confound adjustment is left with a raw correlation pointing the filing's way.

(g) I could not find an evaluation of the cleanest available credential experiment. The 2007 Head Start reauthorization required 50% of teachers to hold BAs by 2013 — a national, dated, enforced credential mandate. I searched for a quasi-experimental evaluation of its effect on child outcomes and did not find one. Per V5 I record this as a failed search, not as a demonstrated absence.

3. Attacking the model directly: the FTE requirement under alternative ratio assumptions

Method: fte_variants.py (in this scratchpad) re-runs /home/user/gubment/childcare/baseline/scripts/fte_model.py with every parameter except RATIO held at the filing's committed value — same population file, same coverage scenarios, same 1.30 overhead, same 1,050,000 baseline workforce, same 30% turnover, same 8-year ramp, same wages, same 0.65 labour share. Ratio regimes are built only from ratios I verified in primary sources, blended explicitly (0–2 band = ages 0/1/2 in equal thirds; 3–4 band = ages 3/4 in halves; 5–12 band split per the governing statute).

EFFECTIVE BAND RATIOS DERIVED FROM VERIFIED LAW (children per FTE educator)
  Florida     0-2 = 5.91   3-4 = 17.14   5-12 = 25.00
  Mississippi 0-2 = 7.61   3-4 = 14.93   5-12 = 21.62
  US median   0-2 = 5.60
  NAEYC rec.  0-2 = 5.14
  FILING      0-2 = 4.50   3-4 = 10.00  5-12 = 14.00   <-- tighter than the US median AND tighter than NAEYC

================================================================================================================
REQUIRED EDUCATOR FTEs, filing's CENTRAL coverage scenario, all non-ratio parameters identical
================================================================================================================
regime                       r02    r34   r512     0-2    3-4   5-12    TOTAL  net new hires/yr   $B/yr
A  filing central           4.50  10.00  14.00    1601    781    424     2807     1757      798   209.0
B  filing tight             4.00   8.00  12.00    1801    977    495     3273     2223      926   243.7
C  filing loose             5.50  12.00  15.00    1310    651    396     2357     1307      675   174.5
D  NAEYC recommended        5.14  10.00  15.00    1401    781    396     2579     1529      735   191.8
E  US median state law*     5.60  10.00  14.00    1287    781    424     2492     1442      712   184.3
F  Florida statutory        5.91  17.14  25.00    1219    456    238     1913      863      552   143.7
G  Mississippi reg.         7.61  14.93  21.62     947    523    275     1745      695      506   129.7

CHANGE vs the filing's published headline (2,807k FTE | 1,757k net new | 798k hires/yr | $209.0B/yr):
regime                        TOTAL          net new         hires/yr           $B/yr        
A  filing central              2807     +0%     1757     +0%      798     +0%   209.0     +0%
B  filing tight                3273    +17%     2223    +27%      926    +16%   243.7    +17%
C  filing loose                2357    -16%     1307    -26%      675    -15%   174.5    -17%
D  NAEYC recommended           2579     -8%     1529    -13%      735     -8%   191.8     -8%
E  US median state law*        2492    -11%     1442    -18%      712    -11%   184.3    -12%
F  Florida statutory           1913    -32%      863    -51%      552    -31%   143.7    -31%
G  Mississippi reg.            1745    -38%      695    -60%      506    -37%   129.7    -38%

================================================================================================================
SHARE OF THE HEADLINE THAT IS THE RATIO PARAMETER ALONE
================================================================================================================
  D  NAEYC recommended       removes    228k of the 1,757k net-new shortfall =  13% of the headline, from the ratio parameter alone
  E  US median state law*    removes    315k of the 1,757k net-new shortfall =  18% of the headline, from the ratio parameter alone
  F  Florida statutory       removes    894k of the 1,757k net-new shortfall =  51% of the headline, from the ratio parameter alone
  G  Mississippi reg.        removes   1062k of the 1,757k net-new shortfall =  60% of the headline, from the ratio parameter alone

================================================================================================================
THE FILING'S OWN DOMINANT LEVER vs THE RATIO LEVER (annual hires needed, central coverage)
================================================================================================================
  filing ratios, turnover 30% (published)            :    798k/yr
  filing ratios, turnover HALVED to 15% (filing's fix):    509k/yr
  Florida ratios, turnover 30% (unchanged pay)        :    552k/yr
  Florida ratios AND turnover halved (both levers)    :    330k/yr

What these numbers say, stated plainly.

• The filing's published headline — 2,807k FTEs, 1,757k net new, 798k hires/yr, $209.0B/yr — is reproduced exactly at regime A, confirming the re-run is faithful.
• Substituting Florida's own statutory ratios — currently legal, in a state the filing names as a durable universal-pre-K success — cuts the requirement to 1,913k FTEs, net new to 863k, annual hires to 552k, and steady-state cost to $143.7B/yr. That is 51% of the entire net-new shortfall erased by the ratio parameter alone, and $65B a year.
• Mississippi's ratios cut net new to 695k — 60% of the headline — and cost to $129.7B/yr.
• Even the wholly uncontroversial substitutions move it: NAEYC's own recommended standard removes 13% of the shortfall, and the median US state's actual law for the 0–2 band removes 18%, with the 3–4 and 5–12 bands left untouched at the filing's values.
• The decisive comparison is the last block. The filing calls halving turnover "the arithmetically dominant lever" and it is: 798k → 509k hires/yr. Adopting Florida's ratios, with no pay increase at all, gets to 552k — within 8% of the same result, for free, without federal legislation. Do both and it is 330k/yr, which is the first number in this whole exercise that looks like something a labour market could actually supply.

The honest accounting of what is parameter and what is finding. The filing's deviations log entry #4 defends the ratio choice on the ground that "National totals robust to ±1 ratio point (sensitivity shown)." That is true and it is not the relevant test. ±1 ratio point is narrower than the legal variation across US states, which for toddlers runs 4:1 to 12:1 — a factor of three. Across the range of ratios that American states actually enforce today, the model's central output moves from roughly 1.7M to 3.3M FTEs. The "2.8M" is therefore about as much a statement about American administrative law as it is a statement about American children, and the whitepaper's Part 2 stamp — Binding constraint — presents it as the latter.

This does not make the number wrong. It makes it conditional, and the condition is a policy variable the filing never treats as one.

4. S5 — Scorecard consequences

Two structural problems, not one contested cell.

1. A missing architecture. There is no a12/a13 "regulatory harmonisation and ratio reform" in the candidate set. On the filing's own 14 dimensions it would score unusually well where the winners score worst: procedural = 5 (state law; no federal vehicle, no Byrd exposure — the constraint that dominates the filing's Part 4), durability = high (statute, not annual appropriation — the exact defect the filing identifies in DC's Pay Equity Fund), federalism = 5 (no holdout problem; it is the states). It would score badly on relief_workforce (1 — funds no compensation) and is contested on distributional. It cannot be scored at all on the dimension that would matter most, because —
2. A missing dimension. Nothing in the 14 scores cost per delivered slot or regulatory burden. Every architecture is scored on how much money it moves and how well it survives, never on how many children a dollar buys. A scorecard built that way cannot rank an intervention whose entire mechanism is lowering unit cost. The stable top-four is therefore a top-four of financing instruments, which is a narrower claim than "four designs survive every ranking."

Sensitivity: under a re-score adding a cost-per-slot dimension, a6 (supply-first) and a3 (Head Start scaled) — both high-unit-cost, standards-inside-the-programme designs — lose ground, and a4 (K–12 extension) and a11 (caregiver choice) gain. Whether the top-four order flips I cannot say from outside the scoring; that the instrument is incapable of testing it is the finding.

5. Honest self-assessment

Where this case is strong

1. The model attack is unanswerable and is the strongest thing here. It requires no contested literature. It is arithmetic on the filing's own committed script, with ratio values verified in the Florida statute and the Mississippi regulations. Half the headline shortfall is a parameter choice, and the parameter is tighter than the median state's law and tighter than NAEYC's recommendation. Nothing in the record acknowledges this.
2. The "what does the regulation buy" evidence is genuinely strong and comes from friendly witnesses. Perlman (PLOS ONE), Dalgaard (Campbell), Early et al. (Child Development), and Vanderbilt's Prenatal-to-3 Center are the early-childhood research establishment, not a deregulation lobby, and they converge: ratio and credential variation within the legally permissible range has little measurable relationship to child outcomes, the causal base is two studies and a 1979 experiment, and the field says so itself.
3. Ali–Herbst–Makridis is the best-identified paper either side has, and it inverts the filing's causal model: tightening ratios reduces childcare employment, so the ratio is not a claim on the workforce but a determinant of the workforce's size.
4. The procedural argument is strong on the filing's own terms. Its longest and most corrected section establishes that the federal path is procedurally treacherous. The regulatory lever needs none of that machinery, and the filing's own sequencing file already has a bucket for exactly this kind of move.
5. The exchange rate is favourable and specific. ~2% of childcare wages (Garcia-Vazquez) against ~50% of the workforce gap and ~$65B/yr.

Where it breaks

1. It does not survive as a case about credentials. Currie & Hotz find training beyond high school lowers non-vehicular accidental child deaths ~18%, with state fixed effects and sensible placebo behaviour (no effect on cancer deaths). That is the best single causal estimate in this whole file and it points the other way. The steelman is strong on ratios and group sizes; on credentials it is weak, and I would not defend credential deregulation on this evidence.
2. It does not survive in home-based settings. Same paper: looser ratios in family day care homes raise accident rates. That is precisely the 0–2 segment the filing's §8 identifies as the modal arrangement.
3. The cross-state evidence does not support it on the surface. My own test, using the filing's own population data, found looser-ratio states have fewer ECE workers per child (r = −0.49). I can show the confound (state wage level, r = −0.54) and show that the employment elasticity is above −1 so capacity rises — but the 95% CI is [−0.99, −0.11], close to the boundary, and it is cross-sectional. A sceptic is entitled to stop at the raw correlation.
4. The wage-causation claim fails. Childcare pay relative to its own state's all-occupation median is uncorrelated with ratio stringency (r = −0.089). Low relative pay is not a regulatory artifact. The filing is right that the wage problem is structural; the steelman cannot claim regulation caused it.
5. The strongest positive US result is a bundle that includes what the steelman would remove. Boston pre-K had BA-credentialed, near-parity-paid teachers. Nothing separates the coaching from the credential.
6. Distributional harm is real and identified. Garcia-Vazquez shows tighter ratios produce large skill gains for poor children unable to substitute away from paid care; loosening takes that from them. Hotz & Xiao's crowd-out cuts both ways and they say so.
7. A citation I could not root-trace, in both directions. CAP's "no correlation" bottoms out in a 2020 NAEYC blog post I could not retrieve (403 + Wayback blocked), so I cannot rule out that its method is better than CAP's description implies. And Gorry & Thomas's headline estimates are second-hand to me.

The adjudication I would record under S4

The steelman partially survives, and it partially survives on the most load-bearing claim in the filing. Not a defeat: the filing is right that at current ratios the labour flow is unprecedented, right that funding without workforce produced queues in five countries, and right that low pay drives the turnover that dominates hiring arithmetic. None of that is touched.

What does not survive intact is the word binding, as the filing's own inquiry document defines it: "a constraint is binding for a design if, with all other constraints at observed values, relaxing it alone would materially raise achievable coverage or ramp rate, and relaxing the others would not." By that definition, relaxing the ratio constraint alone raises achievable coverage by roughly as much as the filing's own dominant lever, at no fiscal cost — so the second half of the test ("and relaxing the others would not") fails on the filing's own terms. Labour is a binding constraint at the ratios the model assumes; it is not the binding constraint independent of them.

The correction I would propose is narrow and does not require withdrawing the recommendation:

Before: "The binding constraint on universal childcare in the United States is not appropriations but labor: full coverage for ages 0–12 requires roughly 2.8 million educator positions…"

After: "Under prevailing staff-child ratios, the binding constraint on universal childcare is not appropriations but labor: full coverage for ages 0–12 requires roughly 2.8 million educator positions — a figure that falls to about 1.9 million under ratios Florida's statute already permits. The ratio is a policy choice, and it is the one lever comparable in force to compensation."

And I would add to the honesty box: the workforce requirement is a function of a regulatory parameter this filing did not examine, whose evidentiary basis in child outcomes is rated "needs further study" by the field's own systematic reviews.

6. Source register

Fetched and read in full or in relevant part by this session:

Not obtained: NAEYC blog (5 Feb 2020) underlying CAP's correlation claim — 403, Wayback blocked. Gorry & Thomas (2017) full text — quoted second-hand from #14. Bowne et al. (2017) EEPA — PDF unreadable; its 7.5:1 threshold claim is not relied on above, and Perlman (#5) tested for and rejected a threshold.