GBMT-1 Childcare — blind re-score

Column key (matrix order): A relief_workforce · B relief_supply · C coverage_02 · D coverage_34 · E coverage_512 · F procedural · G federalism · H durability · I passthrough_risk · J participation_risk · K preference_fit · L integrity_risk · M distributional · N evaluability Row numbering: 1 = a1_ccdf_supercharge … 12 = cash_comparator (matrix order as given in the brief).

1. The matrix

2. Cell-by-cell basis

3. What my matrix implies

On my own numbers the strongest row is a4 (K–12 extension) — it is the only architecture with cited evidence on almost every axis (durable red-state universal pre-K, dedicated Georgia-lottery-style revenue, direct provision on the pass-through and integrity axes, near-parity pay through district salary structures, and explicit school-age scope) — but it carries a hard 1 on coverage_02, which is a disqualifying gap on the age band the rest of the record says is hardest. Behind it sit a3 (Head Start scaled) and a8 (public option), which win on the four axes the record speaks to most loudly (federalism/holdout-proofing, pass-through, participation, integrity, plus ws11's Byrd point about standards inside spending programmes) and lose on preference fit for 0–3; and a6 (supply-first), the only row with a 5 on relief_workforce, dragged down by an evidenced 2 on durability (DC PEF's annual survival battles; the ARPA cliff). The weakest row is a10 (federalized Tri-Share), the only architecture whose real-world instance is measured in the record and comes out at ~1,000 child-years. a5, a2, cash_comparator all land at the same total (43) by different routes — a5 buys procedural cleanliness at the cost of pass-through and integrity, a2 buys legal entitlement at the cost of federalism and provider participation, and the cash comparator buys nothing on the supply side at all, which is what the brief said it was there to show. Large parts of the board are flat at 3, particularly distributional (ten rows at 3) and evaluability (eleven rows at 3) — the evidence base simply contains no architecture-specific findings on either.

4. Evidence-coverage count

Cells scored on cited evidence (non-3 cells plus 3s marked evidenced / evidenced-wash), out of 14:

Board totals: 113 evidence-based, 55 unevidenced-neutral. Of the evidence-based cells, 34 are flagged judgment-call and 20 are evidenced-wash.

5. What I could not score, and what would change my mind

Whole rows I could not score. a7 (social insurance) is 12/14 neutral: the evidence base contains a revenue fact (Georgia lottery earmark verified; Vermont payroll tax "pending primary-source pass") and nothing about what the fund buys, who administers it, or payroll-tax incidence. Its two non-neutral cells rest on the earmark-durability finding and on ws11's budget-window logic. a9 (fallback ladder) is 8/14 neutral: the record is emphatic about the federalism value of a fallback (ws06) and silent about everything downstream of it. cash_comparator is half-neutral because the base contains no CTC or child-benefit evidence at all — it is a comparator with no comparator literature attached.

The a11 maternal-employment trade. The brief says a11's maternal-employment tradeoff "appears in the demand/precedent evidence." I could not find it. ws15 lists "Nordics" among remaining §15 work, and no document in the base contains a home-care-allowance employment finding. I have therefore named the trade here rather than hiding it, but I did not score any cell down on it — that would be scoring an assertion in the row spec, not evidence. Had the Finnish employment evidence been in the base, I would expect it to bite on distributional (foregone maternal earnings) and possibly relief_supply (withdrawal from the formal-care market), and I would likely have scored 11M and 11B lower.

Anchors that fit badly. (a) participation_risk is written entirely about provider acceptance of public rates, which leaves employer-mediated (a10) and family-side (a5, a11, cash) designs scoring well on an axis where their real participation risk lies elsewhere; I disclosed this at 10J. (b) The three coverage anchors measure a care offer, which makes the cash comparator unscoreable in the intended sense — I held it at 3 rather than invent a reading. (c) evaluability reserves 4–5 for staged/lottery rollout, which no architecture on this board specifies, so the column is nearly constant by construction; a4's 4 rests on the observation that oversubscribed district pre-K has in fact produced the record's only causal evidence.

What would move specific cells. A DoD childcare deep-dive (flagged pending in ws15) would settle 8A/8B/8K and probably lift a8. A Head Start age-scope and setting-mix source would settle 3E and 3K, which are currently my reading rather than the record's. Medicaid holdout counts and tribal set-asides (flagged pending in ws06) would firm up 2G, which I scored 2 on the CBO/BBB analogue rather than Medicaid-specific data. The Netherlands case in any detail would decide whether 5L is a 1 or a 2. A primary source on Vermont Act 76 and the NM permanent fund would take a7 off the neutral line. And any evidence on who receives benefits under each design — the distributional column is where this board is most thoroughly blind.