GUBMENTPlain talk · policy frontier
Filings / Childcare / Sources / Independent re-score & reconciliation
GBMT-1 · Research record · No. 1

Independent re-score & reconciliation log (GBMT-1)

childcare/research/scorecard/rescore-log.md
This is a working research document from the childcare filing, published as written — including the parts later corrected. It is the underlying record for Whitepaper No. 1, not a summary of it.

Date: 2026-08-07 Method: Per M6/M10 — structural blinding via scripts/batch-rescore.py. The blind scorer received only a tool-less request containing: architecture specs, the 14 anchored scales, scoring disciplines, and ws03–ws13 + ws15. Excluded (cannot be reached): scores.csv, rationale.md, rankings.txt, red-team log, deviations log, sequencing, site pages, ws14 (pre-score leanings). Raw output: blind-scores-2026-08-07.md. Reconciled cell-by-cell below.

Supersedes deviations #2/#6 (same-analyst 3-architecture sample). That sample is no longer the filing's independence claim.

Headline: the stable set was an evidence-density artifact

The re-score's largest contribution is not any single cell — it is exposing that the published "four designs survive every ranking" finding rested on above-neutral scores without cited effect evidence, especially on the federal-fallback ladder (a9) and on distributional/evaluability columns across several winners. The blind scorer applied the evidence floor both ways. After reconciliation:

Finding Before After
Stable across all 4 weightings a3, a6, a8, a9 a3, a4, a8
In top-4 under some but not all (none claimed; actually two exact ties existed) a6 (3 of 4), a11 (equity only)
Bottom two under every weighting misstated as "cash and Tri-Share" a5 demand-side allowance, a10 federalized Tri-Share
Cash comparator claimed "ranks last" mid-table (7th equal / 7th–10th by weighting)

K–12 extension (a4) enters the stable set on the strength of evidenced red-state durability, near-parity district pay, and direct-provision pass-through — while retaining its definitional coverage_02=1. It is a stable component for the school-age band, not a universal architecture. Federal fallback (a9) keeps federalism=5 (the holdout-proofing finding stands) but loses unevidenced 4s on workforce, supply, durability, and distributional — and leaves the all-weightings top-4.

Cell-by-cell reconciliation (material moves only; identical cells omitted)

47 cells corrected. 25 judgment splits kept-and-logged.

Asymmetric floor — below-neutral without evidence → 3

Cell Orig Blind Resolved Basis
a1 coverage_512 2 3 3 Age scope not in evidence base
a1 durability 2 3 3 Mandatory funding vs no dedicated revenue = wash at 3
a1 evaluability 2 3 3 Observational cross-state variation = anchor 3
a1 participation_risk 2 3 3 Cost-estimation rate-setting = anchor 3
a2 relief_workforce 2 3 3 No compensation mechanism evidenced
a2 integrity_risk 2 3 3 Unevidenced
a2 evaluability 2 3 3 Unevidenced
a5 evaluability 2 3 3 Unevidenced
a7 evaluability 2 3 3 Unevidenced
cash coverage_02/34/512 2 3 3 Cash makes no care offer; anchors fit badly → neutral not 2
cash integrity_risk 2 3 3 Unevidenced
cash evaluability 1 3 3 Unevidenced; 1 was mechanism reasoning

Asymmetric floor — above-neutral without evidence → 3

Cell Orig Blind Resolved Basis
a2 distributional 4 3 3 No incidence evidence for this design
a3 relief_workforce 4 3 3 Grants cover wages indirectly; no parity/pipeline = anchor 3
a3 distributional 5 3 3 Means-tested HS ≠ scaled-universal incidence
a6 distributional 4 3 3 Rural-weighting instruction ≠ measured incidence
a6 evaluability 4 3 3 National staging creates no comparison group in record
a7 relief_workforce 4 3 3 Revenue instrument, not a compensation program
a7 coverage_34 4 3 3 Benefit design unspecified
a8 distributional 4 3 3 Unevidenced
a8 evaluability 4 3 3 DoD case still pending in record
a9 relief_workforce 4 3 3 No compensation mechanism specified
a9 relief_supply 4 3 3 Capacity effects unevidenced
a9 durability 4 3 3 Unevidenced
a9 distributional 4 3 3 Unevidenced
cash distributional 4 3 3 No child-benefit incidence evidence in base

Evidence corrections (blind citation accepted)

Cell Orig Blind Resolved Basis
a2 procedural 2 3 3 Byrd exposure on standards = "reconcilable with major strips"
a2 passthrough_risk 2 4 4 Administered rates; Quebec rations rather than inflates
a3 procedural 3 4 4 ws11: standards inside federal spending programs
a3 passthrough_risk 4 5 5 ARPA supply-side operating grants = anchor 5
a4 relief_workforce 3 4 4 Boston/district near-parity pay (ws07/ws13)
a4 coverage_34 4 5 5 Universal pre-K sustained at scale (GA/OK/FL)
a4 federalism 3 4 4 Red-state durability lowers holdout risk vs match designs
a4 passthrough_risk 4 5 5 Direct public provision
a4 preference_fit 2 3 3 Centre-fit for 3–5 washes against no 0–2 / no home-based
a5 relief_supply 1 2 2 Holds nothing new but is not pure "builds no capacity" void
a5 federalism 4 5 5 No state opt-in machinery
a6 relief_supply 5 4 4 Retention dominates new-build arithmetic (ws04)
a7 durability 5 4 4 Dedicated revenue without institutional entanglement
a8 procedural 2 4 4 ws11 names architecture 8 with a3 as Byrd-defensible home
a8 federalism 4 5 5 Direct-federal = holdout-proof
a8 integrity_risk 4 5 5 Direct provision = anchor 5
a10 durability 2 3 3 Multi-state replication vs pilot scale = wash
a11 passthrough_risk 2 3 3 Kin payments outside priced market vs purchased-care risk = wash
a11 integrity_risk 1 3 3 Netherlands caution vs direct-to-family simplicity = wash

Judgment splits kept (original stands; blind reading logged)

Cell Orig Blind Kept Why
a11 coverage_02 5 4 5 Universal home-based offer matches anchor 5; take-up dock is judgment
a5 distributional 3 2 3 Desert/rate argument conflates supply failure with incidence (sample re-score already corrected this direction once)
a3 coverage_34 / participation / preference 5/4/3 4/5/2 kept Magnitude thresholds on universal pre-K and centre-vs-home fit
a4 procedural / evaluability 3/3 4/4 kept Federal vs state-led vehicle; lottery-evidence credit
a5 coverage / participation / preference / integrity various +1 kept Portable-money readings vs as-implemented inflation record
a6 coverage_02/34 3/3 4/4 kept Credit for eventual universal eligibility vs build-out-only present
a7/a8/a9/a10/a11/cash procedural & risk cells various ±1 kept See blind basis notes; none move a ranking on their own

Reconciled matrix

Architecture Wf Sup 0–2 3–4 5–12 Proc Fed Dur Pass Part Pref Int Dist Eval
a1 CCDF supercharge 3 3 3 3 3 4 2 3 3 3 3 3 3 3
a2 Medicaid entitlement 3 2 4 4 3 3 2 4 4 2 3 3 3 3
a3 Head Start scaled 3 4 4 5 3 4 5 4 5 4 3 4 3 3
a4 K–12 extension 4 4 1 5 5 3 4 5 5 4 3 4 3 3
a5 Demand allowance 1 2 3 3 3 5 5 3 1 3 3 1 3 3
a6 Supply-first 5 4 3 3 3 4 3 2 5 4 3 4 3 3
a7 Social insurance 3 3 3 3 3 2 3 4 3 3 3 3 3 3
a8 Public option 5 4 4 4 3 4 5 3 5 5 2 5 3 3
a9 Fallback ladder 3 3 4 4 3 3 5 3 3 3 3 3 3 3
a10 Tri-Share federal 2 2 2 2 2 3 3 3 3 3 3 3 2 3
a11 Caregiver choice 2 2 5 3 3 4 4 3 3 5 5 3 3 3
Cash comparator 1 1 3 3 3 5 5 3 2 5 4 3 3 3

What survives, what changed

Effects applied

← All Childcare research documents Sources digest Read the whitepaper