Groundwork's readiness feedback on the Build M default (David Trimmer, 2026-07-17): CTC/EITC/SALT much improved; four remaining items gate their sign-off. Measured on the certified export (98bf731) — each maps to a pipeline class this campaign hardened, and none has a Ledger fact today:
1. CDCC adult-care leg → structural zero (absent input)
No adult/disabled-dependent care expense column exists on the export at all — only the childcare legs (spm_unit_pre_subsidy_childcare_expenses, $71.2B). Any CDCC reform binding through adult-care expenses scores exactly $0: the input-coverage "absent input" class (#368). Fix ladder: impute adult-dependent care expenses in the staged base builder (source decision: ASEC/SIPP), add the leaf to the required coverage manifest, add a CDCC-adult-care reform smoke probe so this can never ship silently zero again (the keogh pattern).
2. Self-employed health insurance premiums → structural zero (missing attribution)
Premium mass exists in bulk (health_insurance_premiums_without_medicare_part_b $371.4B; other_health_insurance_premiums $361.3B) but no SE-attributed premium input, so the SE health ALD computes $0. Fix: attribute premiums to self-employed units in the base builder, require in coverage, target the SOI Table 1.4 self-employed health insurance deduction line (a real Ledger fact — same family as the self_employed_pension_contribution_ald fact already in the feed), probe.
3. Tips → thin + untargeted (~4–5× under)
tip_income: 549 carriers, $9.9B on the export, against the IRS W-2 social-security-tips class (order ~$40B; exact value cited when the Ledger fact is sourced). treasury_tipped_occupation_code carries 12,241 flagged workers, so the occupational scaffolding exists — the dollar imputation is thin and nothing pins it. 549 carriers is keogh-class thin: expect to need a carrier/mass protection (#446) or reselection alongside the target. OBBBA no-tax-on-tips scoring undercounts proportionally until fixed. Fix: Ledger fact (IRS W-2 tips aggregate), calibration target, imputation support widening, OBBBA tips probe.
4. Overtime → untargeted, reported undercount
fsla_overtime_premium: 8,748 carriers, $114.3B. No target, no authoritative anchor in the feed; Groundwork reports undercounting. Fix: source the anchor (BLS-derived FLSA premium aggregate / the JCT no-tax-on-overtime scoring convention), Ledger fact, target, OBBBA overtime probe.
Cross-cutting
All four are the #445 lesson generalized: untargeted thin inputs are unprotected against the solve, and unprobed reform paths can ship silent zeros. Items 1–2 need base-builder imputation work before targets can bind (the coverage-gate "Deliverable 2" class); items 3–4 are Ledger-sourcing + target + probe. Slotted with Build N alongside the CD target surface (#449) — Groundwork's bar: "if we can target these in the next deployment we should be ready to go."
Groundwork's readiness feedback on the Build M default (David Trimmer, 2026-07-17): CTC/EITC/SALT much improved; four remaining items gate their sign-off. Measured on the certified export (
98bf731) — each maps to a pipeline class this campaign hardened, and none has a Ledger fact today:1. CDCC adult-care leg → structural zero (absent input)
No adult/disabled-dependent care expense column exists on the export at all — only the childcare legs (
spm_unit_pre_subsidy_childcare_expenses, $71.2B). Any CDCC reform binding through adult-care expenses scores exactly $0: the input-coverage "absent input" class (#368). Fix ladder: impute adult-dependent care expenses in the staged base builder (source decision: ASEC/SIPP), add the leaf to the required coverage manifest, add a CDCC-adult-care reform smoke probe so this can never ship silently zero again (the keogh pattern).2. Self-employed health insurance premiums → structural zero (missing attribution)
Premium mass exists in bulk (
health_insurance_premiums_without_medicare_part_b$371.4B;other_health_insurance_premiums$361.3B) but no SE-attributed premium input, so the SE health ALD computes $0. Fix: attribute premiums to self-employed units in the base builder, require in coverage, target the SOI Table 1.4 self-employed health insurance deduction line (a real Ledger fact — same family as theself_employed_pension_contribution_aldfact already in the feed), probe.3. Tips → thin + untargeted (~4–5× under)
tip_income: 549 carriers, $9.9B on the export, against the IRS W-2 social-security-tips class (order ~$40B; exact value cited when the Ledger fact is sourced).treasury_tipped_occupation_codecarries 12,241 flagged workers, so the occupational scaffolding exists — the dollar imputation is thin and nothing pins it. 549 carriers is keogh-class thin: expect to need a carrier/mass protection (#446) or reselection alongside the target. OBBBA no-tax-on-tips scoring undercounts proportionally until fixed. Fix: Ledger fact (IRS W-2 tips aggregate), calibration target, imputation support widening, OBBBA tips probe.4. Overtime → untargeted, reported undercount
fsla_overtime_premium: 8,748 carriers, $114.3B. No target, no authoritative anchor in the feed; Groundwork reports undercounting. Fix: source the anchor (BLS-derived FLSA premium aggregate / the JCT no-tax-on-overtime scoring convention), Ledger fact, target, OBBBA overtime probe.Cross-cutting
All four are the #445 lesson generalized: untargeted thin inputs are unprotected against the solve, and unprobed reform paths can ship silent zeros. Items 1–2 need base-builder imputation work before targets can bind (the coverage-gate "Deliverable 2" class); items 3–4 are Ledger-sourcing + target + probe. Slotted with Build N alongside the CD target surface (#449) — Groundwork's bar: "if we can target these in the next deployment we should be ready to go."