Mortgage-interest dollars run ~25-30% hot (ht2 +29.5%, JCT expenditure +23.3% on certified O-1) — first-order for deduction-replacement analyses

On buildo-sparse-rmloss100-22bd902-20260722T232627Z (and consistent with N):

Returns fine, dollars ~25-30% hot ⇒ interest per mortgage-holder too high — the tips/medical signature inverted (level, not carriers). Not a #492 cap casualty: under the rational-tail experiment arm the JCT mortgage row got WORSE (+50.6%), i.e. uniform re-spreading trades ht2-family fixes against the reform-materialized JCT rows (finding-5 dynamics; the JCT expenditure family is not in the critical register at all — an adoption-gate gap for #492's step 3, flagging there).

Why it matters now: a client project replacing standard+itemized deductions with a refundable credit binds directly through itemized composition — total itemized is tight (−1.8% to −4.7% across four SOI tables) but the composition is skewed (mortgage +30%, medical +21%), which distorts winners/losers by decile even with the aggregate right (mortgage skews upper-middle; medical skews elderly).

Suspects, in order: the mortgage-interest imputation levels (tax_unit first/second_home_mortgage_balance/interest/origination_year columns — QRF/PUF-donor lineage, the #487 mean-reversion class), or a target-side concept gap (ht2 A19300-class mortgage interest vs what the model column measures). The #496 diagnose_us_target_support tool is built for exactly this — same forensics family as the six-state medical and tips-carrier work (#481/#487).

Start with the #496 diagnose_us_target_support tool and inspect the tax_unit first/second_home_mortgage_balance, interest, and origination_year columns, following the QRF/PUF-donor lineage. Determine whether the discrepancy is caused by imputation levels or a target-side concept gap, then validate the mortgage-interest returns and revenue-loss rows against the cited ht2 and JCT results.