Bind SOI Table 2.1 mortgage targets to the capped deductible concept (#511 target-side leg)

[MaxGhenis]

Summary

Target-side leg of #511 (mortgage dollars +29.5% on certified O-1), plus the release-gate hardening from cross-family review. SOI Table 2.1 "Home mortgage interest" reports the amount deducted on Schedule A — post the section 163(h)(3) acquisition-debt caps — while the Build H mapping (#299) summed the gross home_mortgage_interest export input over itemizers. The target column therefore measured a strictly larger concept than the fact it was pinned to.

Two commits:

1. Remap both Table 2.1 mortgage measures (home_mortgage_interest_amount, home_mortgage_interest_returns) to the engine's capped deductible_mortgage_interest_tax_unit, adding the concept to _SOI_ITEMIZED_ONLY_VARIABLES so any future record set carrying these measures stays itemizer-masked regardless of universe. Regression test pins the mapping (verified to fail on the pre-diff code).
2. Register the rows as critical (sol review, finding 1 — the adoption-gate gap Mortgage-interest dollars run ~25-30% hot (ht2 +29.5%, JCT expenditure +23.3% on certified O-1) — first-order for deduction-replacement analyses #511 flagged): the amount row at an interim 0.20 with a dated ratchet condition (to US_CRITICAL_DEDUCTION_MAX_ABS_RELATIVE_ERROR once the PUF donor home_mortgage_interest carries the E19200 total-interest concept (+7.6% vs the Table 2.1 mortgage-only leg) #515 donor carve lands and a run holds 0.15), the returns row at the standard 0.15, both with no incumbent-improvement escape, matching their itemized siblings. A builder-side regression pins the exact shipped O-1 diagnostics failing the release gate and the expected post-remap fit passing; the contract-side deduction parametrizations now cover the mortgage row.

Receipts (certified O-1, buildo-sparse-rmloss100-22bd902, exact reproduction of the diagnostics rows)

itemizer-masked mass at final weights	value	vs $186.31B aged fact
gross home_mortgage_interest (shipped target column)	$241.27B	+29.5%
capped deductible_mortgage_interest_tax_unit (this PR)	$214.20B	+15.0%
÷1.0758 donor E19200→A19300 concept scale (#515, donor-side leg)	$199.11B	+6.9%

• The cap haircut alone is a $27.1B pure concept gap in the target column; SOI's fact already embeds it.
• Carrier set: a positive-balance unit's deductible share is always positive; gross and structural columns are independently imputed, so carrier equality is a data property verified on O-1 post-itemizer-mask — identical weighted carrier totals (11,929,444.9), minimum deductible share 0.0835, the only gross-vs-structural divergent units are non-itemizers (returns row +2.45% untouched).
• Pull-through preserved: the capped formula is linear in the first/second_home_mortgage_interest export inputs — balances/origination years only set the deductible share — so calibration still binds the export mass Build H wanted to pin.
• Not a carrier/tail defect: diagnose_us_target_support shows 5,762 carriers, top-1 2.6%, top-5 5.4% (design +84.9% → final +29.5% = broad per-holder level excess).

Cross-family review

Two independent sol (gpt-5.6-sol, ultra) passes over the committed diff, read-only, worktree-scoped:

• Round 1 verdict: request changes — (1) Medium: no critical-register entry, a +30% miss could re-certify (fixed in commit 2); (2) Low: "carriers unchanged" comments overclaimed a data-contingent fact as a formula invariant (reworded in commit 2). Confirmed clean: materializer entity handling, base_variable consumers (mortgage rows can't take the person-entity age-band path), no missed old-concept consumer repo-wide, pe-us formula acyclic and computable on the target frame (79,729 finite values), regression test fails pre-diff.
• Independent reproduction in review: gross $241.269B / capped $214.205B / identical return-carrier totals on the certified artifacts.

Scope

• fiscal_targets.py measure/concept maps + comments; us_critical_targets.py register entries; test fixtures in both register mirrors. No build stages, no calibration knobs (Loss shape: flat cap at 1.0 makes past-cap rows free dumping grounds (17 rows pushed out in Build N); bounded-tail experiment + past-cap census diagnostic #492 doctrine).
• Local: populace-data suite green; test_us_fiscal_refresh_builder.py + test_us_fiscal_targets.py + test_us_plan.py + test_gates.py green (worktree sources via PYTHONPATH); ruff format/check clean.
• The JCT mortgage row stays unregistered here — the JCT-family register gap is Loss shape: flat cap at 1.0 makes past-cap rows free dumping grounds (17 rows pushed out in Build N); bounded-tail experiment + past-cap census diagnostic #492's step-3 flag, and the row's fix rides the same two concept legs (see Mortgage-interest dollars run ~25-30% hot (ht2 +29.5%, JCT expenditure +23.3% on certified O-1) — first-order for deduction-replacement analyses #511 verdict comment).

Fixes the target-side leg of #511. Relates #515 (donor concept), #516 (donor integrity hazard found in the same forensics), #299 (Build H mapping this supersedes).

🤖 Generated with Claude Code

[MaxGhenis]

Cross-family review record (sol, gpt-5.6-sol ultra, read-only, committed-diff scoped)

Round 1 (remap commit e667648): request changes.

• Medium/blocking: the corrected mortgage target had no critical-register entry — a +29.5%-class miss could still certify (the adoption-gate gap Mortgage-interest dollars run ~25-30% hot (ht2 +29.5%, JCT expenditure +23.3% on certified O-1) — first-order for deduction-replacement analyses #511 flagged). → Fixed in 67c98ee: both Table 2.1 mortgage rows registered by exact name (amount at interim 0.20 with a dated ratchet to US_CRITICAL_DEDUCTION_MAX_ABS_RELATIVE_ERROR once PUF donor home_mortgage_interest carries the E19200 total-interest concept (+7.6% vs the Table 2.1 mortgage-only leg) #515 lands; returns at 0.15; no incumbent-improvement escape), plus a builder-side regression pinning the exact shipped O-1 diagnostics failing the release gate.
• Low: "carriers unchanged" comments overclaimed a data-contingent fact as a formula invariant. → Reworded to the O-1 post-itemizer-mask verification.
• Confirmed clean: materializer entity handling; base_variable consumers (mortgage rows cannot take the person-entity age-band path); no missed old-concept consumer repo-wide; pe-us formula acyclic, computable on the target frame (79,729 finite values); the regression test fails on pre-diff code. Independent reproduction on the certified artifacts: gross $241.269B / capped $214.205B / identical weighted return-carrier totals (11,929,444.9), minimum deductible share 0.0835.

Round 2 (register commit 67c98ee): approve, three low-severity test-hardening suggestions, all taken in b083869:

1. Pin allow_incumbent_improvement=False inside the 0.20–0.25 band (+22.5% with an improving incumbent must still fail — the +30% case alone also trips the 0.25 hard stop and can't tell the flag regressed).
2. Returns requirement gets its own negative (+20% > 0.15) and removal (required-presence) regressions.
3. Builder O-1 regression asserts relative_error=0.294986 + exceeding 0.2 so the missing-row formatter (same row name + label) can't satisfy it.

• Verified: exact-name-only matching (semantic/pattern selectors can't fire); required-present semantics fail closed on both publisher and builder consumers, with no dense/sparse path where absence certifies (registry compiled once before the split); fixture family counts consistent in both mirrors; the passing arm of the builder regression is non-vacuous.

Local test state after b083869: populace-data suite green (1 unrelated skip), builder release-gate/critical battery green, full test_us_fiscal_refresh_builder.py + test_us_fiscal_targets.py + test_us_plan.py + test_gates.py green, ruff format/check clean.