Liquid assets sit on the lowest-line person, against the stage's declared is_household_head carry; 13.1M SSI-relevant adults test at $0 resources by construction #1215

What is wrong

In the certified US frame (populace-us-2024-spm-20260909, H5 SHA-256 6496cc43…; Build P cae8640), each household's liquid assets (bank_account_assets, stock_assets, bond_assets) sit entirely on one person, the household's lowest-A_LINENO member. All 52,637 rows with a nonzero leaf are that person; every other member holds $0.

Two problems follow.

1. The selector contradicts the stage's own declaration. us/source_stages.json (:1135–1145) declares the carry as head_carry with condition: is_household_head. The runtime (scf_wealth.py _household_head_mask, :670–691 at cae8640, :673–694 on main; SIPP's equivalent in sipp_financial_assets.py:561–577) sorts by household, A_LINENO (missing → 9999) and row order and takes the first. is_household_head is P_SEQ == 1 (relationship_inputs.py). They disagree in 243 households. The docstring's "carried by the CPS reference person" holds only where line 1 is the reference person: 370 carriers have A_EXPRRP outside {1, 2} (for the 2022 and 2023 income years the loader synthesizes A_EXPRRP, assigning 1/2 only to line 1), and 123 carriers are under 18. In household 884, a 71-year-old wife on line 1 carries $17,500 while the line-2 reference person carries nothing.

2. One vector per household, on one person, for an engine that reads person ownership. The SCF draw describes a primary economic unit (the dominant person or couple and their financially dependent members). The SIPP leaves read TVAL_BANK, TVAL_STMF and TVAL_BOND, which are person-level values (THVAL_* are the household totals). Both are drawn once per household, conditioned on the carrier's characteristics, and placed on the carrier. PolicyEngine-US's ssi_countable_resources adds exactly these three leaves, and meets_ssi_resource_test uses the marital unit's sum for a joint claim and the person's own value otherwise (ssi_claim_is_joint requires two ABD heads or spouses in the marital unit). So everyone outside the carrier's marital unit has $0 countable resources by construction:

They include 1.15M people 65+ coded as the reference person's parent, all living with an adult-child carrier, and 5.64M SSI-relevant adult children of the reference person. A rough income screen (annual Social Security plus earnings under $12,336 = 12 × ($943 + $20 + $65)) leaves 6.41M SSI-relevant adults outside the carrier's unit, 3.41M of them in households whose vector exceeds $2,000. These are exposure counts. Whether a given person's SSI eligibility changes depends on rules this screen omits, and on who in the household owns what.

Within a marital unit, a claimant whose spouse is not aged, blind or disabled files an individual claim. The PolicyEngine-US releases pinned with this frame (1.764.6 at build, 2.2.1 in policyengine.py's bundle) count only the claimant's own resources then, so a carrier spouse leaves the claimant at $0. PolicyEngine-US main now deems an ineligible spouse's resources, and parents' resources to children under 18 (policyengine-us#10040, merged 2026-10-11; 20 CFR 416.1202). No rule deems resources between adults outside a marital unit, so the adult children, parents and other relatives counted above test at $0 under every engine version.

Proposed fix

• Now, small: make the runtime honor its declaration. Select the carrier with is_household_head, or change the declaration to the rule actually intended, and add a test that the declared condition and the runtime selector agree. This moves 243 households' vectors. It does not reach the ownership problem, and 104 is_household_head persons are still under 18.
• The real fix needs consistent donor and recipient units: draw SIPP person-level leaves per adult (with spouse dependence), or draw per marital unit and allocate within it; map SCF's economic unit to CPS economic units and give independent adults their own draw. An SPM-unit draw alone would leave 12.15M of the 13.06M exposed SSI-relevant adults sharing the carrier's unit, so allocation within the unit is still needed. INT_VAL and DIV_VAL can guide ownership (4.22M exposed SSI-relevant adults report interest or dividends while holding $0 in the frame) but do not determine balances. Changing the draw unit changes aggregate assets, so it needs validation against SCF and SIPP totals, and a refit.

#1184 (imputed savings put SSI reporters over the limit) is a related symptom of the same stage; #1216 covers the stage's missing provenance record.

Downstream

Every SSI resource-test result on this frame: microcosm-dynamics Registration 19 and 20, the SSI asset-limit analyses (PolicyEngine/ssi-asset-limits), and policyengine.org SSI results through PolicyEngine-US's default dataset. The bias runs toward people passing the resource test, so asset-limit reforms add fewer new eligibles among adults living with relatives than they would with assets placed on owners.

Reproduce

The carrier fact: on the person table, compare A_LINENO == min(A_LINENO) per household with the nonzero-leaf mask and with is_household_head. Exposure: people with age >= 18 whose person_marital_unit_id differs from the carrier's, weighted by household_weight.

Found by read-only structural reads of the frame while registering microcosm-dynamics Registration 19.