SIPP+SCF blend + SSI-aware refit: land the asset magnitudes for the Build J re-certification

[MaxGhenis]

Follow-up to #373 (SCF wealth stage): reach the dense-native SSI magnitude before the Build J re-certification flip.

Two steps, per the #373 probe diagnosis:

1. SIPP blend: the retired pipeline mixed SIPP and SCF financial-asset donors at FINANCIAL_ASSET_SOURCE_SCF_PROBABILITY = 0.5 (retired asset_imputation.py) — SIPP carries realistic low liquid assets; SCF-only overstates the SSI-relevant left tail. Port the SIPP donor half.
2. SSI-aware refit: the +$9.66B probe ran as a post-hoc overlay on weights solved without assets (the Sparse-57k certified release drops SSI asset input columns; SSI resource-test reforms score $0 on bundle 4.18.8 #356-documented +$9-26B overlay band). The certified rebuild must re-solve weights with the asset columns present (and an SSI baseline target if adopted), which the Sparse-57k certified release drops SSI asset input columns; SSI resource-test reforms score $0 on bundle 4.18.8 #356 dense-native record shows lands $10k/$20k at ~+$1.60B.

Acceptance: the #369 reform-coverage smoke stays green AND the $10k/$20k delta lands in the dense-native reference class on the re-certified artifact.

Refs: #368, #373, #356, #369.

[MaxGhenis]

Measured data points for this issue: full scenario runs on the certified Build J default (populace-us-2024-buildj-sparse-rmloss100-75d5add, bundle 4.20.4, pe-us 1.764.6), TY2026, same harness as the #356 record.

run	baseline SSI	$10k/$20k	remove test
Build J native (SCF-only, no refit)	6.71M / $57.1B	+$5.69B / +797k	+$18.19B / +2.45M
dense f0af251 native (SIPP-era, in-build)	8.05M / $63.6B	+$1.60B / +499k	+$16.05B / +3.23M
ECPS published (SIPP-era)	6.84M	—	+$15.75B / +1.77M
ECPS SIPP-2024 rebuild (pe-us 1.706)	6.11M	+$4.87B / +545k	—
external anchors	SSA ≈7.4M / $61B	OACT ≈$1.0B / CBPP 224k	—

Observations for the two workstreams:

1. Refit (step 2): the baseline undershoot predicted on Sparse-57k certified release drops SSI asset input columns; SSI resource-test reforms score $0 on bundle 4.18.8 #356 materialized — with the SCF columns present but weights solved asset-less, baseline SSI lands at 6.71M/$57.1B vs SSA ≈7.4M/$61B (the incumbent asset-less artifacts sat right on SSA because the solve absorbed the missing test). Re-solving with assets present (± an SSI baseline target) should recover most of this.
2. Blend (step 1): my in-situ $10k/$20k measurement (+$5.69B) is consistent with the gate probe (+$5.15B; small config/vintage differences expected). One nuance on the acceptance bar: the SIPP-era estimates themselves span a range — dense-native +$1.60B but the ECPS SIPP-2024 rebuild scored +$4.87B — so "dense-native reference class" is the tight end of the SIPP family. The external anchors (OACT $1.0B, CBPP 224k) sit closest to dense-native. The full-removal scenario is much more robust across asset vintages (+$16–18B on every credible variant) — the $10k/$20k band mass ($2k–$20k) is what the blend actually decides.

Resource-bin, CBPP age-bridge, and OACT/CBPP benchmark tables for the Build J run are reproducible via ssi-asset-limits .local-results/populace-rerun-scripts/ (variant buildj_, plain managed_microsimulation() on 4.20.4).