Ingest ASEC 2024 and 2025 into the US support pool (spine is two vintages behind)

Problem

The US support spine pools raw Census ASEC vintages via build_pooled_asec_unit_frame (tools/build_us_asec_pooled_source_base.py takes --asec-h5 inputs), and the available ingested inputs stop at census_cps_2023.h5 (ASEC 2023, income year 2022). Census has since published:

• ASEC 2024 (asecpub24csv.zip, income year 2023)
• ASEC 2025 (asecpub25csv.zip, income year 2024) — released fall 2025, current latest

Production populace-us builds are therefore two survey vintages behind the newest available microdata.

Proposal

Ingest ASEC 2024 and 2025 into the raw-source lane (census_cps_2024.h5, census_cps_2025.h5) and move the pooled support spine's default year set forward so certified bundles build from the newest vintages (e.g. 2023–2025 pooled), with the usual population-share scaling.

Notes

• The imputation-paper repo (github.com/PolicyEngine/imputation-paper) now reads ASEC 2024/2025 CSVs directly from census.gov (src/imputation_paper/data/cps.py); its loader documents the file conventions (PH_SEQ/H_SEQ join, two implied decimals on MARSUPWT/HSUP_WGT, zero-weight shell records in hhpub) and both bundles' column names verified compatible — useful as a reference for the ingestion.
• Fixed sha256s: asecpub24csv.zip begins cdb39cda; column sets verified identical for the fields populace consumes.

🤖 Generated with Claude Code

Start with tools/build_us_asec_pooled_source_base.py and the build_pooled_asec_unit_frame entry point, then review the imputation-paper loader for ASEC 2024/2025 file conventions and joins. Ingest census_cps_2024.h5 and census_cps_2025.h5, update the pooled default year set, and verify that certified bundles build with the expected population-share scaling.