CHIP enrollment targets are concept-mismatched: CMS total_chip_enrollment includes M-CHIP; model materializes separate CHIP only (20 states zero-support)

Finding (Build F, 2026-07-06)

The dense certified-lineage release failed its zero-support gate on exactly 20 cms_medicaid.month2024_12.state_enrollment.{state}.total_chip_enrollment@2024 targets: AK, CA, DC, HI, IL, KY, MD, ME, MI, MN, NC, ND, NE, NH, NM, OH, OK, SC, VT, WY.

That state list is not a sampling artifact (CA is in it, on a 337,704-household base). It corresponds to the Medicaid-expansion CHIP (M-CHIP) states: CMS's total_chip_enrollment counts M-CHIP + separate CHIP, while the model's materialized enrollment concept produces separate-CHIP-style membership only — so M-CHIP-dominated states materialize zero support for a positive target. A target-concept vs model-concept mismatch, not missing data.

Two consequences:

1. The remaining ~30 states' CHIP targets are ALSO conceptually mixed (their totals include an M-CHIP share the model doesn't represent), so calibration over-pulls child weights there to hit an unreachable-by-concept total. The zero-support gate only catches the pure-M-CHIP extreme.
2. Build F proceeded with a documented feed supersession (v6 = v5 minus exactly these 20 rows, sha 9f8edbb9...), which unblocks the gate but does not fix the concept.

Ask

Map the CHIP target family onto concepts the model actually materializes: either split CMS enrollment into M-CHIP vs separate-CHIP components at the ledger level (calibrating each against its true model concept), or add an M-CHIP-inclusive enrollment concept to the materialization. Related: #233 (Medicaid/CHIP spending targets), #170 (eligibility-to-enrollment diagnostics). Same disease as populace#320's capital-gains finding: targets and columns that encode program-structure concepts must be mapped explicitly, never assumed 1:1.

Start by tracing the CHIP target ledger, enrollment materialization, and calibration flow, then review related issues #233 and #170 for existing assumptions. Done means CMS M-CHIP and separate-CHIP targets are mapped to matching model concepts, rather than relying on the superseded feed that removes 20 rows.