# Minnesota ACA 2027

New, independent study at `site/analysis/minnesota-aca-2027.html`. The original
Minnesota ACA dashboard and its datasets are not inputs to economic conclusions;
its county FIPS identifiers and map geometry are reused and checked against CMS.

## Reproduce

Use Python with numpy for `compute.py`, then `build.py` for the study and deck.
The checked-in `inputs.json` is an extracted source snapshot; default computation
is offline. `compute.py --refresh --source '/Volumes/Elephruit/ACA Work - MN'`
also requires lxml, openpyxl, pypdf, the original filings, and downloaded public
documents in ignored `raw/`. Each extracted source file has a SHA-256 in the
snapshot. Do not commit raw filings or documents marked confidential/trade secret.

Run `python research/mn-aca-2027/test_compute.py` for count bridges, geographic
ownership, rates, model margins, structural zeros, and subsidy boundaries.

## Definitions

- Universe: individual medical plans listed in MNsure's official 2026/2027
  provider-network workbooks and also supported by supplied QHP templates.
  Excludes dental and off-exchange-only plans. Includes catastrophic in plan
  counts, but not Bronze/Silver/Gold rankings. CSR variants are not separate plans.
- A plan is one 14-character base HIOS ID. A county option is one available
  base-plan/county combination. RA unique counts are unions, not choices available
  to every resident. County minimum/maximum counts show this distinction.
- New means a 2027 ID absent in 2026, not necessarily a new economic design.
  Continuing means the same ID in both years. Not offered means the old ID is
  absent; replacement/crosswalk is reported separately from continued same-ID
  coverage. It does not imply uninsured people or a terminated customer relationship.
- Footprints use county service-area sets, not the presence of a rate in an RA.
  All observed on-exchange footprints are whole-county. The computation fails
  rather than silently rank a future partial-county service area at county level.
- LCB includes Expanded Bronze. SLCSP uses the second distinct base Silver plan,
  including an equal-priced plan; it need not be from a different issuer or parent.
  The runner-up gap and nearest competing-parent gap are different measures.
- Parent reach allows the winning parent to use different plans/prices within an
  RA. County-count coverage and modeled membership reach are shown separately.
- UCare is grouped with Medica in 2026 ownership because Medica acquired the
  relevant assets/contracts effective Jan 1, 2026. This is a commercial competitive
  grouping, not a claim about every legal corporate relationship. Quartz is jointly
  sponsored; no single sponsoring health system is labeled its sole parent.

## Enrollment and model limitations

Latest board source located in the public document library: July 29, 2026. It
reports 123,840 June effectuated enrollees and rounded carrier and metal shares.
Its RA distribution is labeled May 31, even though the slide appears in July.
County weights are observed private-plan enrollment from the 2025 annual report.
These are different vintages/populations. They are explicitly not observed
June county or county-by-carrier membership.

Normalize rounded RA and carrier margins (each sums to 99.9%). Apply RA weights
to the June total, distribute each RA using the observed 2025 county proportions,
then fit county-by-brand estimates to county and carrier margins with iterative
proportional fitting. Structural zeros come from 2026 service areas. Counts are
continuous estimates and round only for display; no plan-level allocation occurs.

Equal-share and statewide-carrier-share seeds converge to the same matrix: IPF is
invariant to column scaling. A third seed alternates 2x/0.5x RA/carrier affinities
before fitting to the same margins. The range across seeds is an illustrative
model stress range, not a confidence interval, identified estimate, or forecast.
Do not add carrier ranges to obtain a portfolio range. All carrier and county
totals reconcile in every scenario. 2027 modeled market shares are not forecast.

## Source conflicts retained

The 2026 MNsure workbook uses UCare Community Health Plan IDs beginning 97624;
the supplied UCare filing uses 85736 IDs. Fourteen plans match one-to-one on exact
name and metal. Nine also match county sets. Five M Health Fairview plans include
Aitkin in MNsure's workbook but not in the supplied service-area template. The
MNsure listing controls availability, with Aitkin's 2026 comparison marked
provisional. UCare's 2026 rates and benefits come from the pre-transfer filing;
post-transfer equivalence has not been independently verified. Retain both IDs.

The supplied Quartz 2027 XML crosswalk identifies Wisconsin issuer 37833, so it
is excluded. Same-ID continuations remain directly observable. UCare's brand-level
Medica transition is publicly confirmed, but exact successor IDs are unavailable.
All flags remain in the published methodology and data warnings.

## Subsidies

2027 single-adult FPL base $15,960; IRS Rev. Proc. 2026-26 contributes 10.22% at
300% FPL ($47,880, $407.78/month). APTC is benchmark less contribution, floored
at zero, and net premiums floor at zero. No APTC at 450% FPL. Exact 200% FPL is
within MinnesotaCare's published income limit, so no exchange APTC/net premium is
assumed there. Other eligibility requirements are assumed only in the 300% case.
Income-based CSR does not apply at 300% or 450%; CSR is separate from APTC.
No household-specific tax advice, utilization, claims, risk adjustment, margins,
provider participation equivalence, or future membership is inferred.

## Review and release

Commit source and generated public outputs together after relevant checks. Preserve
unrelated files. Build, OG, site checks, and deck audits precede preview handoff.
Firebase project `elephruit` deployment requires Mike's explicit preview approval.
