American Health Security Act — Structural Audit, August 2026
Scope: _includes/technical/health-security-act.md (EN + ES),
_includes/wtp/health-security-act.md (EN + ES), cross-referenced against
_data/figures.json, sections/06-fiscal-analysis.md, and audits/2026-06-economic-audit.md.
Method: Same adversarial audit applied to the Equal Opportunity in Education Act in
reviews/us-education-engineered-ignorance-review.md §6 — look for mechanisms the Act reproduces
while criticizing, precedents cited that undercut the design they support, causal or fiscal claims
asserted without support, and internal contradictions a hostile scorer would find first.
Status: all findings remediated (August 2026). Every finding below records its Change Applied block with before → after values. Following the convention of the June 2026 economic audit, this document is published now that its findings are resolved.
Executive Summary
The Act’s policy architecture is sound and in several respects unusually well drafted — the explicit design-option structure (Option A/B/C with a stated recommendation and tradeoffs) is better legislative reasoning than most of the mandate. The problems are not in the concept. They are in three places:
- The Act’s Cost Analysis section is an orphan. Its totals match neither
figures.jsonnorsections/06-fiscal-analysis.md, and never have — including before the June 2026 audit that claimed to cascade every dependent total to this page. - The single largest design risk to a public option — adverse selection — is not addressed anywhere in the vault, while the Act’s own market-impact section describes the mechanism that produces it.
- Two precedents are cited in support of provisions they actually undercut, one of which is a misstatement of a holding.
Finding 1 is the most urgent because it is the most checkable. Anyone comparing the Act’s stated net cost to the mandate’s own fiscal analysis finds a factor-of-two discrepancy in under a minute.
1. The Cost Analysis Section Is Desynchronized From the Fiscal Model
The Act computes its own healthcare totals from its own offsets list. Those numbers agree with nothing else in the mandate.
| Quantity | Health Security Act | figures.json / Fiscal Analysis |
|---|---|---|
| Public option net cost, steady state | $74-104B | $150-250B (public_option_cost) |
| Administrative savings | $50-100B | $100-180B (admin_savings) |
| System-wide healthcare savings | $270-430B | $400-570B (total_healthcare_savings) |
| Net healthcare savings | $166-356B | $150-320B (net_healthcare_impact) |
| Drug negotiation savings | ”>$250B” | $250-300B (drug_savings) |
The public-option cost discrepancy is the serious one: the Act states a net federal cost roughly half the figure the mandate’s fiscal model uses to prove the program is affordable. Both numbers are load-bearing and they are used in the same argument.
This is not fallout from the June 2026 audit — it predates it. That audit moved total healthcare savings from $350-570B to $400-570B and net healthcare impact from $100-320B to $150-320B, and lists “Health Security Act (EN + ES)” among pages updated. But the Act’s Cost Analysis block reads $270-430B and $166-356B — matching neither the pre-audit nor the post-audit values. The June cascade found and updated the drug-savings sentence at line 220; it did not touch the derived totals five lines below at lines 223-226. The block has been computing independently for longer than that.
Two further defects inside the block:
- The offsets do not sum to the stated total. Listed: uncompensated care $20-30B, private premium reductions $50-100B, drug negotiation >$250B, administrative simplification $50-100B. That sums to $370-480B, not the stated $270-430B.
- Two of the four offsets are annotated “(included in earlier expenditure analysis).” If those savings are already counted in the mandate’s fiscal model, listing them here as additional offsets double-counts them; if they are not, the annotation is wrong. As written it is impossible to tell which, and the annotation appears on precisely the two largest items.
Why check_figures.py did not catch any of this. The validator checks tracked tokens and
derived-total consistency inside figures.json. The Act writes its cost figures as free text, so
they are invisible to it. The validator passing is not evidence this section is correct — a
structural limitation worth recording, and the direct analogue of check_liquid.py catching build
breakage but not EN/ES semantic drift.
One inherited arithmetic slip. In sections/06-fiscal-analysis.md, drug savings components
($150-200B Medicare negotiation + $50-100B Medicaid/public-option leverage) sum to $200-300B,
but the stated total is $250-300B. The low end is overstated by $50B, and figures.json tracks
the stated total rather than the component sum. This propagates into total_healthcare_savings and
every figure derived from it.
✅ Change Applied (August 2026)
The Act’s Cost Analysis block now uses the tracked figures. Before → after, in
_includes/technical/health-security-act.md (EN + ES):
| Figure | Before | After |
|---|---|---|
| Public option net cost (headline) | $74-104B | $150-250B (public_option_cost) |
| System-wide healthcare savings | $270-430B | $400-570B (total_healthcare_savings) |
| Net healthcare savings | $166-356B | $150-320B (net_healthcare_impact) |
| Administrative savings | $50-100B | $100-180B (admin_savings) |
| Drug negotiation savings | ”>$250B” | $250-300B (drug_savings) |
| Preventive care savings | absent | $50-90B (preventive_savings) |
The offsets list was restructured into two groups so it sums correctly and the double-counting ambiguity is resolved. The three tracked components (drug, administrative, preventive) now sum exactly to $400-570B. Uncompensated care ($20-30B) and private premium reductions ($50-100B) are retained but moved below the total and explicitly labeled as excluded from it — the first accrues to hospitals rather than the federal budget, and the second is partly the same administrative efficiency already counted.
The bottom-up enrollment scenarios are retained, with an added reconciliation note stating that they produce a lower net cost ($74-104B at 40M enrollees) than the planning figure because they assume higher premium recovery and hold per-enrollee cost flat against medical inflation, and that the mandate uses the conservative $150-250B in all fiscal totals. The scenarios were kept rather than deleted because their arithmetic is transparent and they demonstrate the sensitivity of net cost to premium recovery and enrollee mix — which is directly relevant to Finding 2.
The fiscal-conservative rebuttal in the Political Strategy section, which repeated the stale “net cost <$100B offset by system-wide savings $270-430B,” now states the tracked figures.
Inherited slip corrected upstream. In sections/06-fiscal-analysis.md (EN + ES), the drug
savings components summed to $200-300B against a stated total of $250-300B. Rather than lower the
total — which would have reversed the June 2026 audit and cascaded through figures.json — the
Medicare negotiation component was raised from “$150-200B” to “$200B” with the February 2026 CBO
baseline revision stated as the reason, which is the rationale the June audit already recorded.
Components now sum exactly to the stated total. _data/figures.json was not modified and no
downstream figure moved.
Not changed (intentionally): project-2029-original.md, project-update-plan.md, and
HANDOFF.md still contain the superseded values. All three are excluded from the site build and
serve as historical records; editing them would falsify the archive.
2. Adverse Selection: The Missing Mechanism
adverse selection, risk adjustment, risk corridor, reinsurance, and risk pool appear
nowhere in the vault. For a public option, this is the equivalent of the education act’s
unexamined testing regime: the mechanism most likely to defeat the policy is the one not named.
The design creates the conditions directly. The public option offers guaranteed issue, no medical underwriting, no pre-existing condition exclusions, and income-based premiums. Private plans competing alongside it retain latitude over benefit design, network composition, and marketing — all standard instruments for attracting healthier enrollees. Nothing in the Act specifies risk adjustment between the public option and private plans.
The Act describes the mechanism without recognizing it. Its Market Competition Effects section predicts that “some private insurers may exit individual market” and that private insurers “may focus on supplemental coverage (dental, vision) or premium employer plans.” That is a description of private carriers retaining the profitable segments and ceding the rest — which is what adverse selection into the public option looks like from the other side.
The cost model assumes it away. The $5,500 average cost per enrollee is an average-risk assumption. If the public option draws a sicker-than-average pool, per-enrollee cost rises, premiums follow, healthier enrollees leave, and the cycle repeats. This is the documented failure path of the ACA CO-OPs, and it is the standard objection to public-option proposals in the health policy literature. The Act’s Opposition and Counterarguments section anticipates “government takeover,” “Medicare rates too low,” and “unaffordable” — but not this, which is the strongest technical objection it will face.
Remediable. Standard instruments exist: risk adjustment across public and private plans in the same market, a permanent reinsurance program for high-cost enrollees, and uniform benefit-tier definitions limiting benefit-design cherry-picking. The gap is that none are specified, not that none are available.
✅ Change Applied (August 2026)
New Risk Pooling and Adverse Selection Protection section added ahead of the Cost Analysis, where the average-risk assumption becomes load-bearing. Three mechanisms, all with direct ACA precedent:
- Risk adjustment across all plans in each market, public and private, on the §1343 model. Transfers net to zero — a reallocation, not an outlay. Includes a coding-intensity correction normalizing risk scores, which is the specific defect that made Medicare Advantage more expensive than traditional Medicare and which §3.2 identifies.
- Reinsurance above an attachment point, funded by a per-covered-life assessment on all
health plans on the §1341 model rather than by appropriation. Estimated $10-15B annually and
accommodated within the tracked $150-250B planning figure, which already sits well above the
bottom-up scenario.
public_option_costand every figure derived from it are unchanged. - Benefit standardization within metal tiers, plus marketing conduct standards, to limit benefit-design cherry-picking.
Risk corridors were deliberately excluded, with the reasoning stated in the Act: the ACA’s §1342 corridors were enacted without a permanent appropriation, Congress later restricted payment, and Maine Community Health Options v. United States (2020) held the government liable for roughly $12B it had declined to pay. A risk-sharing promise Congress can decline to fund creates unbudgeted liability without delivering stability.
This finding was expected to carry appropriations and cascade through the fiscal model. The assessment-funded design avoids that — the policy gap is closed at no net federal cost.
3. Two Precedents That Undercut the Provisions They Support
Structurally identical to the education act’s citation of No Child Left Behind as precedent for a federal lever whose design it then reproduced.
3.1 NFIB v. Sebelius is misstated — and it is the wrong case to cite here
The Constitutional Authority paragraph states that “the ACA’s individual mandate and Medicaid expansion [were] specifically upheld in NFIB v. Sebelius (2012) as exercises of the Taxing and Spending powers.”
The individual mandate was upheld as a tax. The Medicaid expansion is the portion of the ACA that lost. The Court held that conditioning existing Medicaid funds on acceptance of the expansion was unconstitutionally coercive under the Spending Clause, which severed the enforcement mechanism and made expansion effectively optional. That holding is the entire reason holdout states exist.
This is not a pedantic correction, because the Act’s Key Provisions promise to “provide incentives for all states to expand Medicaid” — a design that must be built around the NFIB limit on coercive conditions. The Act cites as supporting authority the case that constrains the provision most, and states its holding backwards. A single opposing brief makes this expensive.
3.2 Medicare Advantage proves the political point and undercuts the fiscal one
The rebuttal to “government takeover, kills private insurance” is that “private insurance thrives in Medicare Advantage market (30%+ of seniors).”
- The figure is stale. MA enrollment passed 50% of Medicare beneficiaries in 2023. “>30%” is technically true and rhetorically weak.
- The substance cuts the wrong way. MedPAC has repeatedly found that MA costs Medicare more per beneficiary than traditional Medicare would for comparable enrollees, driven by favorable selection and coding intensity. MA is the leading contemporary example of private plans extracting margin from a public program — which is the phenomenon the Act’s own Rationale attacks two pages earlier when it says insurers “profit from denying care.”
The analogy works for the political claim (private insurance survives public competition) and argues against the fiscal claim (public competition reduces system cost). It should be narrowed explicitly to the former, or replaced.
3.3 A weaker precedent problem
Hospital price regulation in non-Medicare markets is justified by analogy to the CMS Hospital Price Transparency Rule (2021). Disclosure is not rate-setting. A rule requiring hospitals to publish prices is thin precedent for a rule capping them, and rate regulation in commercial markets is the Act’s most legally exposed provision. It needs its own authority argument — the Medicare prospective payment system, also cited, is much closer and should carry the weight.
✅ Change Applied (August 2026)
- NFIB corrected. The Constitutional Authority paragraph now states that the individual mandate was upheld as an exercise of the taxing power, and that the same decision held coercive conditioning of existing Medicaid funds unconstitutional — with an added note that this Act’s Medicaid provisions are structured as new-money inducements that do not put existing federal funding at risk.
- Medicare Advantage narrowed. The rebuttal now cites MA at its correct scale (more than half of Medicare beneficiaries) and states explicitly that the analogy is offered for the political claim only — noting MedPAC’s finding that MA costs more per beneficiary than traditional Medicare, and cross-referencing the coding-intensity correction added in §2.
- Hospital rate authority regrounded. Rate-setting now rests primarily on the Medicare prospective payment system (administered rates since 1983). The Transparency Rule is retained only for the narrower disclosure proposition, with “disclosure is not rate-setting” stated in the text, and the provision is flagged as the Act’s most legally exposed.
4. Internal Contradictions
4.1 The premium table does not work. The columns are “Monthly Premium (Individual)” and “Annual Premium (Family of 4).” At 138-200% FPL an individual pays $50/month — $600/year — while a family of four pays $200 annually. At >500% FPL an individual pays $5,400/year against a family of four at $1,600/year. Read as annual, families pay a fraction of individuals at every tier. Read as monthly, the family-to-individual ratio is a plausible 3.4-4×, which strongly suggests the column is mislabeled. Either way the “% of Income” column tracks only the individual figures.
4.2 Benefit phase-in contradicts the timeline. The benefits section recommends phasing in dental/vision/hearing “over Years 2-3.” The Implementation Timeline places “expand benefits if fiscally sustainable (add dental/vision)” in Years 4-5 — and puts launch itself in Year 3, meaning the recommended phase-in would begin before coverage exists.
4.3 Two different “Year 1”s. The Implementation Timeline runs Days 1-180, Months 6-12, Year 2 buildout, Year 3 launch. The enrollment and cost projections are labeled Year 1 / Year 3 / Year 5 with 15M enrollees in Year 1. These cannot both be the same Year 1. Under the timeline, the Year 5 steady state falls in roughly Year 7 of the administration — beyond a first term, in a mandate organized around a 180-day and first-term frame. The two clocks need distinct labels.
4.4 The plain-language version promises more than the technical version delivers. The We The People summary promises “no surprise billing, no networks, no prior authorizations for medically necessary care.” The technical Act recommends Option A, the Medicare provider network, offers a negotiated-network fallback that it concedes carries “potential network gaps in rural areas,” and never mentions prior authorization at all. Balance billing is genuinely prohibited, so “no surprise billing” holds; the other two commitments are not in the design.
4.5 Medicare rate framing is inconsistent. Baseline says Medicare rates are “~40% lower than private insurance average”; the rationale two lines later says “private pays Medicare+40-60%.” If private equals 1.4-1.6× Medicare, Medicare is 29-38% below private, not 40%. The derived claim that Medicare+15% lands “~25% lower than private” holds only at the optimistic end of that range (17.9% at Medicare+40%, 28% at Medicare+60%).
✅ Change Applied (August 2026)
- Premium table rebuilt. Percentage of income is now the governing parameter on the ACA applicable-percentage model, with dollar figures labelled as illustrative monthly amounts at band midpoints. The mislabelled “Annual (Family of 4)” column is gone. A note records why the fixed dollar cap was abandoned: it made the effective contribution rate fall as income rose above the cap — regressive, and inconsistent with the Act’s own stated design principle. The cost-sharing line now reads “0% below 138% FPL rising to 8.5% above 500% FPL,” matching the schedule.
- Benefit phase-in reconciled. Dental, vision, and hearing now phase in during Operating Years 2-3 (Administration Years 4-5), consistent with the Implementation Timeline.
- Two clocks disambiguated. Enrollment and cost figures are labelled Operating Years; the timeline uses Administration Years. A callout states the mapping and concedes that Operating Year 5 falls in Administration Year 7 — beyond a single term.
- Plain-language over-promise removed. “No networks, no prior authorizations” is replaced with what the design actually delivers: a nationwide Medicare-based network, and prior authorization limited to a narrow published list. “No surprise billing” is retained because balance billing is genuinely prohibited.
- Medicare rate framing made consistent. Both statements now derive from the same premise: commercial pays Medicare+40-60%, so Medicare sits 29-38% below commercial, and Medicare+15% lands 18-28% below commercial.
5. Modeling Assumptions That Do Not Survive Scrutiny
5.1 Zero medical inflation across five years. Cost per enrollee is held at $5,500 from Year 1 through Year 5. At 4-5% annual medical cost growth, Year 5 per-enrollee cost is ~20-27% higher, adding roughly $45-60B to Year 5 medical costs — larger than the entire stated Year 5 net federal cost range of $74-104B. The internal arithmetic of each block is correct; the assumption feeding it is not.
5.2 Unexplained premium drift. Premiums collected per enrollee rise from ~$2,000-2,667 in Year 1 to ~$3,000-3,750 in Year 5 — roughly 50% — with no stated cause. A shift toward higher-income employer-group enrollees would explain it, and is plausible given the projected employer adoption, but it is unstated. This assumption is doing significant work holding net federal cost down, and it runs in the opposite direction from the adverse-selection risk in §2. Both cannot be true.
✅ Change Applied (August 2026)
A Stated assumptions callout now precedes the cost projections, disclosing that $5,500 per enrollee is held in constant 2026 dollars, that medical inflation at 4-5% would add roughly $45-60B to Operating Year 5, and that rising premiums per enrollee reflect an assumed shift toward higher-income employer-group members. The callout states plainly that both assumptions run optimistic and the second runs opposite to the adverse-selection risk, and that they are the principal reason these scenarios undercut the $150-250B planning figure the mandate uses.
6. Figures Requiring Sourcing or Routing
None of these are tracked in _data/figures.json:
| Figure | Issue |
|---|---|
| “30 million Americans uninsured” | No date or source. Also appears in project-2029-claude.md. Route through figures.json |
| “nearly twice as much per capita as any other developed country” | True against the OECD average; against the next-highest spenders the multiple is closer to 1.5×. “Any other” overstates |
| MA enrollment “30%+ of seniors” | Stale; passed 50% in 2023 |
| Employer premiums “$650/month individual, $1,800/month family (2024)” | Ambiguous whether these are total premiums or employee contributions — the two differ by roughly 3-4× for family coverage. As written the comparison to public-option premiums cannot be evaluated |
| Medicare rates “~40% lower” | Inconsistent with the Act’s own Medicare+40-60% framing (§4.5) |
✅ Change Applied (August 2026)
- Per-capita spending reframed as “roughly twice the OECD average — and about 1.5 times the next-highest spenders,” replacing “nearly twice as much as any other developed country.”
- Uninsured now given as “roughly 26-30 million (Census/NHIS range, rising as enhanced ACA subsidies lapse)” rather than a bare 30 million.
- MA enrollment corrected to more than half of beneficiaries.
- Employer premiums now distinguish total premium (~$750/mo individual, ~$2,100/mo family) from the employee’s own contribution (~$115 and ~$525), and the comparison claim is corrected accordingly: 30-50% lower against total premium, but not automatically cheaper at higher incomes than an employee’s share of a subsidized employer plan. The honest claim — dramatically cheaper for the uninsured, self-employed, and individual market; competitive rather than uniformly superior for those already holding subsidized employer coverage — now appears in the text.
- Medicare rate percentage resolved in §4.
These remain free-text rather than figures.json tokens because they are contextual comparisons
rather than figures the fiscal model derives from. The tracked fiscal figures are unchanged.
7. What Is Sound
Recording this so the audit is not read as a verdict on the Act as a whole:
- Constitutional grounding is mostly solid. Medicare as the model for public health insurance, IRA authority for drug negotiation, and Commerce Clause authority over interstate insurance markets are all well founded. The defects are the two specific citations in §3.
- The design-option structure is genuinely good drafting. Presenting Options A/B/C with tradeoffs and a stated recommendation, rather than asserting a single design, is more honest and more useful to a legislative drafter than most of the mandate’s provisions.
- Internal arithmetic is clean within the enrollment and cost blocks — enrollment × per-capita cost, 2% administrative overhead, and net-of-premium figures all reconcile. The errors are in the assumptions and in reconciliation with other documents, not in the calculations.
- Balance billing prohibition with CMS exclusion authority is a real enforcement mechanism, not an aspiration.
- Automatic enrollment for $0-premium-eligible individuals is the single highest-leverage coverage provision in the Act and is easy to overlook.
- EN/ES parity holds — 23 bold sections, 7 table rows, 44 bulleted provisions on both sides. Every defect above propagates identically to the Spanish version and must be fixed in both.
Recommended Remediation Sequence
Phase 1 — Reconcile the numbers (highest urgency, lowest difficulty). Replace the Act’s
orphaned Cost Analysis totals with the tracked figures, or state explicitly why the Act’s scope
differs from the fiscal model’s and reconcile the difference in the text. Fix the offsets sum,
resolve the double-counting annotation, and correct the $50B drug-savings low-end slip in
sections/06-fiscal-analysis.md — noting that this last one moves figures.json and cascades
through total_healthcare_savings, net_healthcare_impact, and the deficit figures derived from
them.
Phase 2 — Add the adverse-selection provision. Risk adjustment across public and private plans,
reinsurance for high-cost enrollees, uniform benefit-tier definitions. This is a new subsection, and
unlike the education act’s guardrails it is not cost-free — reinsurance carries real
appropriations and will move public_option_cost and everything derived from it.
Phase 3 — Fix the citations. Correct the NFIB characterization, narrow the Medicare Advantage analogy to the political claim, and give hospital rate regulation its own authority argument grounded in the Medicare prospective payment system.
Phase 4 — Resolve the internal contradictions. Premium table column labels, benefit phase-in timing, the two competing Year 1s, and the plain-language over-promise on networks and prior authorization.
Phase 5 — Assumptions and figures. Add a medical-inflation trend to the cost projection, state the enrollee-mix assumption behind premium drift, and route the untracked figures.
Phases 1, 3, and 4 are corrections. Phase 2 is new policy with fiscal consequences and should not be bundled with them.
Related
reviews/us-education-engineered-ignorance-review.md— the audit method applied hereaudits/2026-06-economic-audit.md— the June cascade whose scope this audit revisessections/06-fiscal-analysis.md— the master fiscal model_data/figures.json— tracked figures and derived totals