cd issue_18

# Download the CMS Medicare Inpatient IPPS PUFs DY2013-DY2024; build the fixed-weight
# Fisher case-mix migration index and the FFS HCC risk-score counterfactual; apply the
# Crespin coding-intensity share to the operating-netted IPPS pool; print the headline
python 01_build_data.py

# Generate all five charts plus hero
python generate_all_charts.py

Key outputs:

• results/savings_estimate.json — Booked $1.04B, range $0.73–$5.15B, waterfall components, recoverability and overlap assumptions
• results/drg_complexity_trend.csv — Fixed-weight Fisher/Laspeyres/Paasche case-mix migration index by year
• results/risk_score_counterfactual.csv — FFS HCC risk-score index by year
• results/ffs_ipps_pool_panel.csv — FFS IPPS pool by year, full and operating-netted
• results/mechanism_specific_upcoding.csv — Top-tier (MCC) share by condition family, 2013 vs. 2024
• results/ma_selection_test.csv — FFS-vs-MA enrollment shift and the selection counter-test
• results/fca_settlement_panel.csv — DOJ FCA confirmed hospital DRG-upcoding settlements 2018–2025
• results/savings_by_component.csv — Component build (pool, Crespin share, recoverability, overlap)
• results/methodology.md — Full methodology, ceiling/floor construction, and the patient-level data-partner ask

• Case-mix migration: national fixed-weight index, each DRG's relative weight held at DY2024 value; Fisher (geometric mean of 2013-anchor Laspeyres +12.3% and 2024-anchor Paasche +18.9%) = +15.5% over 2013–2024, +9.9% over the pre-COVID 2013–2019 window
• Acuity counterfactual: FFS beneficiary-weighted HCC risk score rose +7.8%; the HCC score is itself partly coding-driven, so the risk-net method over-attributes the residual to coding and is used only as the ceiling
• Pool: full DY2024 FFS IPPS = $107.04B; operating-netted (removing ~11% non-scaling outlier + capital add-ons; IME and DSH retained as proportional) = $95.27B
• Central: Crespin 2.0% × $95.27B = $1.91B gross × 0.55 recoverability − $3M FCA overlap = $1.04B booked
• Floor $0.73B (Crespin CI-low 1.7% × pool × 0.45); ceiling $5.15B (Fisher risk-net residual 7.2% × pool × 0.75, over-attributing by construction)
• Selection counter (MA enrollment doubled, FFS share 71.8% → 49.9%): Crespin's patient-level controls absorb it; FFS average HCC risk score actually fell 2.6% 2019–2024, so selection did not inflate measured FFS acuity
• Overlap: Issue #11 MA coding intensity = zero (insurer-side, different program); Issue #5 admin = zero; Issue #15 facility fee = zero (outpatient HCPCS vs. inpatient MS-DRG); Issue #10 Procedure Mill = additive; Issue #24 confirmed fraud netting = negligible

cd issue_17

# Download the CMS Part B Spending by Drug PUF CY2023; build the 597-drug universe
# (excluding skin substitutes, vaccines, immune globulins); apply the ex-manufacturer
# ratio with biosimilar zeroing, 340B netting, and recoverability; print the headline
python 01_build_data.py

# Generate all five charts plus hero
python generate_all_charts.py

Key outputs:

• results/savings_estimate.json — Booked $6.17B, range $2.85–$14.06B, waterfall components, ratio/340B/recoverability assumptions
• results/universe_part_b_drugs.csv — The 597-drug Part B universe with per-drug spend, gap fraction, and booked savings
• results/exclusions_accounting.csv — Skin substitute, vaccine, immune globulin, biosimilar, and generic exclusions applied
• results/manufacturer_rollup.csv — Booked savings by manufacturer (Merck, Regeneron, BMS, Roche)
• results/sensitivity_grid.csv — Conservative/central/aggressive scenarios across ratio, 340B share, and recoverability
• results/cross_validation.csv — Against MedPAC/ASPE ex-manufacturer ratio and MFN/GLOBE scoring
• methodology.md — Full methodology, exclusion rationale, and the per-molecule international-net-price data-partner ask

• Universe: 597 single-source brand drugs ($41.80B) after excluding skin substitutes (56 codes, $3.64B), vaccines (34 codes, $2.50B), and immune globulins/blood products (47 codes, $2.86B) from the 734-code, $50.79B CY2023 Part B PUF
• ASP is net of rebates by statute (SSA 1847A(c)(3); excludes 340B and Medicaid rebates), so no Part D-style rebate adjustment applies on the US side
• Central international ratio: MedPAC June 2019 / ASPE 2018 ex-manufacturer 1.8x (gap fraction 0.444). RAND 3.22x is the aggressive ceiling only (broader basket, US-net vs. international-gross asymmetry)
• Biosimilar handling: 27 Q5xxx biosimilar product codes ($1.435B) and non-Q5 follow-ons carry zero gap; 15 biosimilar-competed-by-CY2023 originator lines ($2.075B) carry a half-gap; monopoly-priced originators (Eylea, Prolia, Soliris) and novel suffixed biologics (Darzalex Faspro, Vabysmo, Ultomiris, Enhertu, Reblozyl) retain the full gap. Gap-bearing spend $39.70B (95.0% of universe)
• Waterfall: $41.80B universe × 0.444 spend-weighted gap = $17.27B gross gap × 0.65 non-340B share = $11.22B × 0.55 recoverability = $6.17B booked
• 340B netting reserves the domestic acquisition spread for the upcoming Issue #19; recoverability 0.55 (range 0.40–0.70) reflects MFN/GLOBE history and buy-and-bill provider resistance
• Scenarios: conservative $2.85B (1.5x, 340B 0.45, recov 0.40); central $6.17B; aggressive $14.06B (3.22x, 340B 0.25, recov 0.70). IPAY-2028 sensitivity: five in-universe drugs ($2.90B) carry a Maximum Fair Price from Jan 2028; excluding them entirely lowers booked to $5.71B
• Overlap: Issues #2 and #16 cover Part D (mechanically separate program); no dollar double-count. Issue #15 facility-fee components contain zero J-code ingredient spend. The booked figure is a system-spend gap, not a federal-budget-savings figure

cd issue_16

# Download CMS Part D PUF + NHS Drug Tariff; build the 100-product residual universe
# (removing IRA Rounds 1-3, Issue #2 drugs, GLP-1, and generic leakers); compute the
# net-to-net per-molecule savings; print the headline
python 01_build_data.py

# Generate all five charts plus hero
python generate_all_charts.py

Key outputs:

• results/savings_estimate.json — Booked $16.36B, range $13.4-$28.3B, components, rebate tiers, recoverability
• results/top_100_drugs_panel.csv — The 100-product residual universe with per-product gross, net, and booked savings
• results/per_drug_savings.csv — Per-drug net-to-net savings build
• results/savings_by_therapeutic_class.csv — Booked savings by rebate tier / therapeutic class
• results/ira_exclusion_list.csv — IRA Rounds 1-3 and Issue #2 exclusions applied to the universe
• results/cross_validation.csv — Issue #2 nine-drug reproduction (within 0.2% of published $25B)
• results/methodology.md — Full methodology, exclusion rationale, and the per-molecule international-price data-partner ask
• results/originality_gate.md — Stage 3.5 originality-gate verdict and adversarial-math notes

• Universe: top 100 single-source brand products (~85 molecules) in the CY2023 Part D PUF after removing all 25 IRA-negotiated drugs, the nine Issue #2 drugs, tirzepatide (Issue #7 overlap), multi-source generic leakers, and tender-procured vaccines; backfilled to 100 from rank 101+ after each exclusion
• Net-to-net method: net_spend = gross × (1 − rebate_fraction); net_savings = net_spend × 0.689 (RAND 3.22x net gap); booked = net_savings × 0.55. Rebate applied once
• Rebate tiers: top-brand 0.49, mid-brand 0.275, biologic 0.20, insulin 0.70 (ASPE HP-2022-22, SSR Health, Trustees Report)
• Booked $16.36B; range $13.4B (recoverability 0.45) to $28.3B (NHS-floor reference on the 11 matched molecules, recoverability 0.85). Gross-gap method (RAND 4.22x list ratio) would give $18.1B; the lower net-to-net figure is used
• Cross-validation: the same method applied to Issue #2's nine drugs reproduces the published $25B to within 0.2%
• Overlap: Issue #2 nine drugs and all IRA drugs excluded by construction; Issue #4 PBM insulin layer netted out by the 70% insulin rebate; Issue #7 GLP-1 excluded; Issue #17 Part B is a separate Medicare program
• The 3.22x blanket applied to 89 products is the most consequential assumption; the data-partner ask is per-molecule international pricing (IQVIA, Australia PBS, Health Canada PMPRB) to replace it

cd issue_15

# Download CMS OPPS Addendum B, PFS RVU25D, and Medicare Physician PUF;
# join by HCPCS; compute per-procedure site-of-service differential for the
# two clean categories; build commercial extension, overlap subtractions,
# and recoverability sensitivity
python 01_build_data.py

# Generate all five charts plus hero
python generate_all_charts.py

Key outputs:

• results/savings_estimate.json — Booked $2.55B, range $0.93–$4.13B, components, overlap subtractions, recoverability
• results/medicare_counterfactual_savings.csv — Per-category Medicare counterfactual savings
• results/per_hcpcs_savings.csv — Per-HCPCS site-of-service differential
• results/savings_by_component.csv — Component build (Medicare base, commercial extension, overlaps, recoverability)
• results/commercial_extrapolation.csv — Commercial extension at 1.5x (range to 2.54x RAND ratio)
• results/cross_validation.csv — Against MedPAC ambulatory aggregate and CBO clinic-visit scoring
• results/methodology.md — Full methodology, exclusion rationale for imaging and drug administration
• results/originality_gate.md — Stage 3.5 originality-gate verdict and adversarial-math notes

• Two booked Medicare categories computed per HCPCS: clinic visits ($1.821B) and minor procedures ($0.146B), a $1.967B base. Rates are CY2025; volume is DY2023 (a 2025-rate counterfactual on 2023 use)
• Commercial extension at a conservative 1.5x Medicare (full RAND 5.1 ratio of 2.54x forms the range ceiling), not the inpatient-weighted hospital-wide average
• Overlap subtractions: Issue #3 (15% of the commercial layer, hospital-pricing transmission), Issue #12 (5% of gross, consolidation premium). Issue #14 distinct mechanism (0%)
• Recoverability factor 0.60 central (range 0.50–0.70) for legislative grandfathering delays
• Booked $2.55B; range $0.93–$4.13B
• Imaging (358 codes, gross $4.68B) and drug administration excluded: the Physician PUF facility flag cannot isolate HOPD-outpatient volume from inpatient, ER, ASC, and SNF settings, and drug-administration HOPD volume is packaged into Comprehensive APCs. Both require claims-level data (CMS LDS/VRDC, state APCD, MarketScan) and form the data-partner ask
• Cross-validation: MedPAC ambulatory aggregate ($6.6B) includes imaging and drug administration; the booked base is a clean conservative subset; CBO clinic-visit scoring ($3–7B over ten years) is consistent
• Editorial framing: the target is the provider-based billing architecture, not individual hospitals or physicians

cd issue_14

# Build the four-component analysis: international compensation gap (OECD-18 peers),
# workforce-mix counterfactual, RVU-misvaluation residual, GME-allocation counterfactual,
# overlap subtractions, and recoverability sensitivity bands
python 01_build_data.py

# Generate all five charts plus hero
python generate_all_charts.py

Key outputs:

• results/savings_estimate.json — Booked $27.62B, range $19.65–35.49B, four components, overlap subtractions, recoverability bands
• results/per_specialty_savings.csv — Per-specialty international compensation gap
• results/savings_by_component.csv — Component A–D raw and recoverable totals
• results/rvu_panel_full.csv — RVU misvaluation residual by code family
• results/international_compensation_panel.csv — US (BLS anchor) vs. OECD-18 specialist and GP medians, PPP-USD
• results/specialty_workforce_panel.csv — BLS-FTE workforce mix vs. OECD median
• results/overlap_subtractions.csv — Overlap accounting against Issues #3, #10, #11, #12
• results/recoverability_sensitivity.csv — Conservative/central/aggressive recoverability bands
• results/cross_validation.csv — Against Laugesen/Glied 2011 and MedPAC
• results/methodology.md — Full methodology, editorial guardrail, and OECD-18 peer-set rationale
• results/originality_gate.md — Stage 3.5 originality-gate verdict and adversarial-math notes

• Four components, each computed from public data: Component A (international compensation gap, productivity-normalized, OECD-18 high-income peers) $35.3B recoverable from $64.2B raw; Component B (workforce-mix counterfactual, BLS-FTE 28% PC share canonical, toward COGME 45%) $2.6B recoverable from $4.7B raw; Component C (RVU misvaluation residual, commercial cascade only; Medicare = $0 by statutory budget neutrality under SSA Sec. 1848(c)(2)(B)) $2.3B recoverable from $2.9B raw; Component D (GME allocation counterfactual) $1.4B recoverable from $2.4B raw
• Productivity normalization: raw compensation gap discounted by ~23% (2.7/3.5 = 0.77 physicians per 1,000) so each US physician's higher throughput is not counted as excess pay; this makes the estimate conservative
• OECD-18 peer set: AUS, AUT, BEL, CAN, CHE, DEU, DNK, FIN, FRA, GBR, IRL, ISL, ITA, NLD, NOR, NZL, SWE, KOR. Median computed as median-of-country-medians (one vote per country). Lower-income OECD members excluded as non-comparable
• US side anchored on BLS OEWS observed wages (the US does not submit to the OECD DF_REMUN dataset); Doximity/Medscape self-employed data run 10–25% higher, so Component A is conservative on the US side
• Pre-overlap recoverable sum $41.6B; overlap subtractions total $14.0B (Issue #3 $6.2B, Issue #10 $1.5B, Issue #11 $2.1B, Issue #12 $4.2B); booked $27.6B
• Range $19.7–35.5B from conservative and aggressive recoverability bands (each computed post-overlap)
• Editorial guardrail: the savings estimate is computed against system-level counterfactuals (different RVU schedule, different GME allocation, different workforce mix), never against any individual physician's compensation. The target is the payment system, not the people who chose medicine

cd issue_13

# Pull per-filer Form 990 Schedule H Part I from IRS bulk XML (2,103 filers)
python 02_schedule_h_pull.py
python 02b_run_batches.py

# Build the EIN to CCN crosswalk (joins IRS filers to CMS hospital identifiers)
python 03_build_crosswalk.py

# Main analysis: tax-exemption valuation, narrow + broad community-benefit tests,
# state-level decomposition, overlap subtractions, recoverability sensitivity
python 01_build_data.py

# Generate all five charts plus hero
python generate_all_charts.py

Key outputs:

• results/savings_estimate.json — Booked $5.38B, range $4.06–$7.11B, overlap subtractions, recoverability sensitivity
• results/gap_panel.csv — 3,005-hospital panel with narrow- and broad-test gaps
• results/per_hospital_tax_exemption.csv — Per-hospital tax-exemption valuation (7 components)
• results/per_hospital_community_benefit.csv — Per-hospital audited charity care plus Schedule H broad subset
• results/per_filer_schedule_h.csv — Per-filer Schedule H Part I pulled from IRS bulk XML (2,103 filers)
• results/ein_ccn_crosswalk.csv — IRS EIN to CMS CCN matched on filer name and state
• results/savings_by_state.csv — State-level decomposition of failing-hospital gap
• results/overlap_subtractions.csv — Overlap accounting against Issues #3 and #12
• results/cross_validation.csv — Cross-validation against Herring 2018, Bai 2021, Plummer 2024
• results/methodology.md — Full methodology including the v3 Schedule H pull patch and recoverability rationale
• results/originality_gate.md — Stage 3.5 originality-gate verdict and adversarial-math notes
• results/schedule_h_pull_coverage.json — Match coverage statistics for the IRS bulk pull

• 3,005-hospital FY2023 panel after removing partial-year filings and extreme outliers (operating-expense range filters)
• Tax-exemption valuation follows the seven-component Plummer/Socal/Bai JAMA 2024 method applied to each hospital's federal Form 990 + CMS HCRIS financial line items; aggregate $46.4B; federal income tax forgone is the largest component at $17.0B (36.6%)
• Schedule H pull (v3 patch, May 2026): per-filer Schedule H Part I from IRS bulk XML for 2,103 matched filers (76% of panel expenses); the remaining 24% (975 hospitals) fall back to HCRIS S-10 charity x 2.5 sector uplift to approximate the Schedule H broad subset
• Narrow test (Herring 2018): audited HCRIS charity care vs. tax-exemption value, 86% fail, $31.3B aggregate failing-hospital gap. This 86% replicates Herring 2018 to within rounding on FY2023 data, the calibration check for the per-hospital tax-exemption math
• Broad test (Bai 2021): audited charity + Medicaid shortfall + community health + subsidized services vs. tax-exemption value, 44% fail, $11.9B aggregate gap. Medicaid shortfall ($44.1B in aggregate across matched filers) is the largest broad-subset component and tilts the broad test toward Medicaid-heavy hospitals
• Recoverability factor (0.53 central, 0.40–0.70 sensitivity range): reflects what the literature on state revocation cases and consent decrees (Provena, UPMC, Pittsburgh, Boston) suggests is realistic to redirect into actual charity care through enforcement or 501(r) tightening
• Booked $5.4B = $10.15B post-overlap × 0.53 recoverability; range $4.1–7.1B from the 0.40–0.70 recoverability sensitivity
• Overlap subtractions: Issue #3 hospital pricing (commercial-vs-Medicare price levels) — $0.60B; Issue #12 consolidation tax (HHI-driven commercial premium) — $1.19B. Different mechanisms but partial overlap on the at-risk commercial spend
• Ownership comparison (charity care share of operating expenses): government 3.56%, for-profit 3.14%, nonprofit 1.86%. This ordering — for-profit hospitals paying full taxes delivering more charity care per dollar of expenses than tax-exempt nonprofit hospitals — reproduces Bai 2021 on FY2023 data
• No overlap with Issue #5 (admin waste counts processing cost, not exemption value) or Issue #6 (supply purchasing variance is operationally distinct)

cd issue_12

# Build the merger-event panel from CMS POS 2018-2025, compute HSA-level HHI shifts,
# join HCRIS commercial-spend exposure, and apply the four-anchor piecewise coefficient
python 01_build_data.py

# HRR-vs-HSA market-definition sensitivity check
python 02_hrr_sensitivity.py

# Generate all five charts plus hero
python generate_all_charts.py

Key outputs:

• results/savings_estimate.json — Booked $13.03B, raw $17.37B, overlap subtractions, sensitivity at 5% and 10% blended uplift
• results/merger_event_panel.csv — 1,155 ownership-change events 2018–2025 with horizontal-merger flag
• results/market_hhi_panel.csv — HSA-level HHI before/after each merger, with shift magnitude
• results/savings_by_market.csv — Per-market booked savings, ordered for Pareto/outlier analysis
• results/commercial_spend_at_risk.csv — HCRIS-derived commercial spend exposure by merger-market HSA
• results/hrr_sensitivity.csv — Same analysis at HRR (vs HSA) market definition
• results/cross_validation.csv — Booked figure against published Cooper, Dafny, FTC, Brot-Goldberg anchors
• results/overlap_subtractions.csv — Per-issue overlap accounting (Issue #3 hospital pricing, Issue #15 facility fees)
• results/methodology.md — Full methodology including the piecewise coefficient construction
• results/originality_gate.md — Stage 3.5 originality-gate verdict and adversarial-math notes

• Panel construction: CMS POS annual snapshots 2018–2025 yielded 1,155 ownership-change events. Horizontal-merger filter (hospital-on-hospital, same or adjacent market) reduced the panel to 530 events; tax-status conversions and government transitions are reserved for the upcoming Issue #13 (The Nonprofit Lie)
• Market definition: Dartmouth HSA is the primary unit; HRR is reported as a sensitivity check (Chart 3 shows the $13B HSA central vs. $16B HRR alternative)
• Coefficient construction: Piecewise HHI-dependent uplift derived from the four academic anchors. The booked figure uses the actual computed HHI shift in each HSA, not a uniform multiplier; chart 1 visualizes the coefficient band against the post-2019 cohort
• Commercial-spend exposure: HCRIS commercial-payer line items, restricted to merger-market HSAs, totaled $218.44B at risk. Booked uplift at the piecewise coefficient = $17.37B raw
• Overlap accounting: Issue #3 covers the 254% commercial-vs-Medicare price level (different mechanism, same denominator). $3.47B subtracted to avoid double-counting. Issue #15 (forthcoming) covers vertical integration / facility-fee uplift on outpatient services; $0.87B subtracted as a defensive overlap estimate
• Booked $13.03B = $17.37B raw − $3.47B (#3) − $0.87B (#15). Range high $50B reflects the upper coefficient anchor (Cooper monopoly) at the full at-risk denominator with no overlap subtraction
• Sensitivity disclosed: at a flat 5% blended uplift the at-risk spend implies $8.19B; at 10% it implies $16.38B. The piecewise per-market approach falls inside that envelope
• No overlap with Issue #11 (MA Overpayment): #11 covers risk-score upcoding inside Medicare Advantage. #12 covers commercial-rate negotiation between merged hospitals and private insurers. Different payer side, different mechanism

cd issue_11

# Build the V24/V28 sensitivity band, HRA decomposition, state allocation, and cross-validation
python 01_build_data.py

# Generate all five charts plus hero (V24/V28 band, HRA share trajectory, state allocation, qui tam timeline, savings tracker)
python generate_all_charts.py

Key outputs:

• results/savings_estimate.json — Headline central, V24/V28 band, range low/high, full Path A by-year structure
• results/coding_intensity_timeseries.csv — Per-year MedPAC central plus V24-only and V28-only counterfactuals, 2023–2026
• results/pool_share_trajectory.csv — Coding-intensity overpayment as a share of the MA Part C payment pool, 2021–2026 (original)
• results/hra_decomposition.csv — Health Risk Assessment yield as a share of national coding-intensity overpayment (original)
• results/state_level_decomposition.csv — State-level allocation of the 2025 anchor (original allocation, not estimation)
• results/qui_tam_settlements.csv — MA risk-adjustment FCA settlement timeline, 2018–2026
• results/cross_validation.csv — Cross-validation against Kronick 2025, OIG, and CMS Risk Adjustment Trend
• results/methodology.md — What is original here vs. what is curated reference data, with the data-partner CTA

• Headline anchor ($28B for 2025) is curated from MedPAC March 2026 Fig 12-6: the issue does not claim the central as original. It is the cleanest public estimate of the coding-intensity slice.
• V24-only-vs-V28-only sensitivity band is the original analytical contribution: the 2025 band is $19.2B (V28-only) to $44.8B (V24-only), a $25.6B spread that reflects how much of the answer hinges on which formula is in force.
• The Kronick-style insurer-level decomposition is not replicable from CMS public files. Kronick used the CMS Chronic Conditions Warehouse, which requires a Data Use Agreement. The newsletter and the methodology file flag this explicitly. Insurer-level decomposition is the data-partner CTA.
• HRA decomposition (HRA yield as share of national coding-intensity overpayment) is original to this analysis, computed from OIG audit yield rates against the MedPAC pool.
• State-level allocation of the 2025 anchor is original, computed from CMS Geographic Variation PUF risk-score deltas; it is an allocation of MedPAC's central, not an independent state-level estimate.
• Coding intensity is the slice booked. Favorable selection and benchmark structure (the rest of the $76B MedPAC gap for 2026) are named in the newsletter and reserved for future issues; not double-counted.
• No overlap with Issue #8 (Denial Machine): #8 covers prior-authorization denial mechanics; #11 covers risk-score upcoding. Different dollars, different mechanisms.
• No overlap with Issue #10 (Procedure Mill): Component C of #10 is a price-multiplier extension for MA volume of low-value procedures; #11's headline is risk-score-driven payment, not procedure-volume driven payment.

cd issue_10

# Pass 1: build the headline analysis from raw PUFs (re-downloads CMS files)
python 01_build_data.py

# Pass 2: PSPS modifier and place-of-service profiling, dedup pass
python 02_component_a_pass2.py

# Pass 3: adversarial-math corrections (mean-share dedup, BLS Medical CPI 1.60x)
python 03_pass3_corrections.py

# Generate all five charts plus hero
python generate_all_charts.py

Key outputs:

• results/savings_estimate.json — Pass 1 booked components and ranges, plus framing benchmarks (Kim & Fendrick, Schwartz, Mafi, Lown, CMS WISeR)
• results/component_a_schwartz_medicare.csv — Per-measure Medicare paid and low-value share
• results/component_b_state_variance.csv — State-level low-value spend per beneficiary
• results/component_c_all_payer.csv — All-payer extension multipliers
• results/component_d_wiser_pilot.csv — WISeR 17-procedure pool by pilot state
• results/component_e_defensive_medicine_did.csv — Difference-in-differences output across three control-state specifications
• results/pass3/savings_estimate_v3.json — Pass 3 corrected headline ($7.628B booked, $13.619B range high)
• results/pass3/methodology_v3.md — Detailed Pass 3 methodology with both corrections documented
• figures/ — All five analysis charts plus hero

• Five booked components, each computed from federal PUF data:
  • Component A ($4.31B): Medicare Schwartz-list pool from PUF spend × measure-level low-value share
  • Component B ($0.42B): state-variance compression to P25 (range to P10 included)
  • Component C ($2.16B): all-payer extension using MedPAC and RAND multipliers
  • Component D ($0.46B): WISeR pilot 17-procedure pool for six pilot states
  • Component E ($0.27B): defensive-medicine difference-in-differences (booked at the conservative midpoint across three control-state specifications)
• Pass 3 corrections (caught by our own adversarial-math review and committed to the published results):
  • Mean-share dedup replaces max-share dedup for the 106 HCPCS that map to two measures (saved $583M from being double-counted)
  • BLS Medical CPI factor corrected from 1.74x to 1.60x (CUUR0000SAM, 2008→2024 annual average)
• Booked $7.628B sits below CMS's own November 2025 finalization of $19.6B in skin-substitute reductions alone, and well below Lown's $75–100B macro estimate. The gap is the explicit data-partner ask: licensed claims data (Optum Clinformatics, MarketScan, IQVIA Pharmetrics) or restricted Medicare extracts (CMS LDS/VRDC) would let us add diagnosis-code filtering, NPI-level analysis, and broader low-value-care detection that public PUF data cannot support
• Range high $13.619B reflects Component A at upper-bound shares + Component B at P10 + Component C at the upper multiplier + the Mello inflation-corrected defensive-medicine ceiling
• No overlap with Issue #5 (admin waste): #5 counts the processing cost of low-value care; #10 counts the paid spending on services that should not have been ordered
• No overlap with Issue #11 (MA Overpayment): Component C's MA extension is a price-multiplier adjustment to the same volume of services, not a separate volume of upcoded patients

cd issue_09

# Build the Schedule A and Schedule C analysis datasets
python 01_build_data.py
python 02_build_data_schedule_c.py

# Generate all four analysis charts (peer variance, broker-vs-admin boxplot, savings decomposition, running tracker)
python generate_all_charts.py

Key outputs:

• results/savings_estimate_v2.json — Booked components and range with all assumptions
• results/schedule_c_admin_variance.csv — Per-peer-group admin fee P10/P25/P50/P75/P90
• results/schedule_a_c_linkage.csv — 425 plans filing both Schedule A and Schedule C, with broker rate and admin fee per participant
• results/overlap_matrix.md — Component-level overlap accounting against Issues #3, #4, #5, #8
• results/meps_ic_verification.md — Cross-validation against MEPS-IC public tables
• figures/ — All analysis charts

• Three booked components, each computed from federal filings:
  • Component A ($2.18B): per-plan broker commission above the 3% DOL benchmark, 7,036 fully insured plans
  • Component B ($2.77B): Schedule A broker rate ($82.68 per life) extended at conservative 0.5 ratio to ~115M employer-sponsored self-insured lives
  • Component C ($1.68B): above-peer-median admin fees, 8,180 plan Schedule C sample at 30% reducibility
• 30% reducibility floor is conservative: post-Tibble v. Edison 401(k) reform compressed retirement plan administrative fees by 20–30% over five years; Ohio Medicaid PBM reform compressed by 12–18% in year one
• Bias caveat: Schedule C is filed primarily by large, trust-funded plans. One cell (jumbo plans with mixed funding, 260 plans) accounts for 65% of in-sample lives and half the disclosed fees. Excluding the cell entirely drops Component C from $1.68B to $0.87B and the headline from $6.6B to $5.8B
• Broker-admin correlation: 425 plans filed both Schedule A and Schedule C. Pearson r = 0.030 (linear), Spearman rho = 0.199 (rank). Admin medians do climb 3.2× across broker quartiles, but within-quartile spread is much larger than the between-quartile shift, so broker rate is a weak predictor of admin fee — Components A and C measure different mechanisms and add without double-counting
• Range ceiling ($12.2B) reflects 50% reducibility on Component C plus a bounded extrapolation to the ~63 million ERISA-welfare participants in plans that file Form 5500 but skip Schedule C; the bounded extrapolation is described in the newsletter's CTA box but is not booked, pending matched plan-level claims data
• No overlap with prior issues: Schedule C dollars flow to TPAs, consultants, brokers, and lawyers — not to hospitals (Issue #3), PBMs (Issue #4), insurer underwriting (Issue #8), or general administrative complexity (Issue #5). See results/overlap_matrix.md for the full accounting

cd issue_08

# Build dataset from CMS-0057-F disclosures and SEC filings
python 01_build_data.py

# Generate analysis charts
python generate_all_charts.py

Key outputs:

• results/ — Per-contract denial rates, appeal analysis, savings model
• figures/ — All analysis charts

• Original CMS-0057-F per-contract extraction: 93 MA contracts, 18.4M PA requests
• Volume-weighted denial rates computed from per-contract data (not headline averages)
• Savings components (restated 2026-07-06): Care Suppression ($13.7B mid), Vertical Integration Arbitrage ($10.3B mid), Risk Adjustment ($0.3B). AI Denial Escalation (previously $5.7B mid) moved out of the booked total — its magnitude rested on a misattributed source with no measured effect size — and is now treated qualitatively
• MLR Gaming (mechanisms real; the prior $11.8–20.7B range rested on a fabricated citation and is dropped) documented but excluded from booked savings
• Booked: $24B (Care Suppression + Vertical Integration + Risk Adjustment); restated from $32B on 2026-07-06 (see issue_08/CITATION_AUDIT_2026-07-06.md)
• Component D (deductible-delay extraction) described in the newsletter's MRI vignette but excluded from booked total pending matched patient-level claims + deductible-exposure data
• No overlap with Issue #5 (admin waste counts processing cost of PA; this counts the denied-care cost)

cd issue_07

# Run the BALANCE Model projection (no downloads needed — all data hardcoded from published sources)
python 01_build_data.py

# Generate analysis charts
python generate_all_charts.py     # Charts 1–4 (market growth, price comparison, cost projection, pricing structure)
python generate_chart5.py         # Chart 5 (savings tracker)

Key outputs:

• results/balance_projection_all_scenarios.csv — 10-year enrollment and cost projections (LOW/MID/HIGH)
• results/sensitivity_analysis.csv — Model uncertainty drivers
• results/health_benefit_roi.csv — Cost-benefit analysis
• results/international_prices.csv — US vs. international GLP-1 price comparison
• results/market_growth.csv — Historical GLP-1 market trajectory (2018–2025)
• results/key_metrics.json — Summary of all headline numbers and assumptions
• figures/ — All analysis charts

• Eligible population: 67.5M Medicare beneficiaries × 35.5% obesity × 55% behavioral engagement = 13.2M
• Enrollment ramp: 5% (2027) → 12% (2028) → 40% plateau (2031+), 88% annual retention
• Three pricing scenarios: LOW ($2,940/yr), MID ($3,600/yr), HIGH ($4,200/yr)
• 10-year cost: $86.2B (LOW) to $170.1B (HIGH); MID = $124.8B
• Health benefit ROI: 33% conservative (prevented CV events and T2DM progression)
• Savings: $40B/yr from 50–70% US GLP-1 price reduction through reference pricing and generic competition
• No overlap with Issue #2 (which covers 9 specific top-spend Medicare drugs) or Issue #4 (PBM extraction)

cd issue_06

# Build dataset from raw HCRIS FY2023/FY2024 flat files
# Downloads ~200MB per year, extracts supply cost centers, computes CMI-adjusted variance
python 01_build_data.py

# Generate analysis charts
python generate_chart1_supply_variance.py   # Supply cost variance by bed size (CMI-adjusted)
python generate_chart2_surplus_nonprofits.py # Medical surplus redistribution
python generate_chart3.py                   # Supply cost decomposition ($170.9B)
python generate_chart4.py                   # Ownership breakdown (for-profit vs nonprofit vs govt)
python generate_chart5.py                   # Implant price variance
python generate_chart7_state_ranking.py     # 50-state supply waste ranking

Key outputs:

• results/expanded_analysis_results.json — State rankings, teaching analysis, FY2023 vs FY2024 comparison
• figures/ — All analysis charts

• 5,480 hospitals analyzed (≥50 discharges, nonzero supply costs) from raw HCRIS FY2023 federal cost reports
• Three supply cost categories: medical supplies (Worksheet A, line 53), implantable devices (line 55), drugs charged to patients (line 56)
• CMI adjustment: per-discharge costs divided by hospital Case Mix Index to normalize for patient acuity
• Bed-size stratification: Small (<100), Medium (100–299), Large (300–499), Major (500+)
• Savings scenarios: Q4→P75 within peers = $28.5B (conservative); above-median→P50 = $58.9B; UCSF 6.5% universal = $11.1B
• Ownership: For-Profit 1,576 hospitals ($18.8B total), Nonprofit 2,993 hospitals ($128.9B total, 75.5% of national spend), Government 911 hospitals ($23.0B total)
• No overlap with Issue #3 (hospital pricing): that covers commercial-to-Medicare price ratios; this covers within-hospital purchasing efficiency

cd issue_05

# Build dataset from raw HCRIS FY2023 flat files
# Downloads ~200MB, extracts admin/overhead cost centers from Worksheet A
python 01_build_data.py

# Generate all analysis charts from the hospital dataset
python generate_all_charts.py

Key outputs:

• results/hospital_admin_costs_fy2023.csv — 4,518 hospitals, 22 columns (admin costs, overhead breakdown, payer mix)
• figures/ — All analysis charts

• Four original analyses: (1) per-capita international comparison (US $4,983 vs 10-peer avg $884), (2) Woolhandler update to 2023 ($1.13–1.66T), (3) prior auth national cost ($21–93B), (4) HCRIS hospital admin variance
• HCRIS analysis: 4,518 hospitals (≥100 discharges, nonzero A&G), $141.2B total A&G, $268.5B total overhead (32.2% of total costs)
• P75/P25 ratio: 6.2× nationally, 2.0–2.7× within bed-size peers (CMI-adjusted: 2.1–3.1×)
• Ownership: For-profit median $458/DC, nonprofit $1,980/DC, government $2,524/DC
• Savings: Q4→P75 = $18.0B, above-median→P50 = $39.8B (hospital admin only; $200B includes full system)
• Published dataset: results/hospital_admin_costs_fy2023.csv (4,518 hospitals, 22 columns)
• No overlap with Issues #1–4: those cover drug prices, hospital procedure prices, and PBM extraction; this covers administrative overhead

cd issue_04

# Generate analysis charts from cited federal data and academic literature
python chart1_pbm_market.py       # PBM market concentration (Drug Channels Institute 2024)
python chart2_harm_spread.py      # Spread pricing extraction mechanisms (FTC, Ohio Auditor)
python chart4_biosimilar_v4.py    # Biosimilar adoption by state PBM law (CMS Part D, JAMA)
python chart5_insulin_prices.py   # Insulin price trajectory (IQVIA, CMS)

Key outputs:

• results/biosimilar_analysis_2023.csv — CMS Part D biosimilar adoption data
• results/key_metrics.csv — Core PBM extraction metrics
• figures/ — All analysis charts

• Savings model is conservative, built from six distinct non-overlapping mechanisms
• Mechanism 1 (spread pricing): $3.0B — Ohio audit extrapolated to national Medicaid managed care
• Mechanism 2 (rebate pass-through): $10.0B — PBMs retain estimated 5–10% above disclosed admin fees on $334B rebate pool
• Mechanism 3 (specialty markup): $1.5B — FTC-documented $7.3B over 5 years, annualized
• Mechanism 4 (formulary reform/biosimilar preference): $10.0B — biosimilar adoption gap vs. states with strong PBM laws
• Mechanism 5 (admin transparency savings): $5.5B — waste from opaque PBM reporting requirements
• Total booked: $30.0B/year (range $30–50B)
• No overlap with Issues #1–3: Issue #2 addresses manufacturer-level drug prices; Issue #4 addresses the intermediary extraction layer on top of those prices
• CAA 2026 (enacted Feb 3, 2026) includes rebate pass-through effective 2029 for commercial plans; FTC settled with Express Scripts Feb 4, 2026 ($700M/yr projected savings from one PBM)

cd issue_03

# Build HCRIS cost report dataset and compute cost-to-charge ratios
python 01_build_data.py

# Generate analysis charts
python 02_visualize.py

Key outputs:

• results/hospital_ccr_2023.csv — Hospital cost-to-charge ratios (3,193 hospitals)
• results/savings_calculation.json — Full savings model parameters
• results/procedure_prices.json — International procedure price comparisons
• figures/ — All analysis charts

• Savings formula: $528B commercial hospital spend × 65% addressable × 21.3% price reduction (254%→200% of Medicare) = $73B
• 3,193 hospitals analyzed from raw HCRIS FY2023 federal cost reports
• For-profit hospitals: 4.11× median markup (highest); nonprofit: 2.46×; government: 2.22×. 37% of all hospitals charge 3× or more
• Correction (2026-03-17): Original release mislabeled CMS ownership codes, swapping nonprofit and for-profit categories. The $73B savings estimate was unaffected (derived from RAND/CMS NHE national data). See issue_03/CTRL_TYPE_AUDIT.md for details.
• Fix mechanism (Commercial Reference Pricing) is already implemented in Montana and by thousands of self-insured employers
• No overlap with Issues #1 or #2 (those cover drug prices only; this covers hospital/procedure prices)

cd issue_02

# Build reference price dataset (NHS Drug Tariff + RAND international averages)
python 01_build_reference_data.py

# Generate analysis charts
python 02_visualize.py

Key outputs:

• results/nhs_vs_medicare.csv — Medicare vs. NHS Drug Tariff price comparisons
• results/kff_drug_comparison.csv — 11-country OECD drug price benchmarks
• results/rand_country_ratios.csv — RAND international price ratios
• figures/ — All analysis charts

• Medicare figures are gross cost (pre-rebate) from CMS Part D Public Use File
• ~49% net rebate adjustment applied for top-spend brand drugs, triangulated from MedPAC and Feldman et al.
• NHS prices are post-patent generic reimbursement rates — representing the molecule's commodity price
• International average = Peterson-KFF 11-country OECD analysis

cd issue_01

# One-time environment setup
chmod +x 01_setup.sh && ./01_setup.sh
source .venv/bin/activate

# Download CMS Part D data (~200 MB)
python 02_download_data.py

# Build local DuckDB database
python 03_build_database.py

# Run analysis
python 04_analyze.py

# Generate analysis charts
python 05_visualize.py

Key outputs:

• results/by_drug_2023.csv — Per-drug Medicare Part D spending and OTC price comparisons
• results/bene_overpayment_2023.csv — Beneficiary-level overpayment estimates
• figures/ — All analysis charts

• OTC unit prices sourced from current retail at major US pharmacies (March 2026)
• 30-unit-per-claim approximation; see issue_01/VALIDATION_REPORT.md for full methodology
• Savings figures are conservative — do not account for PBM rebates or dispensing fees