How MarkupRx actually works.

Each leak figure on the homepage is an example computed from the same margin = (total_amount_reimbursed / units_reimbursed) - nadac_per_unit formula documented in NADAC pricing & underwater detection. Per-pharmacy results vary by claim mix.

The underwater margin formula is computed per claim-NDC pair from SDUD joined against the latest NADAC entry per NDC:

A negative margin = the pharmacy lost money dispensing that prescription.

• NADAC source: data.cms.gov/nadac — refreshed weekly by CMS; ~34,000 NDCs covered.
• SDUD source: medicaid.gov/sdud — quarterly Medicaid fee-for-service and managed care reimbursement at the NDC level; ~700,000 rows per quarter.
• Unit reconciliation: NADAC and SDUD must share the same pricing unit (EA, ML, etc.) for the margin to be valid. NDCs with mixed pricing units are excluded from underwater detection rather than risk a misleading comparison.

Live freshness for all 13 sources is on our /sources page. Underwater rankings by NDC and state are publicly viewable (no login) in the data explorer.

Appeals are generated by Anthropic Claude Haiku 4.5 with structured-output JSON (no streaming, no chatbot). The prompt assembles your claim record, the relevant NADAC entries, and the state Medicaid pricing context from CMS SDUD, then produces an appeal letter tailored to the PBM's appeal process.

The appeal letter cites:

• The specific NDC and fill date from your dispensing data
• The NADAC acquisition-cost benchmark as of the fill date (point-in-time, not current NADAC — the defensible figure for an appeal is what NADAC said on the date of the fill)
• The SDUD reimbursement rate for that NDC in your state
• The per-fill loss amount (NADAC minus reimbursement)

See a full sample on our samples page or read the step-by-step appeal process guide on the blog.

Two signing systems are in use:

• Ed25519 (legacy): bundles signed before our KMS migration use this curve. Download the legacy public key.
• AWS KMS P-256 (current): all new bundles use this hardware-backed key with CloudTrail WORM audit trail. Download the current public key.

Verify any audit bundle yourself with verify-audit-bundle.py and the public key linked above.

We use Claude Haiku 4.5 (claude-haiku-4-5-20251001) with a $50/month spend cap. When the monthly budget is exhausted, AI features gracefully degrade — cached responses still serve, but new analyses pause until the next billing month. Spend is tracked per request and aggregated daily.

The 13 sources we integrate, with their update cadence:

• CMS NADAC — weekly pharmacy acquisition costs, ~34,000 NDCs (data.cms.gov/nadac)
• CMS SDUD — quarterly Medicaid reimbursement at NDC level (medicaid.gov/sdud)
• FDA NDC Directory — weekly drug identification (openFDA bulk JSON)
• FDA Orange Book — weekly patents and exclusivity for GenericWatch
• CMS Federal Upper Limit — monthly Medicaid ceiling pricing
• CMS ASP — quarterly Medicare Part B pricing
• VA FSS — semi-monthly VA Federal Supply Schedule contract prices
• IRA Negotiated Prices (MFP) — as published (~125 drugs, IRA-negotiated)
• CMS Drug Spending (Part B, Part D, Medicaid) — annual spending by drug
• FDA Drug Shortages — weekly supply disruptions via openFDA API
• Drugs@FDA — weekly approval history (7 tab-delimited files)
• RxNorm (NLM) — monthly RXCUI entity resolution (RXNSAT.RRF + RXNCONSO.RRF)
• HCPCS Crosswalk — quarterly NDC → Medicare Part B code mapping

All sources are joined through RxNorm RXCUI — the NLM's canonical drug entity identifier — so a NADAC NDC, an Orange Book patent, and an ASP billing code all resolve to the same drug concept. We never use fuzzy text matching for pharmaceutical name resolution.

Visit our /sources page for the live freshness table — updated every 5 minutes from our pipeline. Or explore the current underwater rankings in the public data explorer — refreshed weekly when CMS publishes new NADAC data.

For background on the NADAC vs. MAC gap mechanism, see the NADAC vs PBM Reimbursement explainer on the blog.