The price-transparency data is real,
mandatory, and deeply messy.
Federal law forces every health insurer to publish what they pay. But the files are sprawling, inconsistent, and padded with rates that mean nothing. This page shows exactly how messy – measured against our own 541-million-row index – and what we do about it. Every number traces to a query or a cited source.
Two fair questions about insurer data
If you build on Transparency-in-Coverage (TiC) files, two suspicions come up fast. We tested both – against the public record and against our own database. Both are true.
Monthly updates are legally required, but compliance is patchy, prior versions are never retained, and not a single payer has ever been publicly fined. CMS is now proposing to drop the requirement to quarterly.
Files carry thin identifiers (NPI + tax ID, no address) and omit real providers – while simultaneously over-listing 84–92% "ghost" rates for providers who would never perform that service. Naïve averages come out distorted.
What the law actually requires
The Transparency in Coverage rule (45 CFR 147.212) has been in force since July 2022. On paper, it is strict.
What the rule mandates
- Three machine-readable files: in-network rates, out-of-network allowed amounts, and drug prices.
- The in-network file must be updated monthly, with the update date clearly stamped (45 CFR 147.212(b)(3)).
- Posted publicly, free, in a standardized non-proprietary format – no login, no fee.
- Negotiated rates for all covered items and services with all in-network providers.
What actually ships
- Files that skip monthly updates, omit tax-IDs, exclude lines of care, and let URLs expire (Georgetown CHIR, 2025).
- No retained history – CMS only "recommends" keeping prior versions, so month-over-month change is unknowable.
- Single files large enough to be practically unusable without heavy engineering.
- Provider lists that diverge sharply from real networks – one study found a major insurer listed ~47% of its advertised physicians (AJMC, 2025).
Most of the rates aren't real-world rates
A "ghost rate" is a negotiated price listed for a provider who would never furnish that service – a podiatrist priced for heart surgery. They're contractual boilerplate, not fabricated numbers, but they bloat the files and distort every naïve average. Independent estimates of how many rates are clinically implausible:
Share of payer in-network rates that are "ghost" / clinically implausible
Estimates vary by method and sample. The peer-reviewed figure (Health Affairs Scholar, 61 insurers, 3.4B rates) is the most defensible to cite.
CMS itself now acknowledges the cause: payer-provider contracts "account for all items and services for all their providers, irrespective of clinical specialty." In a December 2025 proposed rule (CMS-9882-P), the agency proposed excluding these implausible provider-rate combinations and relaxing the update cadence from monthly to quarterly.
"Monthly" isn't really monthly – and nobody's enforcing it
The rule says update monthly. Reality: some payers do, many don't, none are required to keep the old versions, and the penalty mechanism has never been used against an insurer – even as hospitals are fined under the separate hospital rule.
Because no one retains prior monthly files, the 2022–2025 history is effectively gone – for the industry and for anyone who didn't archive it in real time. That absence is the finding.
What this looks like inside Reddenda
We don't just cite studies – we ingest these files at scale and can measure them directly. Here is the honest state of our own data, queried live on 2026-06-23. No estimates: these are exact counts.
Rate rows by payer: the shape of our index
Live from payer_contracted_rates (exact row counts, top payers). Each bar is that payer's share of the full index. Our national build is mid-flight: Anthem's files were ingested first and dominate today; UnitedHealthcare, Cigna and others are still loading. 233 payer files total.
Our index covers 85,488 real providers – and that gap from the ~9.2M-NPI universe is mostly the files' fault, not a bug.
Four honest reasons a provider isn't in the data
The payer simply omits them
TiC files routinely exclude subsets of providers and lines of care. If a payer doesn't list a provider for a code, no amount of engineering can invent it – and that's a real coverage gap in the source.
The match needs an outside key
Files carry only NPI + tax ID – no name, no address, no specialty. Linking to a real, current provider requires the public NPI Registry (NPPES); where that linkage is weak, a provider can fall out of view.
We deliberately drop ghost rates
If a rate is clinically implausible for that provider, we discard it rather than show a fake number. That's the right call – but it means a provider who only has ghost rates won't appear with a usable rate. We'd rather show nothing than something false.
We index what our customers bill
We target the procedure codes and specialties our customers actually use – not all ~10,000 codes for all ~9.2M NPIs. That's a scope choice, not a defect; coverage expands as we add specialties and states.
How we keep the data honest
The messiness is the moat. Anyone can download a payer file. Turning it into something a practice can trust is the hard part – and where we refuse to cut corners.
Real NPIs only – zero placeholders
Our index holds 0 rows keyed to a missing or "0" NPI. We don't carry tax-ID-only or ghost-NPI records that pollute raw files. Every rate is tied to a real, resolvable provider.
Ghost rates discarded, not averaged in
Clinically implausible provider-rate pairs are filtered out before anything reaches a customer, so a benchmark reflects what a real provider is actually paid – not contract boilerplate.
Local-peer median, never a fake default
Rates are scored against the local peer median for that NPI and CPT. Where public data is thin, we label it "insufficient public data" – we never substitute a specialty average and call it specific.
We re-pull and timestamp
We re-ingest payer files and keep our own dated record of each vintage – so when a payer does update, we capture it, and we can show our work instead of trusting a file's own (often missing) date.
Where every number came from
First-party figures were queried live against our Postgres index on 2026-06-23 using exact counts (no sampling, no estimates). National figures are linked to their primary sources.
payer_contracted_rates.
85,488 distinct NPIs – exact COUNT(DISTINCT npi) via an index skip-scan.
0 ghost-NPI rows – COUNT(*) WHERE npi IS NULL OR npi='0' OR npi='' returned 0.
Per-payer rows & NPIs – grouped aggregate over the full table (chart above).
Update cadence – file dates parsed from tic_ingestion_runs.source_url; observed capture window 2026-02 → 2026-06.
Honest caveats: Our capture window begins in 2026, so we cannot reconstruct 2022–2025 cadence from our own files – no one is required to retain them. "Ghost rates" are real contracted prices that are clinically implausible, not invented numbers. National ghost-rate estimates range from ~60% to ~96.5% by method; we cite the peer-reviewed 84–92% as the defensible range. NPI is not a strictly mandated standalone identifier, so tax-ID-only records are schema-permitted, not violations. Numbers reflect a live, growing index and will change.
We'd rather show you the mess than hide it.
That honesty is the product. We turn sprawling, ghost-laden federal files into one clean number a practice can act on – and we tell you exactly where the data is thin.