Versioned evidence policy
A benchmark methodology built to show its limits
The public projection is produced by deterministic source, identity, comparability, aggregation, and publication gates. AI can assist discovery and editorial work; it cannot decide the numerical truth.
- Policy version
- 2026-08-30
- Methodology version
- hpt-facility-median-v1
- Launch source
- Hospital Price Transparency files
- Release ID
- rr-bootstrap-4cb5bb885d8eaf73033237b5
- Approval scopes
- Legal · Security · RCM SME · Editorial
- Recorded without publishing or inventing an approver identity.
Evidence model
Four layers of reimbursement truth
Public rates are a market signal. A contract and the adjudicated remit are different evidence layers.
Read the methodology- 01Public evidence
Public hospital benchmark
Hospital-reported rate-file observations, normalized into comparable aggregate cohorts.
- 02Separate phase
Public professional or group benchmark
Transparency in Coverage observations require a distinct provider-group methodology and are not mixed with hospital data.
- 03Private analysis
Signed contract expectation
The provider contract establishes the expected reimbursement after its terms, modifiers, bundles, and setting rules are applied.
- 04Private analysis
Actual 835 adjudication
The 835 or ERA shows the allowed, paid, adjusted, and denied outcome that can be compared with contract expectation.
01
Approved launch scope
The launch dataset is limited to official Hospital Price Transparency machine-readable files linked by an official hospital domain or its official transparency manifest. A third-party vendor or CDN may host a file only when the official hospital site establishes the relationship.
The public projection contains sanitized aggregates only. It excludes raw rows, hospital identifiers, individual providers, NPIs, TINs, plan names, and any patient or claim data. Professional or provider-group Transparency in Coverage data requires a separate methodology and is not mixed into these hospital cohorts.
Code identifiers can be displayed. Licensed CPT descriptions are not republished by this directory.
02
Acquisition and publication states
A record advances only when the current state completes. Retries, quarantines, and human review cannot be bypassed by the content generator.
- 01
Official source discovered
- 02
Hospital identity verified
- 03
File retrieved and hashed
- 04
CMS schema validated
- 05
Observations normalized
- 06
Entities resolved
- 07
Quality rules evaluated
- 08
Comparable cohort aggregated
- 09
Page eligibility evaluated
- 10
Structured claims validated
- 11
Published and monitored
AI may assist
Discovery and explanation
- Suggest official source locations for verification.
- Propose entity aliases for an approved mapping table.
- Triage exceptions and schema drift for human review.
- Draft prose from a verified, structured fact packet.
AI may not decide
Numbers and publication gates
- Parse authoritative monetary values or calculate percentiles.
- Convert algorithms or percentages into dollar rates.
- Merge payer identities or infer missing geography.
- Decide that different settings or methodologies are comparable.
- Override a failed quality, legal, or indexation rule.
03
Strict comparability before aggregation
An aggregate is meaningful only when the observation cohort retains the dimensions that can change reimbursement.
- Care setting
- Inpatient observations are not combined with outpatient observations.
- Billing class
- Facility observations are not combined with professional observations.
- Code identity
- Code systems, code vintages, codes, and modifier sets must remain compatible.
- Payment unit
- Per-diem, case-rate, fee-schedule, and other dollar methodologies remain distinct.
- Bundle status
- Bundled and unbundled observations are not combined.
- Evidence layer
- Hospital negotiated-charge observations are not mixed with historical allowed amounts, contracts, or 835 payments.
04
Deduplication and distribution
- Remove exact duplicate source rows.
- Collapse identical facility, payer, plan-label, comparable-service, and amount records.
- Calculate one median facility-level point within the comparable cohort so verbose files cannot dominate.
- Calculate the 25th percentile, median, and 75th percentile across facility-level points using deterministic integer-cent arithmetic.
- Publish sample, facility, unrelated-system, plan-label, exclusion, source-date, and receipt evidence beside the distribution.
Potential outliers stay traceable. Exclusions follow versioned rules and are counted rather than silently discarded.
05
Public access is not automatic indexation
An aggregate can remain available in the public explorer while its standalone page is excluded from search-engine indexation.
| Gate | Requirement |
|---|---|
| Qualifying observations | At least 30 |
| Hospital locations | At least 10 |
| Unrelated health systems | At least 3 |
| Plan labels | At least 2 |
| Largest system concentration | No more than 40% |
| Entity mapping confidence | At least 98% |
| Source-file age | No more than 400 days |
| Source health check | Within 30 days |
| Traceable facts | At least 5 |
| Fixed quality score | At least 85 of 100 |
06
Corrections and rollback
A correction begins at the source or normalized dataset and regenerates every dependent aggregate, page, API response, and CSV. A displayed number is never patched on one page while dependent outputs remain unchanged.
Publication controls can withdraw one URL, one source file, one hospital system, one payer mapping, one market, one code system, or the entire reimbursement-rate family. The last approved snapshot remains available for rollback.
Report a data issueCompare the public benchmark with your contract and 835 payments
QuickIntell can support an assisted, private analysis of contract expectations and adjudicated remits. Do not upload contracts, remits, claims, or PHI through the public contact form.