PriceOS Methodology — Complete Transparency

1. The Central Problem

⚠️ Most medical tourism comparisons are wrong.
They divide a bundle by a line item and print a savings percentage. A Cancún hospital quotes $12,000 for "knee replacement" (implant + surgeon + facility + 3 nights + physio). A US MRF gives you a facility DRG rate, a surgeon CPT, and an anesthesia CPT—separately, from different files, sometimes different entities. Dividing one by the other produces a number that is wrong and confidently presented.

Why bundles matter

Every price in this system has an explicit statement of what's inside and a completeness score (0.0–1.0). Before comparing a US price to an international price, we measure completeness:

Completeness gap = |US_completeness − INTL_completeness|
If gap > 0.15 (15%): comparison is NOT_COMPARABLE
If gap ≤ 0.15: proceed to savings range calculation

✓ This rule is the feature. A NOT_COMPARABLE result is more useful than a wrong savings percentage.

2. Canonical Bundles (Episode of Care)

For each procedure, we define the complete episode of care—every component a patient must pay for, regardless of who bills it or in which country.

Example: Total Knee Arthroplasty (TKA)

Component	Weight	Typical Cost	US Code
Surgeon fee	20%	$3,500–$6,000	CPT 27447
Facility charge	45%	$9,000–$18,000	MS-DRG 469/470
Anesthesia	8%	$1,200–$2,400	CPT 01402
Implant device	15%	$4,000–$8,000	—
Pre-op workup	3%	$400–$800	—
Post-op physio	4%	$1,200–$3,000	—
Inpatient nights	5%	$2,000–$4,000	—
Total		1.00 (100%)

Sources: CMS Inpatient Hospital Prospective Payment System (IPPS), 2024; U.S. News & World Report cost survey, 2023; Journal of Arthroplasty (Slover et al., 2016).

Why weights matter

Weights are used only to compute completeness scores. We never impute a missing price component. If a price observation includes only facility_fee (0.45) but not surgeon (0.20), the completeness score is 0.45. The comparison engine will flag it as NOT_COMPARABLE if it conflicts with a more complete international quote.

3. Completeness Scoring (The Honest Rule)

Hard Rule (enforced in code, not in documentation):

if abs(completeness(us) - completeness(intl)) > 0.15:
comparison.status = "NOT_COMPARABLE"
comparison.savings = None

Example comparison: TKA in Mexico vs. US

Price Side	Components Included	Completeness Score	Status
US Hospital MRF	Facility only	0.45	NOT_COMPARABLE
Mexico (Hospital Ángeles)	Facility + surgeon + anesthesia + implant + physio	0.85	NOT_COMPARABLE
Completeness gap: \|0.45 − 0.85\| = 0.40			> 0.15 threshold

Why this result is a feature, not a bug:
We could hide this complexity, divide $12,000 by $18,000, and print "33% savings." Instead, we report NOT_COMPARABLE and show exactly what's missing on each side. This is the single thing that distinguishes this resource from every lead-gen site in the market.

4. When Comparisons Are Possible: Savings Ranges

When completeness scores are similar enough (gap ≤ 0.15), we compute savings as a range, never a point estimate:

    savings_low  = 1 − (intl_high / us_low)

    savings_high = 1 − (intl_low / us_high)

Example: Cataract surgery (comparable bundles)

Price Side	Low	High	Completeness
US (avg)	$2,100	$3,500	0.90
Thailand (Samitivej)	$1,200	$2,200	0.92
Savings low		1 − ($2,200 / $2,100) = −4.8% (intl actually higher in worst case)
Savings high		1 − ($1,200 / $3,500) = 65.7% (intl better in best case)
Result: Savings range −5% to 66% (Wide range reflects uncertainty; negative savings means Thailand may cost more)

5. Data Sources & Provenance

US Data

Trilliant ORIA: 11,326 observations from parsed hospital MRF DuckDB files. Every observation has source URL (Trilliant ORIA DuckDB path) and vintage date (MRF publication date).
Direct MRF scrape: 14 observations from 3 hospital systems. Source URLs link to live MRF files.

International Data

Public websites: Extracted via adapters that fetch, parse, and cache raw HTML (SHA256 hashed).
Raw document archive: Every price observation links to cached raw document. Disputes can be resolved by checking the page as we saw it.
LLM-assisted extraction: Where website content is unstructured, we use Claude to parse bundle descriptions. These fields carry a confidence score (0.0–1.0); anything < 0.8 is flagged for human review.

6. Staleness & Recency

Every price observation has an observation_date—when the price was valid, not when we scraped it. Prices > 18 months old are marked stale: true and excluded from headline comparisons by default.

7. What This System Does Not Do

❌ Impute missing prices ("typical for the region")
❌ Show country averages as prices (OECD PPPs, etc.)
❌ Accept referral fees or paid placement
❌ Submit inquiry forms or impersonate patients
❌ Hide conflicts of interest
❌ Fabricate completeness scores

8. Conflict of Interest Disclosure

The maintainer of this system is based in Roatán, Honduras (Próspera jurisdiction). This may create conflicts of interest regarding Central American facilities. Any page featuring a Honduran or Central American facility carries this disclosure at the top.

9. How to Audit This

All data is committed to git. To reproduce any number on this site:

Check data/observations/*.jsonl for raw observations
Check ontology/canonical_bundles/*.yaml for bundle definitions and weights
Check etl/comparison_engine.py for comparison logic
Verify citations in this document against original sources
Check cached documents in data/adapters/cache/ to see the original website content

✓ A hostile reader can reconstruct every number on this site from the repository alone. That's the standard.