# Methodology

## Universe

The comparison covers all 38 countries in the OECD membership universe used for
this project. Membership is checked against the OECD's current official member
list at publication time. Country-level structural claims must be supported by a
domestic source in an official or routinely used national language wherever one
is available.

## The 14 production layers

Each country is mapped at the same 14 layers: medical education; postgraduate
training; workforce; hospitals; ambulatory and primary care; pharmacies;
pharmaceutical research; pharmaceutical manufacturing; medical devices and
capital goods; healthcare research; insurance and financing; price setting;
billing and transparency; and capital investment. `layers.csv` defines the
central question for each layer.

## The analytical dimensions

A healthcare system is a production chain, not a binary label. Evidence is
collected separately for:

1. **Ownership** — the legal owner of facilities, beds, laboratories, pharmacies,
   manufacturers, and other productive assets.
2. **Operation and employment** — the entity operating the facility and employing
   or contracting clinicians.
3. **Financing** — the origin of funds: government schemes, compulsory
   contributory schemes, voluntary insurance, households, and other sources.
4. **Collection, pooling, and purchasing** — who collects contributions, pools
   risk, and pays providers.
5. **Price formation** — who sets or negotiates hospital budgets, physician fees,
   drug prices, device prices, and patient cost sharing.
6. **Capacity and capital control** — who authorizes facilities, beds, major
   equipment, training places, capital projects, and geographic distribution.
7. **Regulatory control** — licensing, quality, benefit design, coverage mandates,
   and entry rules.
8. **Public production and innovation** — public research, teaching, procurement,
   manufacturing, laboratories, blood products, or other supported production
   links.

“Vertical integration” is reported only when evidence connects multiple adjacent
links under common public ownership, operation, employment, purchasing, or
binding planning authority. General regulation alone is not vertical integration.

## Quantitative rules

- Every percentage names its numerator, denominator, sector, and data year.
- Hospital shares, hospital-bed shares, long-term-care shares, and facility shares
  remain separate measures.
- Government financing and compulsory financing remain separate. Compulsory
  private insurance is not silently relabeled government ownership.
- Missing components are never averaged into a composite.
- No cross-country rank or index is published until like-for-like denominators and
  defensible weights exist. The default output is a multidimensional profile.
- A value is `Unknown` when a current, comparable source has not been verified.

## Evidence hierarchy

1. **Tier 1:** law, regulation, official statistics, ministry, national statistics
   office, statutory payer, regulator, or official administrative dataset.
   Direct Eurostat/OECD administrative data may supply a harmonized financing
   observation, but never substitutes for domestic evidence on ownership,
   operation, employment, price authority, or capacity control.
2. **Tier 2:** national audit institution, parliamentary research service, public
   research institute, public university, or official professional body.
3. **Tier 3:** peer-reviewed research using identified primary data.
4. **Discovery only:** OECD summaries, international observatories, search-engine
   syntheses, journalism, and secondary explainers. These can locate domestic
   evidence but do not establish a country structural claim by themselves.

## Original-language record

For each major claim, the source ledger preserves:

- institution and original title;
- publication date, data year, and retrieval date;
- a short exact passage in the source language;
- a faithful English translation;
- the relevant local institutional term, its literal translation, and its
  functional meaning;
- direct URL, evidence tier, verification state, and confidence.

Translation aids and multilingual search systems may assist discovery. The
published claim is checked against the cited source itself.

## Publication states

- `discovered`: a plausible source lead exists but has not been checked directly.
- `verified`: the cited primary or authoritative source directly supports the
  observation or claim.
- `partial`: the source supports a bounded finding, but one or more definitions,
  ownership classes, jurisdictions, or denominators remain unresolved.
- `conflicted`: authoritative sources disagree or use incompatible categories.
- `unknown`: no defensible current value has been found.

Accepted findings can be verified, partial, conflicted, or unknown; their status
is always published with them. A partial record is never silently promoted to a
complete national estimate. Pending research candidates are excluded from the
published observation files until reviewed.

## Versioning, time series, and corrections

The operational database stores immutable releases, stable series identities,
observation versions, source versions, source checks, methodology breaks, and
explicit revision links. The public `research-releases.csv` and `series.csv`
files expose the current release and series registry. Older observations remain
available when a newer year is added; they are not overwritten merely because a
new vintage exists.

Reusing a release identifier with different input hashes is rejected. A changed
definition, corrected value, or reclassified institution is published as a new
release and linked to the prior observation where applicable.

## Gap and candidate workflow

Every unsupported layer, missing master variable, inaccessible source, stale
vintage, incompatible denominator, and unresolved contradiction can be stored as
an explicit gap. `data-gaps.csv` is the public work queue. It distinguishes an
open gap from one for which a candidate source has been found.

Candidate evidence is staged separately from accepted observations. Independent
review records whether a second researcher reproduced, qualified, contradicted,
or could not verify the candidate. A candidate does not become a published fact
merely because an agent found a plausible number.