# Native website edition / atlantic-r6

## Research extension / exhibits 30-39

The edition now contains 39 exhibits. WGI institutional estimates and published 90% intervals, SIPRI 2025 military data, WDI/ILO employment composition and separately defined national AI surveys extend the original comparison. Restricted Polity/IDEA extracts and complete third-party resources are not included in the public archive.

The Canadian proximity baseline uses 29 indicators, eight equal domains, the fixed 29-country panel and an EU27 member-median profile. Numeric Gower distance averages normalized absolute differences within domains and then across domains. The relative index is 100 * d(Canada,US) / (d(Canada,US) + d(Canada,EU)); 50 means equal distance. This is neither welfare nor a probability.

All countries share one year per indicator. Ordinary common years fall within 2020-2024; PISA 2025 is a declared survey exception. AI survey estimates remain outside the score because their populations, questions and reference windows differ. Measurement uncertainty and researcher-choice sensitivity are not interchangeable.

The extension folder contains the exact inputs, full research results, per-figure CSVs, 24 model specifications, 512 deterministic domain-weight scenarios and source-provenance receipts. The service-corner inset uses the same ternary geometry as the complete triangle, without moving countries.

The Atlantic Seam upgrades exhibit35. Its 56 directed routes transfer effective weight between two domains, preserving total weight and all other effective weights. Donors can lose half their starting weight and recipients can at most double theirs, within the multiplier ceiling3. Exact roots, full path samples and definitions are supplied in seam-thresholds.csv, seam-paths.csv and seam-method.json. These are choices of emphasis, not changes to country observations or probability intervals.

Three additional instruments retain separate stable identities without renumbering the39 exhibits. The metric spotlight compares the three headline economies and labels inputs and age structure descriptively. The Rank Weave follows29 countries through descending output/spending/longevity ranks in a matched2019-2023 year. The Relationship Atlas provides fixed24-variable2019-2023 and21-variable2019-2024 panels, plus the29-variable mixed-year profile. Annual correlations compare countries within each year, not five or six yearly observations within one economy. Spearman ties use average ranks; undefined values are not zero. Leave-one-country-out ranges are sensitivity diagnostics, not confidence intervals.

After extracting the evidence archive, open its extension directory and run `node reproduce/run.mjs`. No dependencies or network requests are needed. This reproduces the baseline, domain contributions, nearest countries and sensitivity ranges. The methods and exclusions are also embedded in extension/research.json.

## Original comparison / exhibits 01-29

This edition preserves the original 29 exhibits and adds 145 contextual display records from the same frozen World Bank extracts. Germany, France, Sweden, Poland and Spain supplement the health-spending/longevity and child-mortality views. Their years and price bases match the primary series; the five countries are inside the EU aggregate, not an independent control group.

The revised geography is a categorical Atlantic locator, not a population choropleth. The investment slope connects two observed shares without reconstructing the intervening path. The PISA subject profiles retain all 87 published confidence intervals on a common score scale; links compare subjects, not time. The age-composition field uses one cell per percentage point, retains fractional cells and derives under65 as 100 minus the age65plus share. The restored chord retains six directional exporter reports and arrowheads; arc sizes summarize the displayed incoming and outgoing links, not world shares.

US data marks remain black, Canada red and the EU blue. Context paths are neutral until inspected. Country flags use their own national-symbol colors. Narrative panels may have an ice-like elevated treatment; plotting surfaces retain contrast. Display labels use no more than two decimals; years are never thousands-grouped. Source precision remains in JSON and CSV.

All 29 original CSVs, source specifications, comparison data, article and source-specific attribution are downloadable alongside the extension. Package-relative acquisition paths below describe the supplied research archive, not additional files in this method download.

## Source research definitions

# Methods, definitions and reproducibility

## Scope and source vintage

This is a descriptive comparison of the current 27-member European Union, Canada and the United States. The evidence snapshot was collected on 17 September 2026. The EU is a union of countries; Canada and the United States are federations. A common visual treatment does not make their institutional structures identical.

Historical World Bank observations run from 2000 through 2024. Every exhibit states its own period. R&D ends in 2023; the joint health-expenditure/life-expectancy comparison ends in 2023; public education expenditure uses 2021; bilateral goods trade uses 2025; PISA is the 2025 assessment published in 2026; the precise release day was not independently reverified during the native audit. There is no fabricated all-current comparison year. IMF Fiscal Monitor data are from the April 2026 vintage and stop at 2024 in the figures; historical IMF entries may include staff estimates.

The motivating [Canada aviation article](https://www.schym.de/blog/canada-is-flying-again-just-less-to-the-united-states) supplies a question, not a causal premise. Security screenings do not identify travelers’ motives. The economic and trade evidence here is acquired independently.

## Files and analytical lineage

`sources.csv` lists the primary publishers, exact URLs, retrieval and reference dates, reuse notes and original-file hashes where a file was available. `claim-ledger.csv` separates each exhibit’s interpretation from its numerical evidence and records the limitations beside the claim. The dated policy events have separate entries.

`data/figure-specs.json` is the frozen display-level specification. Each figure has one matching exact-value CSV in `../02_figures/exact-values/`. Source extracts are retained in `data/source/research-macro/`, `data/source/research-health/` and `data/source/research-trade/`; their individual research notes preserve detailed extraction decisions. Acquisition and calculation scripts remain beside the corresponding data so that their relative paths work. The macro browser installation and development logs are not research inputs and are excluded.

The source-document acquisition manifest records the official PDF/workbook URLs and the hash of the version examined. Numerical PISA summaries are included, together with their extraction code. The full CMEC copyrighted report is obtained directly from its publisher; the report and its full-text extraction are not republished inside this package. The same acquisition route is provided for the IMF methodological PDF. This choice does not affect offline reproduction of the article from the frozen numerical extracts.

## Geography and the EU denominator

The fixed membership is Austria, Belgium, Bulgaria, Croatia, Cyprus, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain and Sweden. The United Kingdom is excluded throughout.

World Bank `EUU` series are retained where they provide a defensible source aggregate. For population, nominal GDP, real PPP GDP and territorial CO₂ totals, the source EUU values were checked against the sum of these 27 countries across the available 2000–2024 observations. Differences were zero or numerical rounding. Aggregation is nevertheless indicator-specific: a published EUU label does not imply one universal weighting rule.

Three exceptions are central to the story:

- **PISA:** every member country is shown; there is no invented EU pooled student mean or distribution. OECD average is not substituted for the EU.
- **Public education expenditure:** the WDI EUU value is the median of 26 reporting members in 2021. It is explicitly labeled an EU member median, not a GDP-weighted bloc expenditure ratio.
- **Fiscal expenditure:** the EU figure is an author calculation from all 27 national general-government ratios, weighted by each country’s nominal GDP in the IMF WEO. It is not the budget of EU institutions.

Country-distribution figures are unweighted collections of countries. They describe cross-country heterogeneity, not the distribution of individual incomes or life spans. Luxembourg and Ireland remain in the GDP-per-person distribution; their values should be read with the known sensitivity of national output to multinational activity and cross-border labor.

Natural Earth generalized boundaries supply geographic orientation. The native Atlantic locator colors membership categorically and shows population separately, without obscuring geography with circles. Map area is not population, and the clipped projection is not a tool for measuring distances or adjudicating borders.

## Output, recovery, investment and public finance

Nominal GDP is divided by one trillion to express current US dollars in trillions. It measures economic scale at market exchange rates; it does not measure real growth. GDP per person uses World Bank constant 2021 international dollars at purchasing power parity. It measures average output per person, not household median income, disposable income, consumption, hours worked or a composite welfare score.

The recovery chart applies

`index(country, year) = 100 × real_PPP_GDP_per_person(country, year) / real_PPP_GDP_per_person(country, 2019)`.

The initial values are therefore all 100, while initial income levels differ. The latest total-GDP and per-person-GDP growth comparison uses the two published annual real-growth indicators. Their difference is not treated as an exact additive estimate of population growth, because compounded growth ratios are multiplicative.

Gross fixed capital formation includes private and public fixed investment. R&D expenditure is a distinct measure of resources committed to research and experimental development. The two measures can overlap and are never added as mutually exclusive components. The R&D bump chart ranks only the three specified economies at the four stated dates. A higher rank does not imply a proportionally larger R&D gap. Exact expenditure shares remain available with the ordinal path; ties would share a rank if present.

Government final consumption covers government-produced or purchased services and related consumption. It excludes most transfers and is not total general-government expenditure. The total-expenditure comparison instead uses IMF Fiscal Monitor ratios. For EU27:

`EU_ratio(year) = Σcountry [nominal_GDP(country,year) × expenditure_ratio(country,year)] / Σcountry nominal_GDP(country,year)`.

All 27 members contribute. This weighted national-government calculation does not add a separate EU-institutions budget or perform an additional cross-border consolidation adjustment. The source countries may differ in cash/accrual accounting conventions. Military expenditure is the WDI/SIPRI concept and should not be equated mechanically with NATO’s defense-expenditure definition. Government consumption, health expenditure, military expenditure, R&D and investment overlap; the story does not construct a false additive national budget from them.

## Health and childhood

Life expectancy at birth is a period statistic under current mortality conditions, not a forecast of the realized life span of a newborn. The World Bank EU aggregate is used consistently. It can differ slightly from Eurostat’s separately constructed EU life table; the two are not silently spliced.

Health expenditure per person uses **current international dollars at PPP**. The 2023 country comparison adjusts for purchasing power. The time trajectory also contains price changes, so it is not a real expenditure growth series. Current health expenditure excludes capital investment in health buildings and equipment. The connected scatter preserves dates and both observed measures, but it is not an estimated production frontier or causal efficiency model.

Life expectancy reflects health care, prices, disease burdens, inequality, behavior, violence, transport risks, living conditions and earlier public-health conditions. This small ecological comparison cannot allocate causal responsibility across those mechanisms or across current governments. No regression through three aggregate points is presented as identification.

Under-five mortality is the probability of dying before age five per 1,000 live births, drawing on UN inter-agency estimation through WDI. It is not a count of all child deaths, and its denominator is not the population of children alive in a calendar year. Official model-based estimates and later revisions remain possible; the figure uses the source estimates without manufacturing confidence intervals that were not acquired.

The financing bars contain domestic government finance, direct out-of-pocket payments and a residual:

`other_financing = 100 − government_share − out_of_pocket_share`.

The residual is called “other financing,” not “private insurance.” Each bar sums to 100 within floating-point tolerance. A low out-of-pocket percentage does not by itself imply a low out-of-pocket amount or equal financial protection. Sex-specific life expectancy uses the published female and male series on the same year and definition.

## PISA skills and educational attainment

PISA 2025 was verified against the OECD release record and the official [CMEC national report](https://www.cmec.ca/Publications/Lists/Publications/Attachments/448/PISA_2025_Public_Report_EN.pdf). The source is not a relabeled 2022 assessment. Direct OECD downloads presented a browser security check; the independently available CMEC official report was used. No security check was bypassed.

The mean-score tables are B.1.2 (science), B.3.3 (reading) and B.3.4 (mathematics). The percentile tables are B.1.5 (science), B.3.5 (reading) and B.3.6 (mathematics). The extraction records the original table, subject, country, rounded mean, standard error and 95% confidence limits. The subject matrix displays the mean and confidence interval; the mathematics plot carries confidence whiskers. Confidence intervals are statistical intervals, not a range of pupil scores.

PISA assesses the skills of tested 15-year-olds. These figures reproduce official summary estimates; this is not a new microdata analysis with plausible values and replicate weights. They do not identify causal school quality. Canada and the United States require response/sampling cautions. Canada’s documented participation and nonresponse problems can affect representativeness. An absent caution flag for another country is not a certification that it has no data-quality issue. Close point ranks should not be interpreted as statistically established ordering.

The reading-distribution exhibit shows source P10, P25, median, P75 and P90 for selected countries. Whiskers represent P10–P90, not Tukey outliers or confidence limits. No individual records or smooth densities are synthesized from these summaries. Canada and the United States both have a displayed reading mean of 490, while their lower and upper percentiles differ. Published percentile differences can differ by one point from subtraction of rounded displayed endpoints. The internally inconsistent OECD-average science-percentile row is preserved in the source extract but excluded from charts and derived conclusions.

The age 25–34 attainment comparison is separate from PISA. EU27 uses Eurostat ISCED 5–8; Canada uses the published tertiary total in Statistics Canada Table 37-10-0130-01; the US uses associate degree or higher from the 2024 Current Population Survey. For the US, counts in age 25–29 and 30–34 are pooled: 23,653 thousand with associate, bachelor, master, professional or doctoral degrees divided by 44,880 thousand total, yielding 52.7028%. Some-college-without-degree is excluded. Canada’s published total is 69%; the separately rounded component categories should not be summed in its place.

The age and broad credential boundaries align, but survey coverage, institutional populations and qualification mappings differ. Short-cycle qualifications are particularly relevant to Canada’s high attainment share. This is a comparison of attained credentials, not a precise education-quality ranking or an OECD harmonized re-estimate.

## Demography and emissions

The proportion aged 65 and over is an age-structure measure, not a direct measure of fiscal dependency. The total fertility rate is a period synthetic-cohort statistic, not completed lifetime fertility. The displayed 2.1 reference is a conventional approximate replacement benchmark; replacement conditions vary with mortality and the population context.

CO₂ is territorial emissions excluding land use, land-use change and forestry, based on the published WDI/EDGAR-related indicator. It is not consumption-based emissions and does not allocate emissions embodied in imports to consumers. The output–carbon trajectory therefore shows territorial decoupling patterns, not the full global footprint of domestic consumption. Aggregate and per-person measures are not silently interchanged.

## Trade, mirrors and policy status

Goods-trade evidence uses a common 2025 year and the European Commission DG Trade factsheets dated 20 May 2026. The source combines Eurostat Comext for EU figures and IMF-labelled national figures. Amounts are already in current EUR million; the displayed values divide by 1,000 to obtain EUR billion. There is no author currency conversion or mixing with the alternative USD extract.

Every directed bilateral ribbon uses the exporting region’s export column. Opposite directions remain separate; corresponding importer records are not averaged into them. The six directed flows are not claimed to form a balanced national-accounts system. Services and investment are excluded from the ribbons.

For each exporter, the Sankey divides world merchandise exports into the other two comparison destinations and a rest-of-world residual:

`ROW = reported_world_exports − exports_to_comparator_1 − exports_to_comparator_2`.

Source widths and destination widths use EUR, so their sums are meaningful. The origin-specific share is `100 × partner_exports / reported_world_exports`; percentage labels are never added across different countries as if they shared a denominator. EU world exports mean extra-EU exports, excluding intra-EU trade. This compares the bloc’s external dependence with national external dependence.

The mirror exhibit compares an exporter’s export value with its partner’s reported import value for that directed pair. Differences can reflect valuation, attribution, transit/re-exports, timing and revisions. They are not confidence intervals, unrecorded economic losses, missing goods or evidence of misconduct. These components have not been separately decomposed.

The policy timeline records selected institutional events and their actual status: provisional application, legal entry into force, signature, strategy roadmap, royal assent, enactment, framework commitment or negotiations. It is not an exhaustive live tariff inventory. Announced procurement, intended investment and headline policy packages are not recorded as realized government expenditure. No observed macroeconomic difference is attributed causally to a selected policy event.

## Reproduction and validation

1. Open the self-contained article offline. Its initial evidence, sources and tables do not need a remote data service.
2. Rebuild figure data and exports from `data/figure-specs.json`, the archived numerical snapshots and the publication builder described in `README-reproduce.md`.
3. To reacquire an original document, use `code/acquire_source_documents.py`. Compare its resulting hash with `data/source-document-acquisition.json`; a revised publisher file is a new vintage and must not silently replace the frozen one.
4. To refresh WDI, run the relevant archived `acquire.py` script. To rebuild fiscal aggregates from the stored IMF JSON, run the macro `enrich.py`. Refreshing can change published results and should occur in a separate copy of the release.
5. To rerun the PISA extraction, download the CMEC PDF, create `pisa2025.txt` with `pdftotext -layout`, then run the archived `extract_pisa.py`. To verify the US attainment calculation, run `extract_attainment.py` with Python and openpyxl.

The source-quality report verifies duplicate observation keys and fixed EU membership aggregates. The exhibit semantic audit verifies real year ordering, finite coordinates, component totals, percentile ordering, mean/interval consistency and trade-flow conservation. The strict chart plan and compiled design contract record the exact geometry and fallback decisions. These checks establish internal consistency; they do not certify the absence of measurement error in official statistics.

No composite winner score, fitted causal model, fabricated microdata, synthetic density, inferred trade motive or forecast-as-observation is included. The defensible result is a set of conditional comparisons whose ranking depends on the outcome, population, unit and denominator being examined.