Methodology and data audit

1. Data definitions

Stored values are always base units (US dollars, persons, percent) — never "billions". Stocks are levels at the end of the period; flows are amounts during the period; rates and indices are dimensionless.

2. Accounting rules and the double-counting problem

Debt is a liability of the borrower and a financial asset of the creditor. Adding "global debt" to "global financial assets" counts the same claim twice, so the platform never publishes such a sum. Every series is tagged with one accounting view and world totals are computed per series only.

• Additive within a view: economic flows (GDP, current account) can be summed across countries for one period, subject to the coverage disclosure.
• Never added across views: Debt liability, Financial asset stock, Equity liability, External liability (debt and equity), Net position. Debt liabilities are not added to financial assets; equity liabilities are not added to debt; gross external liabilities (IIP liabilities) mix debt and equity and are added to neither; net positions are differences and are not summed with gross stocks.
• Money supply is a stock of monetary liabilities (tagged debt liability), not an asset of the economy.
• External position: net IIP = external assets − external liabilities. A world sum of net IIPs should be ≈ 0; the observed discrepancy is a data-quality indicator, not a quantity.
• Rates and ratios (percent of GDP, growth, inflation) are never summed; world rows show counts and coverage only.

3. Source priority

Tier 1: international and national official statistics (IMF, World Bank, BIS, OECD, central banks, statistical offices). Tier 2: official security and ownership records (treasury foreign-holder surveys, depositories, regulatory filings). Tier 3: licensed commercial sources — not used in the MVP; if added, they are never silently combined with official figures. Within a metric the primary source wins; fallbacks are used only where the primary has no value and the source is recorded per observation.

4. Transformation rules

• Scaling to base units. Sources publishing "billions" (IMF WEO) are multiplied by 10⁹ at ingestion; the factor is recorded in the series file (transformations and attributes.sourceScale). Display scaling (T/B/M) is presentation only.
• Percent of GDP. Annual stock at end of period over the annual GDP flow of the same year. When quarterly data arrive, the convention will be quarter-end stock over trailing-four-quarter GDP. Published ratios (e.g. WEO GGXWDG_NGDP) are shown as published.
• Per capita. amount ÷ population for the matching period. Share of world. amount ÷ sum of reported eligible amounts × 100, with the reporting coverage disclosed.
• Derived series (broad_money_usd, gg_gross_debt_usd, niip_pct_gdp, fx_rate_lcu_per_usd) are computed by the pipeline from official inputs of the same vintage using the formulas in the definitions table, never in the browser.
• Rates of change. YoY = amount_t / amount_t−1 − 1; CAGR_n = (amount_t / amount_t−n)^(1/n) − 1. A change is null (not zero) when an input is missing or a documented series break lies between the two periods.

5. Currency conversion

US-dollar series are taken as published by the source (WEO converts at period-average market rates). Where only a domestic-currency stock is published, conversion uses the implied WEO rate gdp_nominal_lcu / gdp_nominal_usd for the same period (series fx_rate_lcu_per_usd, cross-checked against World Bank PA.NUS.FCRF). PPP values use the ICP-based WEO PPPGDP series. Constant-dollar views are not offered in the MVP.

6. Seasonal adjustment

Not applicable: all MVP series are annual. Monthly holder data (TIC) are shown as published, not seasonally adjusted.

7. Missing-data treatment

No imputation. An economy without a value for a period is omitted from the globe, map and table for that period and from the world total. Every world total discloses the number of reporting vs eligible economies and the GDP-weighted coverage share; an "estimated world total" is shown only when an explicit, documented gross-up method exists (none in the MVP, so it is always shown as none).

8. Series breaks

Documented breaks (e.g. GDP rebasing) are stored per country in attributes.extra.seriesBreaks. Change rates that would cross a break are null, the break is flagged on the quality panel, and charts draw a vertical break line.

9. Estimates

Each observation carries a status: final, preliminary, estimated (source-flagged estimate) or forecast (official projection after the source's last actual year). Statuses are rendered distinctly everywhere (solid / dotted / hatched / translucent) and "Latest" views never include forecasts unless the forecast toggle is on.

10. Revision and vintage history

A snapshot is one immutable pipeline run (YYYY-MM-DDTHHMMZ). A source vintage is the publisher's edition (WEO 2026-04, WDI 2026-07-13). Each snapshot keeps the prior vintage of revised series, so every value can show its previous figure and delta (/v1/vintages, the Updates feed and the revision chips). Snapshots are never rewritten; corrections produce a new snapshot and an update event.

11. Model documentation

platform_debt_dynamics_v1

Debt-dynamics identity for general government gross debt (percent of GDP), scenario platform_debt_dynamics_v1 in /v1/projections:

d_t = d_(t−1) × (1 + i_t) / (1 + g_t) − pb_t        (fractions of GDP; stock-flow adjustment = 0)

• d₀ = last observed debt ratio; g_t = nominal GDP growth from gdp_nominal_lcu including WEO forecast years.
• pb_t = primary balance; interest = primary balance − overall balance; i_t = interest_t / d_(t−1) on the model path.
• Beyond the WEO horizon the last growth, primary balance and interest assumptions are held constant for five years.
• Band: p10 / p90 recompute the path with nominal growth +1 pp / −1 pp every year. It is a transparent sensitivity band, not a statistical interval; no backtest exists for this version. Output is graded D (platform model) and is never mixed with official forecasts.

Official forecasts (official_weo) are shown exactly as published by the source with no platform computation.

12. AI usage statement

AI is used only to write optional plain-language explanations of update events (aiSummary) and to help classify source documents. It never produces, edits or publishes a numeric value, a forecast or a quality grade. Every number on this site comes from a cited source file whose hash is recorded in the snapshot manifest, or from a documented deterministic formula. AI text is always labelled as such.

13. Quality grades

Grades are derived from source authority, observation status, derivation, coverage, estimated share and series breaks. A grade never replaces the underlying limitations, which are shown alongside it.

14. Known limitations

• External debt coverage. Gross external debt is published by the WEO for emerging and developing economies only; advanced economies are largely absent, so world totals for this series carry low coverage by design.
• TIC custodial bias. U.S. Treasury holdings are attributed to the country of the custodian (Belgium, Luxembourg, Cayman Islands, United Kingdom), not the beneficial owner; the issuer total is total foreign holdings, not marketable debt outstanding; the June benchmark survey can revise the whole history.
• M1 / M2 / M3 comparability. Monetary aggregates follow national definitions and are not comparable across countries; only broad money as compiled in the IMF MFS framework is used for cross-country views, and even that differs in scope. The national M1/M2/M3 series listed in SPEC §21 are catalogued but not yet ingested (no IMF MFS_MA native aggregates in the current snapshot); they appear disabled in every metric selector rather than being estimated.
• IIP estimates for non-reporters. International investment position data depend on national reporting; aggregates may include IMF estimates, and several economies have no IIP at all.
• COFER confidentiality. Reserve currency composition is published only in aggregate; country-level holdings are confidential and are not shown.
• Holder data scope. The MVP has one bilateral holder dataset (U.S. Treasury securities). Holder structure for any other issuer is reported as unavailable, never estimated.

15. Licenses

• International Monetary Fund — World Economic Outlook (WEO) database: © International Monetary Fund. https://www.imf.org/external/terms.htm
• International Monetary Fund — International Investment Position (IIP): © International Monetary Fund. https://www.imf.org/external/terms.htm
• International Monetary Fund — Monetary and Financial Statistics — Monetary Aggregates (MFS_MA): © International Monetary Fund. https://www.imf.org/external/terms.htm
• World Bank — World Development Indicators (WDI): CC BY-4.0 (https://datacatalog.worldbank.org/public-licenses#cc-by)
• World Bank — International Debt Statistics (IDS): CC BY-4.0
• U.S. Department of the Treasury — Treasury International Capital — Major Foreign Holders of Treasury Securities: U.S. Government work; public domain.
• World Balance Sheet — Platform-derived values and projections: Derived from the cited official sources; subject to their terms.
• Natural Earth — Admin 0 – Countries, 1:110m: Public domain

Platform-derived values inherit the terms of their inputs. See Updates for what changed in the current snapshot.