Stratos governance diagnostic. Version 0.1.
This document catalogues every data source used by Stratos, with its access pathway, refresh frequency, licence, known biases, coverage profile, and fallback options. It exists for three reasons: contributors building ingestion need a single reference, methodology reviewers need to assess source credibility, and users need to understand what is feeding the numbers they are reading.
The audit is updated whenever a source is added, removed, or materially changed. Source-by-source decisions about inclusion are tracked in the issue tracker under the data-source label.
Three principles govern source selection.
Public accessibility. Every source must be accessible without payment or institutional credentials. Sources requiring registration are acceptable if the registration is free and open. Paywalled sources are excluded even where their data would be valuable.
Licence compatibility. Source licences must permit the use case (aggregation, transformation, public display, derived products). Sources with non-commercial restrictions are flagged and their outputs are isolated so they do not contaminate the broader product's commercial viability.
Bias documentation. Every source has known biases. Stratos documents these biases publicly rather than pretending they do not exist. Sources whose biases cannot be characterised at all are excluded.
These are the foundational data sources. Stratos cannot operate without them.
- Coverage: 200+ economies, 1996-present
- Refresh: annual, typically September release
- Indicators used: control of corruption, government effectiveness, political stability, regulatory quality, rule of law, voice and accountability
- Access: World Bank Data API, also bulk CSV download
- Licence: CC-BY-4.0 (Creative Commons Attribution 4.0)
- Known biases: indicators are constructed from aggregated perceptions surveys with country-expert weighting. Survey populations are not random and may overrepresent business and expatriate views. The aggregation methodology has been critiqued for hidden weighting decisions.
- Mitigation in Stratos: WGI indicators are never used as sole evidence for any score. They contribute to composite dimensional scoring alongside indicators from other sources. Cross-source validation flags significant disagreement.
- Fallback: V-Dem governance indicators provide partial overlap if WGI becomes unavailable or unreliable.
- Coverage: 200+ countries, 1789-present (modern coverage stronger)
- Refresh: annual, March release
- Indicators used: liberal democracy index, electoral democracy index, civil liberties, judicial constraints on executive, legislative constraints on executive, media freedom, civil society repression, regime classification
- Access: V-Dem website, R package, bulk download
- Licence: CC-BY-SA-4.0 (Creative Commons Attribution-ShareAlike 4.0). Note the ShareAlike requirement: derivative works that incorporate V-Dem data must be released under compatible licences.
- Known biases: country expert ratings have known cross-country comparability challenges. V-Dem methodology addresses this with item response theory modelling, but expert pool composition still varies. Coverage of small states is uneven.
- Mitigation in Stratos: V-Dem regime classification is used as the primary input to Stratos cluster assignment but is not the sole input. Capacity indicators supplement classification at cluster boundaries.
- Fallback: Freedom House and Polity V provide partial regime-classification overlap with different methodologies.
- Coverage: 190+ countries, varied historical depth
- Refresh: typically annual, occasionally delayed
- Indicators used: Human Development Index, Inequality-Adjusted HDI, Gender Inequality Index, Gender Development Index, planetary pressures-adjusted HDI
- Access: UNDP website with bulk download
- Licence: open with attribution requirement (CC-BY-3.0-IGO)
- Known biases: HDI component metrics (education, life expectancy, income) carry the biases of national statistical reporting. UNDP applies harmonisation but cannot fully correct for source variation.
- Mitigation in Stratos: HDI is used as a composite outcome indicator, not as a sole proxy for governance success. IHDI-to-HDI ratio is treated as a distributional indicator with appropriate confidence.
- Fallback: HDI can be partially reconstructed from World Bank WDI data using the UNDP methodology if UNDP releases become unreliable.
- Coverage: 195+ economies, varied historical depth
- Refresh: twice yearly (April and October)
- Indicators used: GDP, GDP per capita, inflation, unemployment, government debt, fiscal balance, current account balance
- Access: IMF DataMapper and bulk download
- Licence: open with attribution
- Known biases: IMF data integrates national reporting with IMF adjustments. Adjustments for known reporting issues (Argentine inflation pre-2015) are documented but not always perfect.
- Mitigation in Stratos: WEO data is cross-checked against World Bank WDI for major discrepancies. Where the two disagree materially, both are flagged.
- Fallback: World Bank WDI provides substantial overlap with different revision schedules.
- Coverage: 217 economies, 1,400+ indicators
- Refresh: continuous with major annual revision
- Indicators used: extensive, including service access (electricity, water, sanitation, internet), economic indicators, demographic data, education indicators, health spending
- Access: World Bank Data API
- Licence: CC-BY-4.0
- Known biases: WDI aggregates national statistical reporting with limited harmonisation. Series-by-series quality varies dramatically.
- Mitigation in Stratos: WDI indicators are selected individually based on data quality assessment. Not all WDI indicators are used.
- Fallback: domain-specific sources (WHO for health, UNESCO for education) provide partial alternatives.
These sources contribute important indicators in specific domains. Each is replaceable but the substitution would degrade quality on the affected dimension.
- Coverage: 180+ countries
- Refresh: biennial
- Indicators used: environmental performance composite, climate change indicators, biodiversity indicators
- Access: Yale EPI website
- Licence: CC-BY-NC-4.0 (non-commercial). This is a constraint. Derived products incorporating EPI data must respect the non-commercial limitation.
- Known biases: heavy reliance on satellite-derived measures favours large countries. Some indicators have limited cross-time comparability.
- Mitigation in Stratos: EPI is used in the adaptive capacity dimension. The non-commercial licence means any commercial application of Stratos must either replace EPI or isolate EPI-dependent calculations.
- Fallback: ND-GAIN environmental components and individual WDI environmental indicators provide partial substitution.
- Coverage: 185+ countries
- Refresh: annual
- Indicators used: vulnerability and readiness scores across health, food, water, ecosystem services, human habitat, infrastructure
- Access: ND-GAIN website with bulk download
- Licence: open with attribution
- Known biases: composite construction involves weighting decisions that are documented but contestable
- Mitigation in Stratos: ND-GAIN contributes to the adaptive capacity dimension. Component indicators are preferred over the composite where dimensional scoring requires specificity.
- Coverage: 180+ countries
- Refresh: annual
- Indicators used: composite CPI score
- Access: Transparency International website
- Licence: CC-BY-ND-4.0 (no derivatives). This restricts how the index can be modified or aggregated.
- Known biases: perception-based, drawing from expert and business surveys. Critiqued for circularity (countries thought corrupt are scored as corrupt because they are thought corrupt).
- Mitigation in Stratos: CPI is used alongside V-Dem corruption indicators and World Bank Control of Corruption. Cross-source agreement is required for high-confidence corruption findings.
- Fallback: V-Dem corruption indicators provide alternative measurement.
- Coverage: 38 OECD members plus selected partners
- Refresh: biennial
- Indicators used: trust in government, satisfaction with public services, budget transparency, digital government, regulatory quality
- Access: OECD data portal
- Licence: typically open with attribution, some restrictions on commercial reuse
- Known biases: OECD-focused coverage means non-OECD countries are systematically underrepresented in indicators sourced here.
- Mitigation in Stratos: OECD-only indicators are flagged as such and used only in within-cluster comparison among Cluster A countries.
- Coverage: 180 countries
- Refresh: annual
- Indicators used: composite press freedom score
- Access: RSF website
- Licence: copyrighted, public reproduction with attribution allowed for educational and research uses. Commercial use requires negotiation.
- Known biases: methodology relies on expert surveys and incident counting. Country-level scores can be sensitive to specific high-profile incidents.
- Mitigation in Stratos: press freedom is captured through V-Dem media freedom indicators primarily, with RSF as supplementary cross-check.
- Coverage: 193 UN member states
- Refresh: biennial
- Indicators used: composite e-government score, sub-indices for online service, telecom infrastructure, human capital
- Access: UN E-Government Survey website
- Licence: open with attribution
- Known biases: emphasis on central government online services may understate subnational and informal digital governance
- Mitigation in Stratos: contributes to adaptive capacity dimension alongside other digital indicators
- Coverage: 200+ countries
- Refresh: annual
- Indicators used: public employment as percentage of population, wage premium, wage compression, gender pay gap, public sector wage bill
- Access: World Bank Data API
- Licence: CC-BY-4.0
- Known biases: data quality varies substantially with national statistical capacity. Coverage gaps are large in fragile states.
- Mitigation in Stratos: contributes to execution capacity dimension with explicit coverage flagging
These sources are evaluated for inclusion but not yet committed to. They appear here so contributors know they are on the radar.
- Freedom House for cross-validation of democracy classification (CC-BY-NC, restrictive)
- Polity V for historical regime data (open access)
- Uppsala Conflict Data Program (UCDP) for conflict exposure (CC-BY-4.0)
- Armed Conflict Location & Event Data (ACLED) for political violence (license requires registration, commercial use restricted)
- Stockholm International Peace Research Institute (SIPRI) for military expenditure (open with attribution)
- WHO Global Health Observatory for health indicators beyond WDI (open with attribution)
- UNESCO Institute for Statistics for education indicators (open with attribution)
- Climate Watch / WRI for emissions data (CC-BY-4.0)
- GADM for subnational boundaries (academic use free, commercial restricted)
Each candidate is evaluated against the three audit principles before inclusion.
Stratos as a whole is released under MIT for code and CC-BY 4.0 for documentation. The combination with source licences requires explicit handling.
| Source licence type | Compatibility with Stratos | Handling |
|---|
| CC-BY-4.0 | Full | Attribution in source documentation; no further restriction |
| CC-BY-SA-4.0 | Full for derived data | Derived data products inherit ShareAlike; documented in DATA_LICENCES.md |
| CC-BY-NC-4.0 | Conditional | Affected calculations isolated; commercial users must substitute |
| CC-BY-ND-4.0 | Conditional | Source used as input but not modified; documentation flags |
| CC-BY-3.0-IGO | Full | Attribution as required by source |
| Open with attribution (custom) | Full | Source-specific attribution implemented |
| Registration-required | Conditional | Compliance with terms; redistribution of source data may be restricted |
| Copyrighted with fair-use claim | Avoid where possible | Excluded unless explicit permission obtained |
The matrix is reviewed when any source licence changes.
For dimensions with multiple overlapping sources, Stratos implements cross-source validation as a data-quality signal.
Three validation operations are performed:
- Direct comparison where multiple sources report comparable indicators. Significant disagreement triggers a data-quality flag.
- Trend comparison where sources report the same underlying variable at different frequencies or with different methodologies. Trend disagreement is a weaker signal than level disagreement but still flagged.
- Coverage triangulation where alternative sources can fill gaps in the primary source. Triangulated indicators are documented as such.
Cross-source agreement is itself an indicator. It contributes to the per-country data quality score visible on every profile.
Stratos ingestion runs on a schedule designed around source release patterns.
- Monthly: check for source updates, ingest where new data is available
- Quarterly: full re-ingestion of all sources, regenerate all derived products
- Annually: methodology review, weighting reassessment, addition or removal of sources
Source-specific schedules are tracked in INGESTION_SCHEDULE.md.
Sources can be removed from Stratos for three reasons.
Licence change that makes the source incompatible with Stratos's commitments. If a source moves from open to closed licensing, dependent indicators are either replaced or the affected calculations are suspended.
Quality deterioration if a source becomes unreliable. Documented through cross-source validation failures or methodology critiques accepted by the maintainers.
Better alternative becoming available. If a new source provides superior coverage or methodology for the same indicators, the older source may be deprecated.
Source removals are documented in the change log with reasoning.
Every dashboard page displays the sources that fed it, linked to source documentation. Country profile pages list source citations for each dimensional score. The methodology documentation includes a comprehensive source bibliography.
Attribution is not optional. It is part of the open-source commitment and part of the licensing compliance for several sources.
One question is deferred to later phases: should Stratos incorporate citizen-survey data (Afrobarometer, Latinobarómetro, Asian Barometer, Eurobarometer, World Values Survey) as a corrective to official-statistics bias.
The case for inclusion: lived experience data is the primary check against state-controlled statistics. Closed autocracies have the most distorted official data and the most informative survey data.
The case against inclusion: survey data has its own biases (sampling, response, social desirability), is often poorly comparable across countries, and is sparse in the regimes where it would matter most.
The current methodology does not include barometer data. The decision will be revisited in Phase 2 with explicit user input through the issue tracker.