# The Great Financial Convergence — research specification v0.1.0

Research cut-off: 13 October 2026. Prototype release: v0.1.0. This is a hypothesis-led observatory, not proof that all fintechs become banks or all banks become fintechs. The quotation and its attribution to a World Bank president remain unverified and are not published as a quotation.

## Questions and falsifiable hypotheses

1. Are fintech-origin entities acquiring deposit-taking licences, or relying on sponsor banks? Count independently verified changes in legal permissions by entity-country-year. Distinguish applications, restricted authorisations and full licences. A payment licence is not a banking licence.
2. Are incumbent banks increasing digital distribution and technology partnerships? Measure disclosed digital active customers / total active customers, digital transaction shares, and dated partnerships. Definitions differ; compare within a bank before comparing banks.
3. Does digital access expand participation, especially in Africa? Track adult account ownership, mobile money ownership, digital payments, formal savings and demographic gaps. Adoption does not identify provider ownership or establish causality.
4. Does Africa follow a distinct route? Contrast country distributions, not a single continental score; control descriptively for income and connectivity. Test heterogeneity across five African subregions and North Africa.

## Source and dataset inventory

|Source / access|Unit, coverage and cadence|Purpose / status|Risks and reuse|
|---|---|---|---|
|[World Bank WDI / Findex](https://data.worldbank.org/indicator/FX.OWN.TOTL.ZS); public v2 JSON API|Economy-wave; 2011, 2014, 2017, 2021/22, 2024|2021 and 2024 account observations ingested; raw snapshots retained|CC BY 4.0 on indicator page; survey error; revisions and changing coverage|
|[Findex 2025 downloads](https://www.worldbank.org/en/publication/globalfindex/download-data)|About 148,000 adults in 141 economies in 2024; subgroup tables and microdata|Next ingestion: mobile money, payments, savings, sex/income/rural gaps|Check each file's terms and exact questionnaire; no invented standard errors|
|[Findex press release](https://www.worldbank.org/en/news/press-release/2025/07/16/mobile-phone-technology-powers-saving-surge-in-developing-economies)|Published rounded regional summaries, 2024|Editorial context: SSA account ownership 58%, 49% in 2021|Narrative release, not a country-level extract; regional scope must be preserved|
|[World Bank digital inclusion analysis](https://blogs.worldbank.org/en/allaboutfinance/digital-technology-is-unlocking-financial-inclusion)|Published rounded Findex summaries|SSA mobile money 40% in 2024, 27% in 2021; world account 79% / 74% on 2025 release basis|Older 2021 publication reported world 76%; do not splice releases|
|[IMF FAS](https://data.imf.org/Datasets/FAS)|Annual supply-side country series; banks, ATMs, agents, mobile money|Planned, not ingested|Reported accounts are not unique people; varying response coverage and definitions; inspect metadata and licence|
|[GSMA 2026 report](https://www.gsma.com/sotir/)|Industry supply-side reporting for 2025|Planned, not ingested: 30-day activity, agents, transaction value|Registered accounts are not adult owners; subscriptions can duplicate people; redistribution terms need review|
|[BCBS Digitalisation of finance](https://www.bis.org/publications/digitalisation-finance)|2024 analytical report|Conceptual architecture, not numeric input|Provides mechanisms and supervisory implications; cannot create a country ranking|
|National central banks and regulators; audited bank filings|Entity-country-event, annual bank disclosures|Planned manual evidence ledger|Start with CBK, CBN, SARB/Prudential Authority, BCEAO, Bank Al-Maghrib; global comparison via PRA/FCA, ECB and BCB. Record legal entities, not brands; keep originals and effective dates; second reviewer required|

## Metrics and denominator contracts

Account ownership: FX.OWN.TOTL.ZS, percent of people aged 15+, account at a financial institution or mobile money provider; categories overlap. Difference across waves = value_2024 − value_2021 in percentage points, only when both exact-year values exist. Never carry forward 2022 to 2021. No interpolation or imputation.

Mobile money ownership: adults reporting an account per the wave questionnaire; do not equate it with mobile banking. Both-account share, when exact compatible indicators are available, follows inclusion-exclusion: FI + mobile − any; test bounds before use. Gender gap = male minus female percentage points; income gap = richest 60% minus poorest 40%. Digital payment and saving denominators must be explicit (all adults versus account holders).

For production continental aggregates: use a frozen UN M49 Africa membership list, including Northern Africa, and matched survey year. Population-weight with compatible 15+ populations, retain weights and covered-adult share. Rest of world excludes all Africa; world includes Africa and is not an independent comparator. Do not average published percentages unweighted or subtract regional percentages. Prototype compares country distributions for Africa and elsewhere; it does not manufacture continental rates. Regional summaries shown separately use the publisher's exact labels.

Institutional convergence: keep a three-dimensional evidence matrix (permissions, distribution, infrastructure/partnerships). No composite convergence index in v0.1. Licence events: regulator, legal entity, jurisdiction, permission scope, announcement and effective date, restrictions, evidence URL and status. Partnership events: counterparty, function, start/end, evidence; classify as announced versus operating. Digital bank metrics require explicit active-user window and denominator. Missing disclosures are unknown, never zero. Survivorship, disclosure and English-language biases must be tracked.

## Caveats and inference rules

Repeated cross-sections do not follow individuals. Survey timing, modes, country coverage and population weights can change. Rounded press-release values and full-precision API observations are separate series. No confidence intervals or significance claims without appropriate design information. Access is not usage, welfare, affordability or financial health. Two trends alone cannot demonstrate causal convergence. Africa is not interchangeable with SSA or the World Bank MENA region. The prototype Africa membership is a documented explicit ISO3 set of sovereign African states; it is not yet the production M49 territories universe.

## UX and visual architecture

An editorial cover introduces a question, followed by three scroll chapters: access globally, Africa's mobile route, and institutional evidence still needed. A sticky graphic changes with chapter intersection. Keyboard chapter controls and reduced-motion handling offer equivalent access. Amber identifies Africa/SSA, pale blue identifies global/elsewhere; text also names each category.

The research workspace continues the same page: geography, exact year, search, sort, and coverage controls; ranked bars and a complete table include missing values. Country detail opens evidence and exact-year changes. CSV export contains filtered rows, units, indicator ID and source. Sources/method notes and a downloadable framework remain one click away. No choropleth until geographic boundary provenance is reviewed; no fictitious network of institutions.

## Technical architecture and reproducibility

v0.1: buildless HTML/CSS/JavaScript, locally bundled JSON, accessible SVG and native table. Browser filtering requires no account or API key. Immutable raw WDI snapshots → deterministic Node transform → checked JSON + SHA-256 manifest → versioned static deployment. Ship framework, provenance, source links, exact release number and cut-off. No runtime calls to data providers.

Production: scheduled ingestion in a separate job; raw object storage keyed by hash; validated canonical country-wave and entity-event tables in SQLite/PostgreSQL; schema migrations and release manifests; publish reviewed static extracts. Use server APIs only when dataset size or access control requires them. Fetch jobs retry transient failures and preserve last good release. Automated checks enforce ranges, null semantics, uniqueness, stable geography, source IDs, exact-year comparisons and link availability. Human research review gates narrative changes and licences. Monitor refresh failures, source revisions, accessibility and browser errors. Credentials belong in server secrets, never frontend code.

## Phased roadmap and acceptance criteria

1. Evidence prototype (delivered v0.1): verified snapshots, scrollytelling, country explorer, source-aware CSV, provenance hashes and local checks. Institution-level findings explicitly pending.
2. Comparable data foundation (estimated 1–2 weeks, researcher + data engineer): full Findex waves/subgroups and M49 mapping, matched Africa/rest weights; document licences and coverage; independent reproduction of three headline charts required.
3. Institutional ledger (estimated 2–4 weeks, researcher + regulatory reviewer): a purposive Africa/global panel with predeclared sampling, dual-reviewed licence/partnership records and bank disclosure definitions. Publish coverage, unknowns and selection biases before any comparison.
4. Expanded editorial release (estimated 1–2 weeks, designer + engineer): country maps with vetted geometry, source-linked event timeline and linked cross-filters; usability sessions, WCAG 2.2 AA review and mobile performance checks required.
5. Maintained observatory (ongoing): quarterly evidence review and source-dependent refresh, signed release notes, revision diffs, regression checks and archived snapshots. Reassess hypotheses rather than forcing the narrative.

## Release limits

v0.1 implements access exploration and published regional mobile-money context. Full continent-versus-rest aggregates, entity ledger, bank digitalisation metrics, and statistical uncertainty remain research milestones, not simulated data. This document is the implementation contract, produced before prototype code.