How this was built, and what it will not do

Method

This platform makes claims about the future of 54 countries. That obliges it to show its working. Everything below is the working — including the parts that limit what the platform can honestly say.

The rule that governs everything

No number on this platform was invented. Every projection is either a projection published by someone else and carried through unchanged, or an arithmetic operation on observed history whose method is stated in full and reproducible by hand.

Four categories are kept separate, all the way to the screen:

• evidence an observed value, with a source, a year and a publisher.
• interpretation this platform's reading of that evidence. Set in serif so it never looks like a measurement, and always citing the values it fired on.
• projection arithmetic on observed history under a stated assumption.
• gap a known unknown. Shown as a gap, never filled.

Where the data comes from

74 indicator series covering 110,510 observations, pulled live from the World Bank Indicators API, which republishes the primary compilations of the UN Population Division, the ILO, UNESCO, the WHO, UNICEF, the IEA, the ITU, the FAO, the IMF, UNHCR, IDMC, SIPRI and the Worldwide Governance Indicators consortium. Full register on the sources page, with the database's own last-updated stamp for each.

Nothing is imputed, smoothed or back-filled. Coverage is genuinely uneven — central government debt, for example, is reported by fewer than half of these countries — and the platform shows that unevenness rather than papering over it.

How much history, and how current

The platform holds 36 years (1990–present) of annual history for every series the World Bank publishes it for — not a single recent snapshot. That full run drives the sparklines on every country page, and the scenario bands are fitted over the most recent 15 years, not the latest year.

"Latest year" is not one number across the platform, because statistical agencies publish on different cycles. National accounts land within months; household surveys and health accounts can lag two or three years. Every value on this platform is displayed with the year it was observed, and here is the honest distribution across all 74 series:

Population projections are excluded from this table: they carry a value for every future year by construction, and counting them here would make the platform look more current than it is.

How the three scenarios are built

The bands are calibrated to each country's own realised history, not to an outside imagination of what an African country might do. Acceleration is not "what if this country became Korea". It is "what if this country sustained the pace it has actually reached in its better years, for a decade rather than a year or two". That makes the band defensible, comparable across all 54, and impossible to inflate.

The procedure, in full:

1. Take annual real GDP growth and annual population growth for the most recent 15 years with paired data.
2. Subtract, year by year, to get real GDP per capita growth.
3. Compute the mean (μ) and standard deviation (σ) of that series.
4. Momentum = μ. Acceleration = μ + 0.85σ. Disruption = μ − 0.85σ.
5. Clamp each to the country's own 5th–95th percentile over the window, then to an absolute [−6%, +9%] — no country sustains a decade outside that range.
6. Compound the latest observed GDP per capita forward at that rate to 2026, 2031, 2036.

If fewer than six paired years exist, no band is produced and the country's page says so. 54 of 54 countries currently support a band.

Access measures saturate — and are modelled that way

Electricity access and internet use do not advance in a straight line. The first half of a population is comparatively easy to reach; the last stretch — remote, poor, elderly, or offline by choice — is the hardest and slowest. A linear extrapolation overshoots badly at the top end.

The model now measures how fast a country closes the remaining gap to a ceiling:

gap(t)   = ceiling − value(t)
k        = (gap_last / gap_first) ^ (1 / span)   ← share of the gap left each year
value(h) = ceiling − gap_last · k^(h − last)

The path approaches the ceiling and never reaches it, which is how every real adoption curve behaves. Acceleration closes the gap 1.5× as fast, Disruption at 0.65×. Ceilings are set where universal use is genuinely attainable: 100% for electricity access, which many countries do reach, and 98% for internet use, since the best-performing countries anywhere plateau around 97–99%.

Where a country recorded no improvement across the window, no gap-closure rate exists. The platform says so and carries the last observation forward unchanged in all three scenarios — a statement about the absence of a trend, not a forecast of stasis.

Population is not modelled here at all. It is the UN World Population Prospects projection, carried through unchanged — which is why it is identical in all three scenarios. Ten-year demographic momentum is close to fixed, and saying so is more useful than pretending otherwise.

What this method deliberately cannot do

The platform also does not assign probabilities to scenarios. A probability implies a model of the world good enough to be calibrated against outcomes, and no such model exists for ten-year national trajectories. It publishes a qualitative confidence level instead, componentised so you can see what drives it.

How confidence is calculated

Four components, scored out of 7. It answers "how much weight does the evidence bear?", never "how likely is this?".

• Indicator coverage — how much of the core spine is present (0–2 points).
• Recency — how stale the most recent observed value is. Projection databases are excluded from this test, since they carry a value for every future year and would make every country look current (0–2 points).
• Volatility — the standard deviation of growth. A steadier record makes a central path more meaningful (0–2 points).
• Evidence review — whether country-specific policy and project evidence has been read and cited (0–1 point).

Readings — and why they look different

A reading is the platform's own judgement. It may only fire when specific observed data supports it, and it must cite the values it fired on — so you can always ask "on what basis are you telling me this?" and get an indicator, a value, a year and a source.

Readings are set in the serif face and marked as interpretation throughout. They are not evidence. They are what this platform thinks the evidence means, which is a weaker claim and should look like one. Every rule fails closed: no data, no reading. A country with thin data gets fewer readings, not invented ones.

Research depth is uneven, and classified

Depth is computed from the breadth of the evidence, not its volume. Five items about energy is a narrower review than five items across five sectors, and the platform refuses to let the first look like the second. The five target areas are macroeconomic and fiscal position; infrastructure and productive capacity; human development; technology and AI; and one locally consequential sector.

11 country reaches Deep country review (DR Congo, Egypt, Ethiopia, Ghana, Côte d'Ivoire, Kenya, Morocco, Nigeria, South Africa, Tanzania, Zambia). Every other country falls short on breadth, not on interest. Promotion happens only when the published criteria are met — no country is moved up because it seems important enough to deserve it.

41 of 54 countries have had policy and project evidence read and cited; the rest carry an indicator baseline mark, and their pages state plainly that no national development plan, budget or funded project has yet been reviewed for them.

This is the honest position for a first edition, and the architecture is built for it: the curated evidence layer is a separate input that merges per country, so depth can be added without touching the engine. Depth follows the availability of evidence, never the importance of the country.

The artificial-intelligence profile, and the two states that are admissions

Every country carries a profile of 20 dimensions across four groups — physical foundation, governance, capability and consequence. Each dimension states one question, the answer this platform can support, why the question matters for 2036 rather than for now, and what evidence would change the answer.

Each dimension carries an evidence state, and the five states are not a scale of quality:

The distinction between not evident and unexamined is the reason the profile exists in this form. One is a finding about the world: we looked and found nothing. The other is a finding about this platform: we have not looked. Earlier editions collapsed them, which meant a country with nothing recorded was indistinguishable from a country nobody had researched — silence read as a result. Every country page now counts the two separately, and the summary line above the profile states how many dimensions have not been examined.

No score is produced. Twenty states are twenty states. Adding them, weighting them or ranking countries by them would manufacture a number nobody measured, which is exactly what a readiness index is.

Local-language technology, and what counts as evidence of it

A model that performs well in English and poorly in a market's working language excludes most of that market from anything built on it, and does so invisibly — the failure looks like a user problem. This platform therefore treats language capability as a measurable claim with a strict standard:

A language appearing on a vendor's supported-languages list is marketing, not a measurement, and is never accepted here as evidence of capability in it. What counts is a published benchmark on a named language, a released corpus with a stated collection method, or an evaluation by a research group. Published surveys report that four African languages — Amharic, Swahili, Afrikaans and Malagasy — are consistently represented in public datasets while over 98% are not, which means a benchmark score reported for "African languages" can be carried almost entirely by languages a minority of Africans speak.

Creative and cultural intelligence — and why it has its own rules

Creative-economy figures are the most definition-sensitive numbers on this platform, and the platform treats them accordingly. Three bodies measuring "the creative economy" in the same country routinely produce numbers differing by a factor of three, because they are not measuring the same thing:

Every creative item on this platform therefore carries its measurement frame as a required field, printed above the figure rather than beneath it. Figures from different frames are never added, averaged or compared. South Africa is shown twice for exactly this reason: at 4.3% of GDP under a cultural, creative and sports satellite account for 2021, and at 1.7% of GDP under a creative-only mapping for 2018. Neither number supersedes the other and neither is wrong.

The strictest rule is geographic. A continental statistic is never applied to an individual country. UNESCO's African fashion trade figures and UNCTAD's creative goods and services totals are real and are held on this platform — in the continental systems layer, under The Weave, attached to no country. An Africa-wide total quoted beside one national flag is the most common way a creative-economy claim becomes false, and the build fails if a continental figure is filed in a country file without an explicit note stating that it is not that country's number.

Nine further validation rules run on every creative item at build time. They reject: a programme envelope described as a delivered outcome; an ageing estimate written as though current; an employment claim with neither methodology nor stated limitation; a market-size claim sourced to commercial promotion; a copyright-industries figure described as "the creative economy"; a single subsector standing in for the whole sector; and any headline economic figure carried without its measurement limitations. Programme targets — i-DICE's projected 6 million jobs, for instance — are recorded in a separate field that the interface labels projected, not achieved, and no scenario on this platform adopts them.

The third edition of this platform reported that no country-specific creative-economy sources meeting its standard could be found. That conclusion was wrong; it rested on a single unproductive search. It is recorded as an error on the Changes page rather than quietly replaced.

Territorial scope — which 54 countries, and why

This platform models the 54 African states that are members of the United Nations. That is a dataset and model-scope decision, and it is worth stating plainly rather than leaving inside a phrase like "all African countries".

The African Union has 55 member states. The difference is the Sahrawi Arab Democratic Republic, an AU member that is not a UN member state. Sovereignty over Western Sahara is disputed.

The reason for the 54 is practical, not political. Every quantitative series on this platform — the 74 indicator series and the population projections behind every scenario — is compiled and published against UN member states. There is no equivalent series for the Sahrawi Arab Democratic Republic, and this platform does not estimate values to fill a gap. Including it would mean either inventing its numbers or showing an empty country, and the first is forbidden here while the second would misrepresent the reason for the emptiness.

Nothing in this scope decision takes a position on sovereignty or territorial status. Western Sahara is drawn on the map as a distinct hatched territory and labelled a United Nations Non-Self-Governing Territory with disputed sovereignty. It is not selectable, because there is no series to select. Erasing it from the map would be as inaccurate as colouring it in.

If comparable published series for the Sahrawi Arab Democratic Republic become available, the scope can widen and the change will be recorded on the Changes page like any other.

Cartography

Boundaries are Natural Earth 1:50m (public domain), projected to Lambert Azimuthal Equal-Area centred on 20°E 5°N. Web maps almost always ship Web Mercator, which inflates high-latitude landmasses and shrinks equatorial ones — the most common way Africa is misrepresented at a glance. Here, area on the map is proportional to area on the ground.

Uncertainty is drawn rather than disclaimed: each country carries a stipple whose density tracks its confidence level, so a country the platform knows little about is visibly less resolved. The map is only as sharp as the evidence behind it.

Western Sahara is rendered as a distinct hatched territory, labelled a United Nations Non-Self-Governing Territory with disputed sovereignty, and is not selectable. It is not among the 54 UN member states this platform covers. Erasing it would be as inaccurate as miscolouring it.

What would make this better

The architecture already separates researched evidence, editorial interpretation, engine output and published content. The next increments, in order of value:

1. Curated evidence for all 54 — national development plans, budgets, funded projects, with dates and links.
2. A change log that records why a projection moved, preserves the superseded version, and shows the reader what changed.
3. Forecast accuracy tracking: scoring the 2031 paths against outturns when they arrive, without silently rewriting the originals.
4. Sub-national data for the largest countries, where a national average conceals more than it reveals.