downstream outcomeobserved
- Raw value
- 14949.03articles per million people
- Year
- 2023. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (IP.JRN.ARTC.SC)
- History
South Africa
No dimension reaches 50 on usable evidence. Raise Experimentation first, at 17.0. Trust needs more evidence: confidence is 0.26.
Raise, lowest first
| Dimension | Confidence | Score |
|---|---|---|
| Experimentation | 0.53 | 17.0 |
| Building | 0.57 | 29.4 |
| Learning | 0.48 | 32.3 |
| Adaptability | 0.62 | 38.3 |
| Anticipation | 0.46 | 42.8 |
Measure before managing
| Dimension | Confidence | Score |
|---|---|---|
| Trust | 0.26 | 83.5 |
| Shared Purpose | 0.29 | 43.4 |
| Coordination | 0.39 | 79.0 |
| Agency | 0.41 | 57.8 |
These lists are the country's capability agenda, computed from the scores. Read the full agenda. It names comparable evidence from other countries and the 18 gaps in the registry.
Why this country is included: Unequal, institutionally complex middle-income comparison case.
Anticipation42.8Confidence0.46usableHow capable is the country of identifying and preparing for emerging change?
South Africa
Read South Africa beside up to three other countries, down to the indicator behind every row.
- Usable or good evidence
- Thin evidence, hollow point. The dashes open further as confidence falls.
- The point still sits at the score, because confidence never moves it.
- Click any axis to see every country on that dimension.
A score from 0 to 100 is a position inside the frame that all the benchmark countries set: zero is the weakest on a dimension and 100 is the strongest. A score of 10 puts a country near the floor of that frame and does not mean 10 percent of a capability.
| Trust | 0.26 | 3 of 7 | 0.75 | no trend | 51.0 | -32.5 | |
| Coordination | 0.39 | 3 of 5 | 0.81 | no trend | no data | no estimate | |
| Agency | 0.41 | 3 of 5 | 0.72 | no trend | no data | no estimate | |
| Shared Purpose | 0.29 | 2 of 6 | 0.92 | +8.6(2)1 at edge | 44.0 | +0.6 | |
| Anticipation | 0.46 | 2 of 4 | 0.96 | +5.0(2) | 45.0 | +2.2 | |
| Adaptability | 0.62 | 4 of 6 | 1.00 | no trend | no data | no estimate | |
| Learning | 0.48 | 3 of 4 | 0.69 | no trend | no data | no estimate | |
| Building | 0.57 | 4 of 6 | 0.90 | -1.9(3) | no data | no estimate | |
| Experimentation | 0.53 | 6 of 9 | 0.88 | +1.9(3) | no data | no estimate |
How to read this page
Each dimension has a score from 0 to 100. It shows where this country sits among all 125 countries in the comparison frame, not a percentage. Confidence sits beside the score. A dashed radar edge and hollow point mean the evidence is thin. The records below show each indicator, its raw value, year and source.
A panel estimate, when present, is a separate judgment about what the indicators miss. It does not change the score or confidence. The panel's provenance is shown beside the estimate.
- direct capability measuredirect capability measure Measures the thing itself.
- capability inputcapability input Measures something that supports the capability.
- downstream outcomedownstream outcome Measures a result that usually follows from the capability.
- perception proxyperception proxy Records what people or experts say, not what they did.
Confidence
- very thin0.00 to 0.25
The score rests on one or two indicators. Do not quote it on its own.
- thin0.25 to 0.45
A minority of indicators, or evidence old enough to have moved. Read with care.
- usable0.45 to 0.65
Enough evidence to compare countries, with known holes.
- good0.65 and above
Most indicators observed, recent, from official or intergovernmental sources.
A row marked no dataset exists is a gap: the benchmark wants data that nobody publishes. A retired row has a source this project chose not to use. Both reduce confidence. The glossary defines each term, and the limits page lists where these numbers are known to be wrong about the world.
Anticipation
How capable is the country of identifying and preparing for emerging change?
+5.0 over 10 years, from 37.8 in 2016 to 42.8 on 2 indicators seen in both years.
Trends use fewer indicators than the current score and use today's frame. A longer span reaches further back.
direct capability measureobserved
- Raw value
- 81.288index 0-100
- Year
- 2024. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (IQ.SPI.OVRL)
- History
Government foresight capacity
direct capability measureno dataset exists, 1 record- Raw value
- no dataindex 0-100
- Year
- no data
- Normalized
- no data
- Source
- none
- History
- no data
Note: No comparable international dataset. OECD and UN work is descriptive, not scored. Envisioning would have to build this.
Long-horizon research commitments
direct capability measureno dataset exists- Raw value
- no data% of new grants
- Year
- no data
- Normalized
- no data
- Source
- none
- History
- no data
Note: Redefined 2026-10-02 (D152). The row asked for the share of R&D spending classed as basic research, which is the make-up of a spending stock, the construct D142 retired business R&D share for. OECD MSTI publishes that share for its members only, and UNESCO UIS no longer serves R&D by type of research (construct audit, 2026-10-02). The behaviour behind it is a funder committing to a horizon: a grant awarded for five years or more is a decision to wait for a result. Untried: no memo has triaged a grant register. National funders publish award lists one by one with their own durations, and commercial grant aggregators fail the inspectability rule. Wireable on a grant-level register with award and end dates, on one definition of a competitive public grant, for half the frame or more.
Conditions, not scored
What the country has to work with on this capability. The values are shown as the source published them and are not part of the score, the confidence or the trend. The rank counts the countries with a value, best first.
R&D expenditureWorld Bank (GB.XPD.RSDV.GD.ZS), 2022
0.62 % of GDP, rank 54 of 118. Gross domestic expenditure on research and development as a share of GDP.
A condition, not scored: money committed to research is what a country spends, and spending it does not show the country reads the future well (D122).
Researchers in R&DWorld Bank (SP.POP.SCIE.RD.P6), 2022
444 per million people, rank 69 of 114. Full-time-equivalent researchers per million people.
A condition, not scored: a standing stock of people who could investigate is what a country has, not a record of what they found (D122). Correlates with income but survives size normalization.
Secure internet serversWorld Bank (IT.NET.SECR.P6), 2024
16,658 per million people, rank 35 of 125. Servers using encryption in internet transactions, per million people.
A condition, not scored: a diffusion stock of digital adoption that money buys, not an act of anticipation (D122). Spans four orders of magnitude, so its rank is read on the published value and its correlations on the logged one.
Evidence for missing indicators
Each record describes a case the indicators above cannot see. It does not affect the score or confidence because one case is not comparable across countries. A gap can become an indicator when a comparable series covers at least two countries.
Labour Market Intelligence Programme and forward skills planning
South Africa's Department of Higher Education and Training ran a Labour Market Intelligence Programme from 2012, published the fifth iteration of its national occupations-in-high-demand list by April 2024, and identified 138 occupations for the emerging hydrogen-economy value chain.
5 iterations, national occupations-in-high-demand list iterations, 2024-04-16. 138 occupations, occupations identified for the hydrogen-economy value chain, 2024-04-16. Department of Higher Education and Training, Government of South Africa, retrieved 2026-09-02.
How it worked: DHET institutionalised a research cycle that combines labour-market data, sector studies and consultations with business, labour and government, then turns the findings into recurring occupations-in-demand lists and skills assessments for education planning.
Limits: The programme's own ministerial account reports its outputs, and a list of occupations is a planning instrument rather than proof that enrolment, funding or curricula changed. The source describes the second phase and a third phase planned for 2025; the record therefore establishes a recurring foresight function and delivered reports, not the quality or current use of every forecast.
Panel estimate, round 2, how this layer works
- 45.0 institutionalist@claude-opus-5-5, self-confidence 0.45
Stats SA, the Reserve Bank, Treasury and the National Planning Commission are capable institutions, and the genomic surveillance network that flagged Beta and Omicron, which panelist B raised, is a durable anticipatory institution. Against that, the electricity shortfall was foreseen for over a decade before load-shedding, so I hold at 45.
- 45.0 bottom_up@claude-opus-5-5, self-confidence 0.45
Hold. The panel agrees that strong statistics and planning institutions sit beside a decade of foreseen electricity shortfall that did not become preparation; genomic surveillance that flagged Beta and Omicron is genuine distributed foresight that offsets the low article count.
- 43.0 wealth_sceptic@claude-sonnet-5-5, self-confidence 0.40
Strong statistics, a capable treasury and distributed genomic surveillance are real capability. A decade of foreseen power shortfalls ending in load-shedding shows warning did not become preparation, so I land just above 42.
Agency
How able are individuals and organizations to turn an intention into action?
No trend is shown because too few indicators have values at both ends. The rows below still show the available history.
direct capability measureobserved
- Raw value
- 11.458per 1,000 aged 15-64
- Year
- 2024. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (IC.BUS.NDNS.ZS)
- History
direct capability measureobserved
- Raw value
- 40days
- Year
- 2019. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (IC.REG.DURS.MA.DY)
- History
direct capability measureobserved
- Raw value
- 7count
- Year
- 2019. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (IC.REG.PROC.MA.NO)
- History
Adults doing digital tasks
direct capability measureno dataset exists- Raw value
- no data% of adults
- Year
- no data
- Normalized
- no data
- Source
- ITU / UNESCO UIS (SDG 4.4.1, ICT skills by type)
- History
- no data
Note: Redefined 2026-10-02 (D152). The row asked whether adults can perform standard digital tasks, which reads a skill level, the kind of stock D122 moved out of the scores with internet users. It now counts the act: a person who did the task. It is not the internet users condition beside it, which counts access. Candidate: ITU ICT skills by type (SDG 4.4.1) through the UNESCO UIS API, probed 2026-10-02. Sent a message with an attached file reaches 33 of the 53 countries then in the frame, 25 at 2018 or later, with no United States, India or China and most of Africa and Central America missing; two other task items reach 31 and 16. The items are self-reported and the task list has changed between rounds. Wireable on one task observed for half the frame or more at 2018 or later.
Perceived control over life
perception proxymissing- Raw value
- no datamean 1-10
- Year
- no data
- Normalized
- no data
- Source
- Joint EVS/WVS (A173)
- History
- no data
Note: A perception: the publisher-weighted mean, on a 1 to 10 scale, of how much freedom of choice and control respondents feel they have over the way their life turns out, from the official Joint EVS/WVS 2017-2022 results table, release 5.0.0. It observes the felt capacity to act, not action. The table prints the mean to one decimal and no top-box share, so the row stores the mean and ties are common. Fieldwork years differ by up to six years: the 37 countries emitted were surveyed between 2017 (eight, the US among them) and 2023 (India), and every value is stamped with the release year 2022. Response style on 10-point scales differs across cultures, and a closed autocracy can read high: Vietnam and Nicaragua sit near the top of the frame. Countries with separate EVS and WVS samples (Germany, the United Kingdom, the Netherlands) are held, as for interpersonal trust. See D127 and A15.
Conditions, not scored
What the country has to work with on this capability. The values are shown as the source published them and are not part of the score, the confidence or the trend. The rank counts the countries with a value, best first.
Individuals using the internetWorld Bank (IT.NET.USER.ZS), 2024
78.4 % of population, rank 75 of 125. Share of the population that used the internet in the last three months.
A condition, not scored: access to the basic tool of modern agency is what people have to act with, not a record of them acting (D122). Saturates near 95%, which compresses the top of the range.
Financial account ownershipWorld Bank (FX.OWN.TOTL.ZS), 2024
81.1 % aged 15+, rank 57 of 122. Adults with an account at a bank or mobile money provider.
A condition, not scored: holding an account is the precondition for almost any economic action, not the action (D122). Global Findex, published every three years.
Credit to the private sectorWorld Bank (FS.AST.PRVT.GD.ZS), 2024
89.4 % of GDP, rank 23 of 124. Domestic credit provided to the private sector by financial corporations.
A condition, not scored: financial depth is a stock of money available to actors, not what they do with it (D122). High values can also signal a credit bubble, so more is not always better.
Coordination
How effectively can independent actors organize around shared objectives?
No trend is shown because too few indicators have values at both ends. The rows below still show the available history.
Government effectiveness
perception proxyretired, see notes- Raw value
- no dataz-score -2.5 to 2.5
- Year
- no data
- Normalized
- no data
- Source
- World Bank (GOV_WGI_GE.EST)
- History
- no data
Note: Retired 2026-08-26. The WGI aggregate the opinions of experts and firms, and in this set the four WGI series correlate with each other between 0.93 and 0.98 while correlating with log GDP per capita at 0.91. That is one perception of national wealth counted in three dimensions. Cross-agency delivery records are the replacement and they are a declared gap. See D23 and A4.
Regulatory quality
perception proxyretired, see notes- Raw value
- no dataz-score -2.5 to 2.5
- Year
- no data
- Normalized
- no data
- Source
- World Bank (GOV_WGI_RQ.EST)
- History
- no data
Note: Retired 2026-08-26. Same measurement as government effectiveness, filed under a different name, correlating with it above 0.93. See D23 and A4.
Logistics performance
perception proxyretired, see notes- Raw value
- no dataindex 1-5
- Year
- no data
- Normalized
- no data
- Source
- World Bank (LP.LPI.OVRL.XQ)
- History
- no data
Note: Retired 2026-08-26. The LPI is a survey of international freight forwarders, so it records how a country looks to global logistics firms. A small country with a small port scores low whatever its state can organise, which is artefact A9. Container throughput and border time are observable and stay. See D23.
direct capability measureobserved
- Raw value
- 92hours
- Year
- 2019. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (TRD.ACRS.BRDR.EXPT.BRDR.COMP.HR.DB1619)
- History
direct capability measureobserved
- Raw value
- 102.9percentage points from approved budget
- Year
- 2022. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (GF.XPD.BUDG.ZS)
- History
University-industry collaboration
direct capability measureno dataset exists, 1 record- Raw value
- no dataindex 0-100
- Year
- no data
- Normalized
- no data
- Source
- Global Innovation Index
- History
- no data
Note: GII carries it, but the underlying item is a WEF executive opinion survey whose microdata is not inspectable. Excluded on principle 4.
direct capability measureobserved
- Raw value
- 0.776index 0-1
- Year
- 2025. The rail runs from 1960 to 2026.
- Normalized
- Source
- V-Dem (v2x_cspart)
- History
- one year
Public-private collaboration
direct capability measureno dataset exists, 1 record- Raw value
- no dataindex 0-100
- Year
- no data
- Normalized
- no data
- Source
- none
- History
- no data
Note: No comparable dataset. PPP investment databases measure infrastructure finance, not collaboration capacity.
Behavioral check, not scored
Each number below is published beside the score and stays out of it. It is not in the frame, the average, the indicator count or the confidence.
Customs clearance time for exportsWorld Bank (IC.CUS.DURS.EX), 2020
8.219 days, lower is better. Average number of days to clear direct exports through customs, as reported by manufacturing firms that export directly, in the World Bank Enterprise Surveys.
Days a manufacturer that exports directly says its goods took to clear customs, from the World Bank Enterprise Surveys. It is the time the border took on firms' own shipments, current where the Doing Business border time this capability scores is frozen at 2019. It is not scored for two reasons. The number of firms behind each country's figure is published only in the survey microdata, and on a proxy from the published tables 37 of the 53 countries in the frame when it was checked clear 30 exporting firms; Mexico, the Philippines, Israel and nine others may not. And the question changed: surveys fielded from 2024 also ask how long release took across all border agencies, and drop an answer where customs alone took longer, so a country surveyed on both sides of the change can move severalfold with nothing at its border changing. India reads 17.3 days in 2022 and 2.5 in 2025, on more than 9,000 firms each time. No country in the frame has two surveys on the new questionnaire yet. Income is not why it is left out: it barely tracks log GDP per capita.
Evidence for missing indicators
Each record describes a case the indicators above cannot see. It does not affect the score or confidence because one case is not comparable across countries. A gap can become an indicator when a comparable series covers at least two countries.
Renewable Energy Independent Power Producer Procurement Programme
South Africa's IPP Office reported 95 independent power producers operational with 7,335 MW online by 31 March 2025, alongside 9,618 MW procured from 116 projects at financial close and R292.2 billion attracted for energy infrastructure.
7,335 MW, operational ipp capacity, 2025-03-31. 292,200,000,000 South African rand, investment attracted by projects at financial close, 2025-03-31. Independent Power Producers Office, South Africa, IPPPP Quarterly Report Q4 2024/25, retrieved 2026-09-02.
How it worked: The state set capacity and development rules, ran repeated competitive tenders and gave private producers bankable power-purchase agreements, while Eskom and government agencies carried the grid and policy interfaces.
Limits: The IPP Office reports the programme it manages, and its figures combine several procurement windows and technologies. Operational MW and financial close show a functioning public–private delivery channel, not affordability, reliability, local ownership or emissions outcomes; terminated projects are excluded from current totals and the investment figure is attracted capital rather than completed public value.
TIA Technology Stations Programme, and its erosion
South Africa's Technology Innovation Agency runs a network of university-based Technology Stations that gave 2,907 small firms science, engineering and technology services in 2024/25, against 3,167 at the five-year high in 2021/22, while the programme's budget fell from R105.9 million in 2020 to R93.98 million in 2025 and only 80 percent of the stations were operational.
2,907 SMMEs, smmes accessing set services through the technology stations network, 2024/25. 3,167 SMMEs, smmes accessing set services through the technology stations network, five-year peak, 2021/22. Technology Innovation Agency (TIA), Annual Report 2024/25, retrieved 2026-10-02.
How it worked: The state funds equipped technology stations inside universities, mostly universities of technology, whose staff and students sell subsidised testing, prototyping and process improvement to small firms that could not afford the equipment.
Limits: The numbers sit inside a 188-page PDF, in the outcome table on PDF page 41 (printed page 39), and the budget figures in the Outcome 3 narrative on PDF page 42. The peak is the highest of the five years TIA reports, not a long-run maximum, and the fall from it is about 8 percent, which is computed here. The yearly count is noisy (1,990, 3,167, 2,671, 3,114, 2,907), and 2024/25 sits above two of the five years, so the erosion rests on what TIA reports about capacity: the budget cut, a fifth of the stations not functioning and no new high-end equipment, rather than on the count alone; the budget cut is in nominal rand and larger in real terms. A count of firms served measures university-to-firm technical service, testing and prototyping, which is a thinner form of research collaboration than joint projects: the same report says 73 percent of station equipment time goes to firms. The agency also reports a separate SMME figure (3,775) across all its programmes; this record uses only the Technology Stations line. A 2022 National Treasury spending review recommended moving the programme out of TIA altogether. The start year is taken from the programme's establishment under the former Department of Science and Technology and is not stated in the cited report.
Trust
How much cooperation is possible beyond immediate personal networks?
No trend is shown because too few indicators have values at both ends. The rows below still show the available history.
Rule of law
perception proxyretired, see notes- Raw value
- no dataz-score -2.5 to 2.5
- Year
- no data
- Normalized
- no data
- Source
- World Bank (GOV_WGI_RL.EST)
- History
- no data
Note: Retired 2026-08-26. A WGI perception composite correlating with control of corruption above 0.95 and with log GDP per capita at 0.83. Court throughput and case clearance rates are the observable replacement and are a declared gap. See D23 and A4.
Control of corruption
perception proxyretired, see notes- Raw value
- no dataz-score -2.5 to 2.5
- Year
- no data
- Normalized
- no data
- Source
- World Bank (GOV_WGI_CC.EST)
- History
- no data
Note: Retired 2026-08-26. A WGI perception composite. It measures reputation for corruption, which is not the same thing as corruption, and it tracks income closely. Prosecution and audit records are the observable replacement and are a declared gap. See D23 and A4.
direct capability measureobserved
- Raw value
- 600days
- Year
- 2019. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (ENF.CONT.DURS.DY)
- History
downstream outcomeobserved
- Raw value
- 1.538% of firms
- Year
- 2020. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (IC.FRM.BRIB.ZS)
- History
downstream outcomeobserved
- Raw value
- 3.116scale 0-4 (never to always)
- Year
- 2025. The rail runs from 1960 to 2026.
- Normalized
- Source
- V-Dem (v2jucomp_osp)
- History
- one year
Intentional homicide rate
downstream outcomeretired, see notes- Raw value
- no dataper 100,000 people
- Year
- no data
- Normalized
- no data
- Source
- World Bank (VC.IHR.PSRC.P5)
- History
- no data
Note: Added 2026-08-26 by D23 as the observable replacement for two retired WGI perception composites, and retired 2026-08-27 by D44 when the diagnostic built in D42 measured what it was doing. It raised this dimension's correlation with log GDP per capita by 0.288, from 0.096 to 0.385, the largest single wealth contribution anywhere in the model. It was added to fix wealth contamination and it was the wealth contamination. The dataset is sound and the objection is to what it measures here: homicide is driven heavily by organised crime, a society can be physically safe while trusting very little, and across this country set the variation it carries is mostly income. Behavioural trust measures remain a gap.
Generalised interpersonal trust
perception proxymissing- Raw value
- no data% agreeing
- Year
- no data
- Normalized
- no data
- Source
- Joint EVS/WVS (A165)
- History
- no data
Note: Share answering that most people can be trusted in the official Joint EVS/WVS 2017-2022 results table, release 5.0.0. The adapter uses the publisher-weighted country results, emits only countries with one survey row, and holds countries with separate EVS and WVS samples until the respondent-level weights can be pooled reproducibly. See D57 and D64.
Trust in public institutions
perception proxyno dataset exists- Raw value
- no data% expressing confidence
- Year
- no data
- Normalized
- no data
- Source
- OECD Trust Survey / WVS
- History
- no data
Note: The OECD survey covers members only. Mixing it with WVS items for India and South Africa would break comparability. The Joint EVS/WVS confidence battery (E069) reaches 37 countries from one release, and confidence in the courts (E069_17) is published beside Trust as a check under this id rather than scored: every item of the battery tested (courts, civil service, police, parliament and government) reads higher in the closed and electoral autocracies of the frame than in its democracies, which is deference as much as confidence (A13). The gap stays open for a measure that does not reward a court that cannot rule against the state. See D132.
Cooperation beyond the in-group
perception proxymissing- Raw value
- no data% expressing trust
- Year
- no data
- Normalized
- no data
- Source
- Joint EVS/WVS (G007_34_B)
- History
- no data
Note: The distinction that makes this dimension worth measuring: whether trust reaches a person with no tie to the respondent. Generalised trust asks about "most people", which a respondent can fill with their own circle; this item names the stranger. China is the clearest case, high on most people and low on a person met for the first time. The value is the publisher-weighted share answering trust completely plus the share answering trust somewhat, the one sum of two printed columns the adapter stores, exact to about 0.1, over all respondents including don't know and no answer. EVS and WVS ask the same question on the same four-point scale; where one country has both samples, the WVS share is a median 2.2 points higher. Trust in people of another nationality (G007_36_B), which the row once also named, is left out because it reads attitudes to foreigners as much as trust. Fieldwork runs from 2017 to 2023 and every value carries the release year 2022. Germany, the United Kingdom and the Netherlands are held, as for interpersonal trust. See D140.
Court case clearance rate
downstream outcomeno dataset exists- Raw value
- no data% of incoming cases
- Year
- no data
- Normalized
- no data
- Source
- CEPEJ / OECD / national court statistics
- History
- no data
Note: Whether a court finishes what it starts, counted from case records rather than asked in a survey. D23 named it as the observable replacement for the two retired WGI composites and it has stayed unfilled since. CEPEJ publishes it for Council of Europe members and OECD for its own members, so no single publisher reaches this country set, and a harmonised series needs the project second source adapter. See D57.
Panel estimate, round 2, how this layer works
- 51.0 institutionalist@claude-opus-5-5, self-confidence 0.40
Independent courts that the executive obeys, as the Zuma contempt ruling showed, and low firm bribery are strong institutional signals. State capture showed corruption that firm surveys missed, and generalised trust is among the lowest measured, so I stay well below the indicator; panelist C's point about courts ruling against the executive earns one point.
- 50.0 bottom_up@claude-opus-5-5, self-confidence 0.40
Hold. Panelist C's 55 gives more weight to low firm bribery than state capture findings allow, since the firm survey missed corruption of that scale; independent, obeyed courts and stokvels that pool money among non-kin keep it at the middle despite low survey trust.
- 52.0 wealth_sceptic@claude-sonnet-5-5, self-confidence 0.40
Independent courts that rule against the executive and very low reported firm bribery are real, and I do not mark far down. State capture that firm surveys missed, high violent crime and low generalised trust argue against 83; I move down from 55 on the panel's point about survey blindness.
Learning
How effectively does the country acquire, distribute, and update knowledge?
No trend is shown because too few indicators have values at both ends. The rows below still show the available history.
direct capability measureobserved
- Raw value
- 0.425index 0-1
- Year
- 2020. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (HD.HCI.OVRL)
- History
direct capability measureobserved
- Raw value
- 7.896% of firms
- Year
- 2020. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (IC.FRM.TRNG.ZS)
- History
Adult learning participation
direct capability measureno dataset exists, 2 records- Raw value
- no data% of adults
- Year
- no data
- Normalized
- no data
- Source
- Eurostat / OECD PIAAC
- History
- no data
Note: The best available measure of continuous learning capacity, and it exists only for European and OECD members.
downstream outcomeobserved
- Raw value
- 1.504ratio to world average
- Year
- 2021. The rail runs from 1960 to 2026.
- Normalized
- Source
- OpenAlex (citation_normalized_percentile.is_in_top_10_percent)
- History
- one year
Conditions, not scored
What the country has to work with on this capability. The values are shown as the source published them and are not part of the score, the confidence or the trend. The rank counts the countries with a value, best first.
Tertiary enrolmentWorld Bank (SE.TER.ENRR), 2023
23.5 % gross, rank 90 of 124. Gross enrolment ratio in tertiary education.
A condition, not scored: enrolment counts people in formal education and records nothing learned (D122).
Public education expenditureWorld Bank (SE.XPD.TOTL.GD.ZS), 2024
6 % of GDP, rank 14 of 125. Government expenditure on education as a share of GDP.
A condition, not scored: public spending on education is a pure input, and learning is what the dimension asks about (D122).
Vocational share of secondaryWorld Bank (SE.SEC.ENRL.VO.ZS), 2018
7.2 % of secondary, rank 84 of 125. Vocational pupils as a share of total secondary enrolment.
A condition, not scored: the share of secondary pupils on a vocational track describes how a school system is built, not what anyone learns in it, and higher is not unambiguously better (D141). It marks a route into skilled work that does not run through university. Learning reads research citation impact, firm training and the human capital index instead.
Evidence for missing indicators
Each record describes a case the indicators above cannot see. It does not affect the score or confidence because one case is not comparable across countries. A gap can become an indicator when a comparable series covers at least two countries.
SETAs and workplace-based learning delivery
South Africa's Sector Education and Training Authorities reported 45,146 learnership completions, 83,578 skills-programme completions and 99,778 workplace-based learning opportunities for the 2024 post-school education and training intake.
45,146 completions, learnerships completed, 2024-01-22. 99,778 opportunities, workplace-based learning opportunities created, 2024-01-22. Department of Higher Education and Training, Government of South Africa, retrieved 2026-09-02.
How it worked: Sector authorities levy and coordinate employer-linked training, then place learners into learnerships, skills programmes and workplace experience tied to occupational demand rather than leaving adult learning entirely to formal colleges.
Limits: The ministerial statement aggregates 21 SETAs and counts opportunities and completions, not a harmonised share of adults participating in formal or non-formal learning. It does not report unique learners, demographic reach, completion quality, employment after training or the share of participants who were already in education; the figures are government-reported rather than independently audited here.
Community Education and Training colleges, and their erosion
South Africa's Department of Higher Education and Training merged the public adult learning centres into nine Community Education and Training colleges on 1 April 2015; the colleges enrolled 273,431 people in 2016 and 120,081 in 2023, against a National Development Plan target of one million by 2030.
120,081 students, students enrolled in public cet colleges, 2023. 273,431 students, students enrolled in public cet colleges in the first year after the merger reported by dhet, 2016. Department of Higher Education and Training, Statistics on Post-School Education and Training in South Africa 2023, retrieved 2026-10-01.
How it worked: DHET took provincial adult learning centres into nine national CET colleges to professionalise public adult education. Enrolments kept dwindling, and some centres had lecturers but no students.
Limits: The two figures come from two DHET publications, and the peak is not established: 2016 is the earliest CET-college figure found, and the same Monitor puts the predecessor adult basic education system at 297,491 in 2010, so the decline began before the merger. Enrolment is a headcount, not the share of adults in learning that the indicator asks for, and the 2023 count rests on an 83.5 percent response rate, so it is understated by an unknown amount. DHET says enrolments have been dwindling since data moved to audited unit-level records, so part of the fall may be the removal of unaudited or phantom enrolments rather than learners lost. Most enrolments are Grade 12 and GETC, which is second-chance schooling rather than adult training; the largest group is aged 20 to 24; and DHET suggests part of the 2023 Grade 12 drop moved to the Department of Basic Education's second-chance programme. The sector's own report notes centres with lecturers and no students. Both numbers are in PDFs.
Experimentation
How easily can new approaches be attempted, tested, abandoned, and improved?
+1.9 over 10 years, from 7.3 in 2016 to 9.2 on 3 indicators seen in both years.
+0.7 over 20 years, from 8.5 in 2006 to 9.2 on 3 indicators seen in both years.
Trends use fewer indicators than the current score and use today's frame. A longer span reaches further back.
downstream outcomeobserved
- Raw value
- 1804per million people
- Year
- 2021. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (IP.PAT.RESD)
- History
downstream outcomeobserved
- Raw value
- 25786per million people
- Year
- 2021. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (IP.TMK.RSCT)
- History
downstream outcomeobserved
- Raw value
- 728per million people
- Year
- 2021. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (IP.IDS.RSCT)
- History
downstream outcomeobserved
- Raw value
- 417406per million people
- Year
- 2026. The rail runs from 1960 to 2026.
- Normalized
- Source
- GitHub Innovation Graph (data/repositories.csv)
- History
- no data
Venture deals
direct capability measureno dataset exists- Raw value
- no dataper million people
- Year
- no data
- Normalized
- no data
- Source
- OECD / national VC associations
- History
- no data
Note: Redefined 2026-10-02 (D152). The id keeps its first name so evidence records, links and published files still resolve. The row asked for venture capital deployed as a share of GDP, a level of financial depth of the kind private credit is (D122). A count of first rounds counts attempts started with outside risk money, which is what this capability asks about, and a large round counts once. Candidates, none wireable: the OECD SME and Entrepreneurship Financing scoreboard carried venture capital for 6 of the 16 countries the frame held on 2026-08-26, as amounts in national currency, latest year 2022, with Brazil, India, South Africa and Singapore absent (D21); national venture capital associations publish counts on their own definitions; commercial deal databases cover the world and are not inspectable (D10). Wireable on an inspectable deal count with one definition of a first round for half the frame or more. Read A1 before treating this dimension as measured.
direct capability measureobserved
- Raw value
- 14.7% aged 18-64
- Year
- 2025. The rail runs from 1960 to 2026.
- Normalized
- Source
- Global Entrepreneurship Monitor (APS key indicators, TEA rate)
- History
- one year
perception proxyobserved
- Raw value
- 50.4% not deterred
- Year
- 2025. The rail runs from 1960 to 2026.
- Normalized
- Source
- Global Entrepreneurship Monitor (APS key indicators, fear of failure rate)
- History
- one year
Firms through regulatory sandboxes
direct capability measureno dataset exists, 1 record- Raw value
- no datafirms per million people
- Year
- no data
- Normalized
- no data
- Source
- national regulators
- History
- no data
Note: Redefined 2026-10-02 (D152). The row asked for the number and breadth of live sandboxes, which is a policy stock: a regime can exist with nobody in it. It now counts what passes through: firms a regulator let test under relaxed rules, and firms that came out, authorised, withdrawn or failed. Regulators publish cohort lists and exit reports one by one. The evidence corpus holds 22 records against this row, most of them counts of firms admitted or exited, each on the definition and window of its own regulator, and no register assembles them on one definition (O1 triage sweep, 2026-10-01). A realistic candidate for Envisioning to build from those lists. Wireable on a register that counts admissions and completions on one definition for half the frame or more.
University spinouts
direct capability measureno dataset exists- Raw value
- no dataper million people
- Year
- no data
- Normalized
- no data
- Source
- national TTO associations
- History
- no data
Note: Reported nationally with incompatible definitions of what counts as a spinout.
Business share of R&D
capability inputretired, see notes- Raw value
- no data% of R&D
- Year
- no data
- Normalized
- no data
- Source
- OECD MSTI / RICYT
- History
- no data
Note: Retired 2026-10-02. The only working source, OECD MSTI joined with RICYT for Latin America, covers 34 of the 53 countries then in the frame, but the row reads the make-up of a spending stock: who performs the R&D a country pays for. R&D spending itself is a condition since D122, and the share is high both where business research is large (Israel) and where total research is tiny and one firm dominates it (Thailand). State enterprises count as business, so China and Vietnam read their ownership model. UIS stopped publishing the series in 2023. Firms trying new things are read through patents, trademarks, designs and entrepreneurship instead. See D142.
Evidence for missing indicators
Each record describes a case the indicators above cannot see. It does not affect the score or confidence because one case is not comparable across countries. A gap can become an indicator when a comparable series covers at least two countries.
South Africa's Intergovernmental Fintech Working Group, a joint body of the financial regulators, opened a cross-regulator sandbox in April 2020 that took 54 unique applications and admitted nine sandbox tests from eight firms in its first cohort, then moved to a rolling intake.
9 sandbox tests, sandbox tests admitted to the first ifwg cohort, 2020. 54 applications, unique applications to the first cohort, 2020-05-15. South African Reserve Bank for the Intergovernmental Fintech Working Group, First Regulatory Sandbox Report, October 2022, retrieved 2026-10-01.
How it worked: The Reserve Bank, the FSCA and other financial regulators ran one sandbox through a joint working group, so an innovator whose product crossed mandates such as crypto cross-border payments was tested by every relevant regulator at once.
Limits: Nine tests is a small count, and the source is the regulators reporting on their own sandbox. The report does not show that any rule changed because of a test: it says two tests confirmed compliance with existing rules and seven produced insights for possible future frameworks. The 'operating' status rests on the October 2022 statement that the IFWG voted to move to a rolling intake. No official page that opened on the retrieval date counted participants after the first cohort, and the IFWG's own site (ifwg.co.za) did not respond to scripted requests. The sandbox covers financial services only, so it shows nothing about controlled trial regimes in other sectors. The figures are in a PDF.
Adaptability
How effectively can the system respond when circumstances change?
No trend is shown because too few indicators have values at both ends. The rows below still show the available history.
downstream outcomeobserved
- Raw value
- 32.391% of labour force
- Year
- 2025. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (SL.UEM.TOTL.ZS)
- History
downstream outcomeobserved
- Raw value
- 76.5% of unemployed
- Year
- 2024. The rail runs from 1960 to 2026.
- Normalized
- Source
- ILOSTAT (DF_UNE_TUNE_SEX_AGE_DUR_NB)
- History
- one year
direct capability measureobserved
- Raw value
- 0.153index 0-1, lower = more diversified
- Year
- 2025. The rail runs from 1960 to 2026.
- Normalized
- Source
- UNCTAD (US.ConcentDiversIndices)
- History
- one year
downstream outcomeobserved
- Raw value
- 3.75% of products not exported competitively at the start
- Year
- 2024. The rail runs from 1960 to 2026.
- Normalized
- Source
- Harvard Growth Lab (hs92_country_product_year_4.csv (doi:10.7910/DVN/T4CHWJ))
- History
- one year
Disaster preparedness and recovery
direct capability measureno dataset exists, 1 record- Raw value
- no dataindex 0-100
- Year
- no data
- Normalized
- no data
- Source
- INFORM / UNDRR
- History
- no data
Note: No source observes a country preparing for or recovering from a real shock. INFORM publishes its lack of coping capacity apart from hazard and exposure, but that dimension averages government self-assessments and governance perception indices with stocks such as roads, phones and physicians, which is the perception layer D23 retired plus conditions money buys (D122). Sendai and WHO reporting are governments grading themselves. A warning-reach survey item is a check candidate. See the disaster preparedness memo.
Institutional responsiveness
direct capability measureno dataset exists, 1 record- Raw value
- no dataindex 0-100
- Year
- no data
- Normalized
- no data
- Source
- none
- History
- no data
Note: No dataset exists. Measurable in principle from legislative and regulatory timestamps, which no one has assembled comparably.
Conditions, not scored
What the country has to work with on this capability. The values are shown as the source published them and are not part of the score, the confidence or the trend. The rank counts the countries with a value, best first.
Labour force participationWorld Bank (SL.TLF.ACTI.ZS), 2025
60.6 % aged 15+, rank 103 of 125. Share of the working-age population in the labour force.
A condition, not scored: how much of the population is in the labor force at all is what a country has to reallocate, a level set by norms, schooling and age, not a record of it reallocating (D141). Participation norms confound it, especially for India. Adaptability reads reallocation through long-term unemployment and export diversification instead.
Fixed broadband subscriptionsWorld Bank (IT.NET.BBND.P2), 2024
5.3 per 100 people, rank 89 of 125. Fixed broadband subscriptions per 100 people.
A condition, not scored: installed infrastructure is what a country has to change with, not a record of it changing (D122). Penalizes mobile-first countries such as India and South Africa.
Electricity transmission lossesWorld Bank (EG.ELC.LOSS.ZS), 2023
10.2 % of output, rank 71 of 117. Power lost in transmission and distribution, including unbilled use.
A condition, not scored: the state of the grid, and whether its operator can bill what it delivers, is infrastructure a country has to change with, not a record of it changing (D141). Read it beside broadband, the other installed network beside Adaptability.
Informal employmentILOSTAT (DF_SDG_0831_SEX_ECO_RT), 2024
34.8 % of employment, rank 39 of 102. Informal employment as a share of total employment, SDG indicator 8.3.1.
A condition, not scored: how much work carries no contract, registration or social protection describes the labor market a country reallocates through, not it reallocating (D150). Read it beside the unemployment rate: where most work is informal, a person who loses a job takes any work within weeks, so a low rate is not a fluid market. Its direction for Adaptability cannot be defended, since informal work is both the slack that absorbs a shock and the precarity that keeps a worker from moving up, so the rank orders values and is not a verdict. Comparability is weak: the European values come from an income survey (EU-SILC) with a proxy definition, and the rest from labor force or household surveys processed by the ILO. ILOSTAT has no value for the United States, Canada, Australia, Japan, Singapore, Israel, Malaysia, China, the Philippines or Cuba, and some values are old (Haiti and Nicaragua 2012).
Evidence for missing indicators
Each record describes a case the indicators above cannot see. It does not affect the score or confidence because one case is not comparable across countries. A gap can become an indicator when a comparable series covers at least two countries.
Gauteng e-tolls, and their cancellation after public resistance
South Africa operated Gauteng's e-toll system from December 2013 until the government withdrew the toll declaration effective 11 April 2024, ending urban e-toll charging after roughly 10.4 years.
10.36 years, years of e-toll collection before cancellation, 2024-04-11. Department of Transport, Government of South Africa, retrieved 2026-09-02.
How it worked: After motorists resisted the user-pay system and the funding model stalled, a joint transport, finance and provincial process replaced toll revenue with an alternative debt and maintenance arrangement and repurposed the gantries.
Limits: The duration is calculated from the ministry's stated December 2013 start and 11 April 2024 end rather than published as a single statistic. Cancellation shows that government eventually changed a rule after resistance and a debt impasse, but it does not show how quickly the response was reached, the road programme's net value or the fiscal cost of replacing toll revenue.
South African Weather Service radar network, and its erosion
The South African Weather Service reported that the availability of its weather radar infrastructure fell from 73.83 percent in 2021/22 to 52.23 percent in 2022/23 and 44.25 percent in 2023/24, against a 75 percent target, as load shedding, missing spares and ageing C-band radars took sites down.
44.25 %, availability of radar infrastructure, average of quarterly performance, 2023/24. 73.83 %, availability of radar infrastructure, 2021/22. South African Weather Service, Annual Report 2023/24 (performance information, Programme 3), as tabled in Parliament, retrieved 2026-10-02.
How it worked: A national weather service built and ran a countrywide radar network that underpins severe-weather warnings, then let it degrade when power supply, spares contracts and maintenance budgets failed together.
Limits: The figure is in the PDF of SAWS's own annual report (printed page 70, PDF page 72), and the copy that opens is hosted by the Parliamentary Monitoring Group, a civil-society mirror of documents tabled in Parliament; the SAWS website did not serve the report to a script. 2021/22 is not shown to be the historical peak: the 2024/25 plan presented to Parliament gives 73 percent for 2020/21 as well, so the record shows a fall from a plateau, not from an all-time high. Availability is the share of time radars produced data, not forecast or warning quality, and the population covered by a working radar is not reported. The main cause named is national load shedding, an Eskom failure outside SAWS's control (see zaf-eskom-loadshedding), compounded by spares shortages and vandalism. SAWS says the ageing C-band radars will be dropped from the statistic from 2024/25, which will raise the reported figure without repairing them. The start year is SAWS's establishment under the South African Weather Service Act of 2001, which is not stated on the cited page.
Building
How capable is the country of turning plans and knowledge into functioning systems?
-1.9 over 10 years, from 20.9 in 2016 to 19.0 on 3 indicators seen in both years.
Trends use fewer indicators than the current score and use today's frame. A longer span reaches further back.
downstream outcomeobserved
- Raw value
- 12.255% of GDP
- Year
- 2025. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (NV.IND.MANF.ZS)
- History
downstream outcomeobserved
- Raw value
- 5.709% of manufactured exports
- Year
- 2024. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (TX.VAL.TECH.MF.ZS)
- History
Logistics infrastructure quality
perception proxyretired, see notes- Raw value
- no dataindex 1-5
- Year
- no data
- Normalized
- no data
- Source
- World Bank (LP.LPI.INFR.XQ)
- History
- no data
Note: Retired 2026-08-26. An LPI sub-index, so the same freight-forwarder survey as logistics performance and the same problem. Electricity connection speed stays as the observable infrastructure measure in this dimension. See D23.
direct capability measureobserved
- Raw value
- 60.435score 0-100
- Year
- 2019. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (IC.ELC.TIME.DFRN)
- History
downstream outcomeobserved
- Raw value
- 0.511index
- Year
- 2024. The rail runs from 1960 to 2026.
- Normalized
- Source
- World Bank (EF.EFM.UNIV.XD)
- History
Large project delivery
direct capability measureno dataset exists, 2 records- Raw value
- no data% overrun
- Year
- no data
- Normalized
- no data
- Source
- none
- History
- no data
Note: The single best measure of execution capability and there is no comparable international dataset. Assembling one is a defensible Envisioning project. Building mixes industrial output with state delivery capacity, and this indicator carries the delivery side: the other six indicators cannot see a national programme that was delivered. Documented deliveries are recorded in data/evidence and never scored, see D20.
Firm scale-up rate
direct capability measureno dataset exists- Raw value
- no data% of firms
- Year
- no data
- Normalized
- no data
- Source
- OECD DynEmp
- History
- no data
Note: OECD business dynamics work covers members irregularly. Nothing comparable for India, Brazil or South Africa.
Conditions, not scored
What the country has to work with on this capability. The values are shown as the source published them and are not part of the score, the confidence or the trend. The rank counts the countries with a value, best first.
Output per workerWorld Bank (SL.GDP.PCAP.EM.KD), 2025
48,694 constant 2021 PPP $, rank 60 of 123. GDP per person employed, constant purchasing power parity.
A condition, not scored: output per worker is income per worker, the wealth the benchmark tests capability against, so it is published beside Building as context (D122).
Evidence for missing indicators
Each record describes a case the indicators above cannot see. It does not affect the score or confidence because one case is not comparable across countries. A gap can become an indicator when a comparable series covers at least two countries.
South Africa ran one of the world's cheapest reliable grids and extended it to most of the population after 1994, then under-invested and lost the surplus, reaching 335 days of load shedding in 2023 before recovering to 83 days in 2024.
335 days, days with load shedding in the worst year, 2023. 83 days, days with load shedding after the partial recovery, 2024. CSIR, utility-scale power generation statistics in South Africa 2024, retrieved 2026-08-27.
How it worked: The erosion is the mechanism recorded here: a state monopoly was told to electrify without being allowed to charge for or build ahead of demand, new-build decisions were delayed for a decade, and the two giant plants finally ordered were too large for the delivery capability that remained.
Limits: The day counts are read from the annual chart in the CSIR report rather than a stated table, and a day with any load shedding counts the same as a day of stage six, so the unit understates severity differences. The 2024 recovery also leans on private rooftop solar that households bought in self-defence, which the report itself notes, so the improvement is not all Eskom's. What the record shows is a delivered capability, generation adequacy, lost over roughly fifteen years of deferred maintenance and failed new build, with Medupi and Kusile late and over budget.
MeerKAT radio telescope, a delivered national research facility
South Africa's National Research Foundation and SARAO completed the 64-dish MeerKAT radio telescope in March 2018 on schedule and on budget, then began science operations as a precursor to the Square Kilometre Array.
64 dishes, meerkat dishes completed, 2018-03. South African Radio Astronomy Observatory, National Research Foundation, retrieved 2026-09-02.
How it worked: A national research agency led a staged telescope build, using prototype arrays, local engineering and university and industry partners to test components before integrating the full 64-dish facility in the Karoo.
Limits: SARAO's institutional account states that the array was on schedule and on budget but does not publish an audited cost or baseline schedule in this record. Sixty-four dishes prove physical completion, not scientific productivity, local industrial spillovers or the performance of the larger SKA programme into which MeerKAT is being integrated.