Skills, top talent and the 2047 target

Skills, top talent and the 2047 target

A companion note to the talent-identification evidence review. Version of 6 October 2026. Every figure below was read in the source document itself, listed at the end.

About this paper. This is an internal analysis by Avanti Fellows, shared for transparency. It has not been peer-reviewed. It was prepared with substantial help from AI tools (Anthropic’s Claude), which wrote much of the code and drafted much of the text under the direction of Avanti staff. The numbers are produced by code from the underlying data, and the sources and methods are documented so they can be checked, but errors are possible. Read the findings as preliminary, and contact Avanti Fellows before citing them.

India’s stated goal is to be a developed nation by 2047. NITI Aayog’s February 2026 scenarios put numbers on it:

• GDP from USD 4.18 trillion in 2025 to USD 30 trillion in 2047, about seven times larger in 22 years.
• Income per head above USD 18,000, against about USD 2,700 today (IMF).
• Real growth averaging 7.0% a year through 2047, resting on an investment rate near 34% of GDP.

India has averaged about 6% real growth over the last ten years, a period that includes the COVID year. The gap is large in dollars:

• On its current path, India’s economy is about USD 25 trillion in 2047, about USD 4.6 trillion short of the goal. That is the path of a country whose skill levels stay where they are, growing about 6.2% a year.
• Lifting the top tier from about 1% to 5% of each cohort is worth about USD 2-3 trillion of that gap by 2047, on the cross-country growth evidence, and more each year after as these cohorts reach their working peak.
• Lifting both numbers to the best record any country has achieved puts the economy at about USD 29-34 trillion in 2047, in line with the goal.

These figures apply NITI’s own conversion from real growth to dollars (its 7.0% real path reaches USD 30 trillion) to the growth rates in the table further down. Closing that gap depends on two numbers in schools: how many 15-year-olds reach a basic level of skill, and how many reach the top.

## India is near the bottom on both numbers

Across countries, the share of 15-year-olds who can do basic mathematics and, even more, the share who perform at a world-class level go with faster growth over the long run. India is low on both.

• The basic level. About 22% of Class X students are proficient in maths in the government’s own National Achievement Survey (NAS 2021). India’s last PISA test, in two states in 2009, found 12-15% of 15-year-olds at the basic level, so the two measures, twelve years apart, tell the same story. Japan, Korea and Taiwan are at 81-85%, Viet Nam at 58%.
• The top. About 1% of Indian students reach an international advanced benchmark, from the one study that tested this (Das and Zajonc, in two states). Korea and Japan are at about 22%, Taiwan 32%. India has no national measure of its top tier at all, so this study is the anchor for the level; tests normed on private-school takers, such as EI’s ASSET, show trends, not the level.

Government policy has set targets only for the floor: universal foundational literacy and numeracy by Grade 3 (NEP 2020, NIPUN Bharat). On the top, India has moved the other way since 2021.

• NTSE, the national talent search that gave about 2,000 scholarships a year since 1963, has been stalled since its approval lapsed in March 2021 and is still “under review” (Lok Sabha, March 2025).
• KVPY, the science-aptitude fellowship, ended in 2022. The government’s reason was the cost of running an exam for about 300 fellowships a year; selection moved to board marks and JEE and NEET ranks.
• INSPIRE-SHE, the scheme that absorbed KVPY’s students, has not opened a call since 2024 and is officially “under redesign”.
• NMMSS, the means-cum-merit scholarship, has paid ₹12,000 a year since 2017, and the number of scholarships paid roughly halved between 2017-18 and 2021-22.
• The gifted-education guidelines NEP 2020 promised (para 4.43) have not been issued. The Olympiads still run, but for a few hundred students, partly on private money.
• Board exams stopped separating the top. Until 2017 boards routinely raised marks after the exam: CBSE added 16 marks to Mathematics in 2016 with a cap at 95, so a student with 79 and one with 94 both got 95 (Bhattacharji and Kingdon 2017). In 2022 NCERT removed whole chapters from Class X, among them the periodic table, listing “high difficulty level” as a reason. And CBSE now declares no merit list and awards no divisions, “to avoid unhealthy competition”.

Growth needs movement on both ends

The economies that grew fastest for longest have both. Korea, Taiwan, Singapore, Japan and China’s coastal provinces all have more than 80% of 15-year-olds at the basic level and more than 20% at the top. None of them got there with only one.

The research says the same thing across countries. Hanushek and Woessmann followed about 50 countries over 40 years and found that both shares go with faster growth, each in its own right. Every 10 percentage points more of students at the basic level goes with about 0.3 points more growth a year, and every 10 points more at the top with about 1.3 points. When the data was carried forward to 2016, the ratio of roughly five to one held.

The two ends do different jobs. A broad base gives an economy a workforce that can use and spread new technology, and it is also how able children get found in the first place. The top tier produces the engineers, scientists and founders who create new ideas, and it counts for more in countries that are still catching up - which is where India is. India’s own history points the same way: across 16 states, a 1% rise in tertiary education did as much for growth as a 13% fall in illiteracy.

This is a correlation across countries, not an experiment, and in rich countries on their own the top-tier effect fades. But for a country at India’s stage, the evidence and the examples agree that the 2047 goal needs both ends to move.

On its current path, India is unlikely to move either number much

India’s measures have barely changed. The 2009 PISA states and NAS 2021 tell the same story, ASER’s foundational measures have been flat for a decade, and NAS Class X maths reportedly fell between 2017 and 2021.

Other countries that started where India is have mostly stayed there. Of the 80 school systems with a PISA trend, 25 began with at most 40% at the basic level and 2% at the top (Appendix A8). Their typical gain was about 0.2 points a year at the basic level and nothing at the top. Several went backwards, and not one of the 25 built a top tier: the best, Qatar, went from 0.6% to 3.1% over nineteen years.

The last ten years tell the same story, and faster economic growth did not fix it on its own. Ten economies over $100bn stood near India’s level in PISA 2015. Seven of them grew their income per head by less than 3% a year to 2025, and all but Qatar ended with a smaller share at the basic level and a smaller top tier. Of the three that grew faster, Indonesia went backwards on both numbers despite growing 3.3% a year, and only Türkiye moved both up.

If India follows that path, it reaches about a quarter at the basic level and stays at about 1% at the top in 2047. That goes with growth of about 6.2% a year - short of the 7% the goal needs.

Matching the best improvers would put the 2047 goal in reach

A few countries have moved these numbers, and their pace sets what is realistic for India.

If India becomes like…	Basic level in 2047	Top in 2047	Growth a year
a typical country with a similar start	~26%	~1%	~6.2%
the best low-base improvers (Georgia to Qatar)	43-59%	3-4%	6.6-7.5%
the best record on both numbers	45-59%	~5%	6.9-7.6%

The best record on both is about 45-60% at the basic level and about 5% at the top. No country starting low has grown its top tier faster than about 0.2 points a year over the long run (the UAE and Romania), and at that pace India reaches about 5% by 2047. Short bursts of a point a year have happened in Poland, Portugal and Taiwan, but each of them fell back.

So the targets this note proposes - 45-60% at the basic level and 5% at the top by 2047 - are not modest. They ask India to match the fastest any country has gone, on both numbers at once. That is what it takes to meet the 2047 goal.

The countries that grew a top tier did it deliberately

The countries that moved their top tier share a pattern (Appendices A1-A7).

Türkiye built a selective tier and kept adding seats. Its science high schools, public boarding schools in every province entered by a national exam, teach about 5% of 15-year-olds and hold nearly half of the country’s top performers in maths. Places grew from about 29,000 students in 2010-11 to 204,000 in 2023-24, and its top tier rose with them while the share reaching the top inside these schools held steady. An independent test, TIMSS, shows Türkiye’s share of Grade 8 students at the advanced level rising from 6% in 2015 to 17% in 2023. A network of after-school gifted centres, screening children nominated by every class teacher in Grades 1-3, grew alongside to about 100,000 students.

The UAE made the top tier something schools are accountable for. Every school is inspected on how it identifies and supports able children, and Dubai’s schools benchmark themselves against international tests every year.

Romania sorted students by a national exam into selective high schools, and set up county centres of excellence in law to run after-school Olympiad groups. Teachers’ merit pay counts their students’ Olympiad results.

Poland and Portugal raised their top tiers through curriculum and exams built around reasoning and a national maths plan, and Taiwan has kept gifted provision in law since 1984. Poland’s and Portugal’s gains reversed when their systems were reorganised; Taiwan’s has held.

None of these programmes has a causal evaluation, so what we have is timing and association. Where countries built a selective tier, the gap between the top and the bottom often widened: in Türkiye a quarter of students sit in vocational schools where almost nobody reaches the top, and private schools take most of the perfect scores on the selection exam. And some rises are simply changes in who was tested, as in Qatar.

Taken together, the countries that grew a top tier did four things: they sized a selective tier to the cohort, they found children through screening that reaches everyone, they measured and rewarded performance at the top, and they kept the system stable for long enough to work.

India has the seats for its current top tier, but not the practice

India already runs a network of selective government residential schools: Navodaya, Eklavya and the Telangana, Andhra and Karnataka residential societies, about 3,400 schools that take children by entrance test, board them free and teach them to Class 12. Together they hold roughly 2.2 lakh children a cohort, rising to about 3.1 lakh as new Eklavya schools mature. That is about 1% of a cohort - roughly the size of India’s current top tier, and a fifth of Türkiye’s selective share.

These schools select on poverty, region and category as well as merit, so their seats are not the same as a top tier. But they are the nearest thing India has to Türkiye’s science high schools, and they already exist. What is missing is the practice around them. Navodaya’s main entry is one opt-in test in Class 5 for Class 6. Later tests at Classes 9 and 11 fill only the seats left vacant, a few thousand at most at Class 9 against about 46,000 at Class 6. Within Navodaya, NCST at Class 10 selects students for free JEE and NEET coaching in Classes 11 and 12, and NVS first ran it nationally in 2026. Outside those coaching places, children follow the same syllabus at the same pace as everyone else, whenever they joined. And no national survey measures the top tier, so no one can say whether it is growing or where it sits.

Recommendations

To meet the 2047 goal, India would need to do seven things:

1. Set a top-tier target alongside the FLN target: about 5% of each cohort at an internationally top level by 2047, with 45-60% at the basic level.
2. Measure the top tier now, by embedding released TIMSS or PISA items in tests that already run (NCST, ASSET, a JNVST-linked screen), and build the same into the Grade X assessment the Economic Survey has proposed.
3. Screen every child, more than once: a universal, low-cost screen in Classes 3-5 that feeds JNVST and the state residential tests, and a real second entry point around Class 8, with seats of its own rather than only the vacancies today’s Class 9 lateral-entry test fills.
4. Grow selective seats towards 4-5% of a cohort, and add accelerated and enriched tracks to the selective residential schools, with district excellence centres for able children who stay in day schools.
5. Prepare and recognise teachers for the top, with separate training for teachers of accelerated groups.
6. Keep exams that reward the top, and put gifted provision in law so it outlasts any one government.
7. Design equity in from the start: quotas and local norms in selection, and results reported by state, school type, gender and family background.

The prize is large. Matching the best record on both numbers goes with about one more point of growth a year - the difference between missing the 2047 goal and meeting it.

Notes on the evidence

• Navodaya’s entry points come from NVS’s 2027-28 lateral-entry prospectuses for Class IX (https://cbseitms.nic.in/2026/nvsix_9 download) and Class XI (https://cbseitms.nic.in/2026/nvsxi_11 download), both admitting “only against the vacant seats”. Neither publishes a seat total; the Class 9 figure is a hand count of the vacancy annexures (about 4,000 for 2026-27 and 1,500 for 2027-28; EVIDENCE-REVIEW.md, Section 7.2). NCST is described in the same review’s glossary and Section 8.2.
• The growth estimates are cross-country correlations, and skills reach the economy only as cohorts start work, mostly after 2040.
• NAS “proficient” is not the same bar as PISA’s basic level, and India’s 1% comes from two states in 2005. India’s baseline should be read as an order of magnitude until it is measured on the international scale.
• Paths in the table add the skill effects to India’s 6.1% average growth over 2015-2025 and compare with NITI’s 7% path.
• The dollar figures in the opening scale NITI’s USD 30 trillion by each path’s growth rate over 2025-2047, using NITI’s own implied gap between real and dollar growth (about 2.2 points a year). They treat the skill effect as arriving in full, whereas in practice it builds as cohorts enter work, so most of it lands after 2040; the 2047 figures overstate how much is realised by then and understate what follows.
• PISA trends are noisy. Long-run top-tier rates average each country’s first two and last two rounds, so that one odd year cannot drive the result.
• Where top students sit is not proof that their schools made them. Türkiye selects at 14 and PISA tests at 15.
• Finding talent is not keeping it: 62% of India’s top 100 JEE scorers live abroad.

Appendix: the model countries

Each appendix gives the PISA maths trend (OECD, PISA 2025 Results, annex Table I.B1.2a.35 download), whether the rise is real, the policies in force with their dates and scale, what evidence exists on their effect, and what might transfer to India. “Basic” is the share at Level 2 or above; “top” is the share at Level 5 or above. Policy facts were read in the official documents linked; figures marked press come from news reports only.

A1. Türkiye

PISA maths	2003	2006	2009	2012	2015	2018	2022	2025
Basic, %	48	48	58	58	49	63	61	64
Top, %	5.5	4.2	5.6	5.9	1.1	4.8	5.4	7.9

Is the rise real?

• 2015 was an anomaly. OECD’s 2018 snapshot calls the 2015 results “anomalous” and says neither the fall nor the recovery reflects the long-run trajectory. TIMSS Grade 8, tested the same year, was stable (Advanced benchmark 7% in 2011, 6% in 2015). Quote trends from 2012, never 2015.
• PISA top tier: flat 2012–2022, up in 2025. The 2012–2025 change (+2.1 points) is not statistically significant; 2022–2025 (+2.5) is. The 2025 sample had more 15-year-olds in Grade 11 (13.7%), which may lift the top.
• TIMSS confirms a real rise in the top tail: Grade 8 Advanced benchmark 6% (2015), 12% (2019), 17% (2023); High benchmark 20% to 37%. Türkiye excluded 20% of students from TIMSS 2023, so the size should be discounted.
• Coverage did not create it. PISA’s coverage of 15-year-olds rose from 36% (2003) to 68% (2012) and has been flat at 70–74% since (Annex A2, Table I.A2.1 download). As a share of all 15-year-olds, the top tier roughly tripled, from about 2% (2003) to about 5.7% (2025).

Where the top performers are (Avanti analysis of OECD student microdata, matched to MEB’s published figures):

• Science high schools (Fen Liseleri) teach 5.6% of 15-year-olds and hold 46% of all maths Level 5+ students (2022); 44% of their students reach Level 5. Selective Anadolu high schools hold most of the rest; vocational schools, a quarter of students, almost none (0.1%).
• The 2018–2022 rise came from more seats, not a broad lift. The science-high-school share of 15-year-olds rose from 4.0% to 5.6% while the share reaching Level 5 inside them held at about 44%. In 2025 these schools averaged 606 in maths, at the Level 5 line (MEB PISA 2025 Türkiye Raporu download).
• The system is highly stratified: 55% of the variation in maths scores lies between schools, against 32% across the OECD (MEB PISA 2022 Türkiye Raporu, p.12 download).

Policies

• Science high schools (Fen Liseleri), 1964 onward. Ankara Fen Lisesi opened in 1964 with Ford Foundation funding; 13 schools by 1990. Public boarding and day schools in every province, entered by national exam. Public science high schools grew from 115 schools and 29,058 students (2010-11) to 366 schools and 143,152 students, plus 673 private ones with 60,898 (2023-24) (MEB Millî Eğitim İstatistikleri 2010-11 download and 2023-24 download; history: Suna et al. 2020 download). The ministry added about 35% more science-high-school seats in 2020 (LGS 2020 report download).
• Gifted centres (BİLSEM, Bilim ve Sanat Merkezleri), 1995 onward. After-school centres for children identified as gifted. Teachers nominate Grade 1–3 pupils (capped at 20% of a class); a tablet-based group screen and then an individual IQ assessment follow. Selected children get project-based enrichment of 4–12 hours a week in groups of 2–20, outside school hours, through five stages. Growth: 61 centres and 9,215 students (2010-11) to 392 centres, 102,580 students and 4,267 teachers (2023-24). Nominations rose from about 250,000 to 900,000 when screening moved to tablets in 2019. (BİLSEM directive, 2022 download; 2025-26 identification guide download.)
• The national transition exam. LGS (1998–2004), OKS, SBS, then TEOG (2013–2017, compulsory, six subject exams) and LGS again from 2018. Since 2018 the exam is optional and fills only selective places (science and social-science high schools, “project” schools); about 90% of students are placed locally without it. Yet about a million Grade 8 students sit it each year (994,358 in 2026), for about 200,000 exam-based places (2025). Maths and science carry the same weight as Turkish. (MEB LGS reports 2018 download, 2025 download, 2026 download.)
• National gifted strategy, 2013–2017. Ordered by the Science and Technology High Council in 2009; estimated 375,000–562,500 gifted children and set out identification and provision (MEB Özel Yetenekli Bireyler Strateji ve Uygulama Planı download).
• Science Olympiads (TÜBİTAK 2202). 19,062 applicants in 2022 (download).
• Tutoring centres (dershane) closed or converted from 2014. No causal study of the effect was found; tutoring had a small positive effect on university-entrance maths in a regression study (Berberoğlu and Tansel 2014 download).
• Spending did not drive the rise. Real spending per student fell between 2015 and 2022 (OECD Education at a Glance 2025, Türkiye note download).

Evidence of effect. None causal. A review of 107 theses on BİLSEM finds description and opinion surveys, no outcome evaluation (Oruç and Çağır 2022 download). Science-high-school students reach university at far higher rates (54% got a bachelor’s place in 2023, against 21% from Anadolu high schools), but they were selected for it.

For India. The analogue is a national exam-selected tier of residential science schools, sized to the cohort, with results published by school type. What not to copy: hard vocational tracking, and a high-stakes exam captured by private schools and tutoring (57% of 2025’s perfect scores came from private schools teaching 7% of pupils).

A2. United Arab Emirates

PISA maths	2003	2006	2009	2012	2015	2018	2022	2025
Basic, %	–	–	49	54	51	54	51	60
Top, %	–	–	2.9	3.5	3.7	5.4	5.3	7.0

Is the rise real? Mostly yes. PISA covers 83–98% of 15-year-olds throughout. The immigrant share of the sample was flat from 2015 (58%) to 2025 (57%), and both immigrant and non-immigrant students gained about 28–30 maths points. Immigrant students average about 80 points more, so the top tier probably sits mostly in private and international schools; the UAE does not publish the top tier by nationality (OECD PISA 2018 country note download; immigrant trend tables download).

Policies

• National Agenda, 2014. A target to rank in PISA’s top 20 and TIMSS’s top 15.
• Unified school inspection framework, 2015-16. Every school is graded on how it identifies gifted and talented students and how it supports them; definitions follow Gagné’s model, with enrichment, extension and acceleration inside mainstream classes (UAE School Inspection Framework 2015-16, §5.2.3–5.2.4 download).
• Dubai’s benchmarking. Each private school has PISA and TIMSS targets and must benchmark itself externally every year; Dubai’s regulator credits these for its gains (KHDA, Dubai PISA 2015, §2.4 download).
• National gifted-care policy, Ministerial Decree 209/2020. Early identification, enrichment and acceleration programmes, a national database and periodic evaluation (Ministry summary page download; the decree text itself could not be obtained).
• Hamdan Award (Dubai, 1998). Teacher screening, then ability and achievement tests; small, about 50 students at its 2023 summer camp (download).

Evidence of effect. None found.

For India. The cheapest levers: grade every school on identifying able children, and benchmark schools against international tests every year. Report the top tier by group, so a rise driven by private schools is visible.

A3. Romania

PISA maths	2003	2006	2009	2012	2015	2018	2022	2025
Basic, %	–	47	53	59	60	53	51	48
Top, %	–	1.3	1.3	3.2	3.3	3.2	4.0	4.2

Is the rise real? Yes, but the system widened rather than rose. The top tier jumped in one step (2009–2012) and drifted up after, while low performers rose from 41% (2012) to 52% (2025) and gaps by family background widened. Coverage of 15-year-olds (66–76%) did not create it: as a share of all 15-year-olds, the top tier still rose from 0.9% to 3.2%.

Policies

• Grade 8 national exam and computerised placement into high schools by score (in place by the early 2000s). OECD calls it a gatekeeper exam that identifies an elite; tracking happens at 14 (OECD Reviews of Evaluation and Assessment: Romania, 2017, pp.49, 101 download).
• Law 17/2007 on gifted education. Differentiated provision for gifted pupils through specialised classes, weekend schools, camps, mentoring and psychological identification (download).
• County centres of excellence (Centrele Județene de Excelență). About 42, one per county plus Bucharest, funded by the ministry; they run after-school “excellence groups” in Olympiad subjects (regulation, 2020 download; Iași centre notices download). National enrolment is not published.
• Olympiads tied to teacher pay. Students’ Olympiad placings count in the teacher merit-grade bonus. In 2023-24: 73 national Olympiads, 436 students at international competitions, 221 distinctions (Starea învățământului 2023-24 download).

Evidence of effect. Getting into a better school by just clearing the admission line adds about 0.05 SD on the school-leaving exam (Pop-Eleches and Urquiola, regression discontinuity across about 2,000 cut-offs download). No evaluation of the excellence centres or Olympiads.

For India. District excellence centres in law, run by the best local teachers, and teacher recognition tied to students’ results. The warning: without attention to the bottom, the gap widens.

A4. Poland

PISA maths	2003	2006	2009	2012	2015	2018	2022	2025
Basic, %	78	80	80	86	83	85	77	74
Top, %	10.1	10.6	10.4	16.7	12.2	15.8	9.4	10.0

Is the rise real? It was real but did not last: high in 2012–2018, back to its 2003 level by 2022. Coverage was steady (89–94%).

Policies (Jakubowski, in Crato, ed., Improving a Country’s Education, 2021, pp.142–152 download)

• 1999 structural reform. Tracking delayed by a year to 16, with a three-year lower-secondary school (gimnazjum), external exams from 2002 and per-pupil funding. It lifted students who would otherwise have gone to vocational school by about 100 points, and did not lift likely academic-track students (World Bank WPS 5263 download).
• 2008 curriculum. Learning outcomes rather than content lists, with modelling, strategic thinking and reasoning built into lower-secondary maths and its exam. The 2009–2012 gain came mostly on complex, analytical questions.
• 2017 reorganisation. The gimnazjum was abolished; the 2022 cohort was the first tested afterwards, and the top tier fell (government statement download).

Evidence of effect. Associational for the top tier.

For India. A curriculum and exam built on reasoning and modelling is the lever most closely tied to the top. Gains reverse when the system is reorganised.

A5. Portugal

PISA maths	2003	2006	2009	2012	2015	2018	2022	2025
Basic, %	70	69	76	75	76	77	70	65
Top, %	5.4	5.7	9.6	10.6	11.4	11.6	6.7	5.8

Is the rise real? It held for four cycles (2009–2018) and then fell back to its 2003 level. Coverage rose over the period, which works against a rising top share.

Policies

• Grade 9 national exams in maths and Portuguese, high-stakes from 2005.
• Maths Action Plan (Plano de Ação para a Matemática), 2006. Teacher in-service training, a resource bank and textbook evaluation (download).
• New maths programme, approved 2007 and generalised from 2009.
• 2011–2015: more maths time, learning standards (Metas), Grade 4 and 6 exams; these exams were abolished in 2016.

Evidence of effect. The timing of the 2006–2009 jump fits the 2005 exams, but no causal effect has been shown. Attribution of the later fall is disputed between a former minister and the maths teachers’ association (IAVE national report download; APM response download).

For India. National maths plans with teacher training can lift the top quickly; they need to be kept.

A6. Taiwan (Chinese Taipei)

PISA maths	2003	2006	2009	2012	2015	2018	2022	2025
Basic, %	–	88	87	87	87	86	85	85
Top, %	–	31.9	28.6	37.2	28.1	23.2	31.7	31.8

Is the rise real? The 2018–2025 rise is a recovery to the long-run level, not new growth. The 2015–2018 dip followed a 2014 change that turned the Grade 9 exam score into three broad bands, weakening the incentive to excel at the top, alongside falling birth rates that eased competition for places (Lin et al., in Crato 2021).

Policies

• Gifted provision in law since 1984. The Special Education Act defines six gifted categories and requires resource rooms, itinerant or special programmes for gifted pupils, and lets them take courses a grade ahead (Articles 4, 40, 44; amended 2023 download).
• Funding and scale. About NT$307–334 million a year for gifted education in 2005–07 (MOE Gifted Education White Book download); 0.75% of students identified in 1996.
• 12-year basic education (2014) and the 2019 curriculum (download).

Evidence of effect. None causal; the stability of the top tier over two decades coincides with continuous gifted provision.

For India. Put gifted provision in law so that it outlasts governments, and keep exam scores fine-grained at the top.

A7. Cautionary cases: Qatar and Azerbaijan

Qatar

PISA maths	2003	2006	2009	2012	2015	2018	2022	2025
Basic, %	–	13	26	30	41	46	44	46
Top, %	–	0.6	1.8	2.0	2.2	2.9	2.6	3.1

Part of the rise is composition. Immigrant students rose from 46% to 62% of the sample and score 66–83 points higher; OECD attributes about a third of the 2009–2018 reading gain to composition (PISA 2018 country note, p.3 download). Among non-immigrant students the basic share did improve. The reform of 2002 onward (“Education for a New Era”: independent schools and national testing of all grades) does not address gifted education (RAND 2007 download); its STEM high school (QAST, 2018) takes about 40 students a year.

Azerbaijan. PISA 2018 and 2022 tested Baku only, so there is no usable trend. The national series (0.9% top in 2006, 4.7% in 2025) came with a falling average (476 to 422) and a much wider spread. It runs a centralised exam body and an Olympiad funnel of about 45,000 students into about 25 international team members (2013 education strategy download).

The lesson from both: a rising top share can reflect who is tested rather than what the system develops. India should track its top tier within groups, by state, school type, gender and family background.

A8. All 80 PISA systems with a trend

Every system in the OECD’s PISA trend table (PISA 2025 Results, annex Table I.B1.2a.35 download) with at least six years between its first and last maths results. ★ marks the 25 systems similar to India: at their first PISA round, at most 40% of 15-year-olds were at the basic level (Level 2 or above) and at most 2% at the top (Level 5 or above). India would sit in this group (about 22% and 1%).

• “Change a year” is the change from the first to the last round, divided by the years between, in percentage points. These are whole-period rates; Section 4’s top-tier rates instead average the first two and last two rounds to damp single odd years.
• Rounds differ by system: many joined in 2015 or 2018, so their trends are short.
• Read individual rows with the cautions in A1–A7: Türkiye’s 2015 round was an anomaly; Qatar’s rise is partly a change in who was tested; Azerbaijan’s 2018 and 2022 rounds tested Baku only and are not in this table, so its row compares 2006 with 2025; the asterisk on Albania is the OECD’s own flag.
• The similar group’s median is +0.19 points a year at the basic level and zero at the top. Across the other 55 systems the median is negative on both (−0.41 and −0.09): many richer systems declined over the period.

	System	Years	Basic: first → last	Top: first → last	Basic change a year	Top change a year
★	Zambia	2018–2025	2% → 15%	0.0% → 0.0%	+1.87	+0.00
★	Qatar	2006–2025	13% → 46%	0.6% → 3.1%	+1.74	+0.13
★	Albania*	2009–2025	32% → 55%	0.4% → 1.5%	+1.44	+0.07
★	Cambodia	2018–2025	10% → 18%	0.0% → 0.0%	+1.18	-0.00
★	Georgia	2009–2025	31% → 47%	0.6% → 2.0%	+1.01	+0.09
★	Saudi Arabia	2018–2025	27% → 32%	0.2% → 0.3%	+0.69	+0.02
★	Kyrgyzstan	2006–2025	11% → 19%	0.1% → 0.0%	+0.43	-0.00
★	Moldova	2009–2025	39% → 44%	0.7% → 1.0%	+0.31	+0.02
★	Philippines	2018–2025	19% → 21%	0.1% → 0.0%	+0.26	-0.01
★	Peru	2009–2025	26% → 30%	0.6% → 0.3%	+0.25	-0.02
★	Montenegro	2006–2025	40% → 45%	0.8% → 0.9%	+0.24	+0.00
★	North Macedonia	2015–2025	30% → 32%	0.8% → 0.7%	+0.22	-0.01
★	Paraguay	2018–2025	8% → 10%	0.0% → 0.0%	+0.19	+0.00
★	Brazil	2003–2025	25% → 28%	1.2% → 0.4%	+0.16	-0.04
★	Colombia	2006–2025	28% → 29%	0.4% → 0.2%	+0.05	-0.01
★	Dominican Republic	2015–2025	9% → 9%	0.0% → 0.0%	-0.01	-0.00
★	Jordan	2006–2025	34% → 31%	0.2% → 0.1%	-0.15	-0.01
★	Mexico	2003–2025	34% → 30%	0.4% → 0.1%	-0.17	-0.01
★	Indonesia	2003–2025	22% → 17%	0.2% → 0.0%	-0.22	-0.01
★	Guatemala	2018–2025	11% → 8%	0.0% → 0.0%	-0.39	+0.00
★	Argentina	2006–2025	36% → 24%	1.0% → 0.4%	-0.61	-0.04
★	Kosovo	2015–2025	22% → 15%	0.0% → 0.0%	-0.74	-0.00
★	Lebanon	2015–2025	40% → 31%	2.0% → 0.6%	-0.92	-0.14
★	Morocco	2018–2025	24% → 16%	0.1% → 0.1%	-1.18	-0.01
★	Ecuador	2018–2025	29% → 20%	0.0% → 0.0%	-1.32	-0.00
	Türkiye	2003–2025	48% → 64%	5.5% → 7.9%	+0.75	+0.11
	United Arab Emirates	2009–2025	49% → 60%	2.9% → 7.0%	+0.73	+0.26
	Brunei Darussalam	2018–2025	52% → 55%	3.0% → 2.4%	+0.44	-0.10
	Kazakhstan	2009–2025	41% → 44%	1.2% → 1.2%	+0.21	+0.00
	Macao (China)	2003–2025	89% → 90%	18.7% → 28.8%	+0.07	+0.46
	Romania	2006–2025	47% → 48%	1.3% → 4.2%	+0.05	+0.16
	Italy	2003–2025	68% → 69%	7.0% → 7.0%	+0.03	-0.00
	Singapore	2009–2025	90% → 89%	35.6% → 37.4%	-0.09	+0.11
	Japan	2003–2025	87% → 84%	24.3% → 22.3%	-0.13	-0.09
	Chinese Taipei	2006–2025	88% → 85%	31.9% → 31.8%	-0.13	-0.00
	Poland	2003–2025	78% → 74%	10.1% → 10.0%	-0.17	-0.01
	B-S-J-Z (China)	2018–2025	98% → 96%	44.3% → 54.2%	-0.17	+1.41
	Bulgaria	2006–2025	47% → 43%	3.1% → 2.8%	-0.20	-0.01
	Israel	2006–2025	58% → 54%	6.1% → 4.8%	-0.21	-0.06
	Chile	2006–2025	45% → 41%	1.5% → 0.4%	-0.21	-0.06
	Portugal	2003–2025	70% → 65%	5.4% → 5.8%	-0.23	+0.02
	Thailand	2003–2025	46% → 41%	1.6% → 0.8%	-0.24	-0.04
	Estonia	2006–2025	88% → 83%	12.5% → 13.3%	-0.27	+0.04
	Hong Kong (China)	2003–2025	90% → 83%	30.7% → 20.7%	-0.30	-0.45
	Lithuania	2006–2025	77% → 71%	9.1% → 6.0%	-0.30	-0.16
	Ireland	2003–2025	83% → 76%	11.4% → 6.1%	-0.34	-0.24
	United Kingdom	2006–2025	80% → 74%	11.1% → 12.7%	-0.35	+0.09
	Malaysia	2009–2025	41% → 35%	0.4% → 0.6%	-0.36	+0.01
	Mauritius	2009–2025	50% → 44%	1.6% → 0.8%	-0.39	-0.05
	Croatia	2006–2025	71% → 64%	4.7% → 5.3%	-0.40	+0.03
	Switzerland	2003–2025	85% → 77%	21.2% → 15.1%	-0.40	-0.28
	Austria	2003–2025	81% → 72%	14.3% → 7.8%	-0.41	-0.29
	Latvia	2003–2025	76% → 67%	8.0% → 4.5%	-0.41	-0.16
	Korea	2003–2025	90% → 81%	24.8% → 23.2%	-0.41	-0.07
	United States*	2003–2025	74% → 65%	10.1% → 8.0%	-0.43	-0.10
	Uruguay	2003–2025	52% → 42%	2.8% → 0.9%	-0.45	-0.08
	Spain	2003–2025	77% → 67%	7.9% → 3.9%	-0.48	-0.18
	Hungary	2003–2025	77% → 66%	10.7% → 5.2%	-0.48	-0.25
	Greece	2003–2025	61% → 50%	4.0% → 2.1%	-0.48	-0.08
	Serbia	2006–2025	57% → 47%	2.8% → 2.3%	-0.53	-0.03
	Belgium	2003–2025	84% → 72%	26.4% → 9.7%	-0.54	-0.76
	Slovak Republic	2003–2025	80% → 68%	12.7% → 9.1%	-0.56	-0.16
	Czechia	2003–2025	83% → 71%	18.3% → 9.8%	-0.56	-0.39
	Germany	2003–2025	78% → 66%	16.2% → 7.8%	-0.58	-0.38
	Denmark	2003–2025	85% → 72%	15.9% → 5.4%	-0.59	-0.48
	New Zealand*	2003–2025	85% → 71%	20.7% → 11.4%	-0.64	-0.42
	Malta	2009–2025	66% → 56%	7.7% → 5.4%	-0.64	-0.14
	Australia	2003–2025	86% → 71%	19.8% → 9.9%	-0.67	-0.45
	Luxembourg	2003–2025	78% → 63%	10.8% → 6.9%	-0.69	-0.18
	Canada*	2003–2025	90% → 74%	20.3% → 11.1%	-0.72	-0.42
	Sweden	2003–2025	83% → 66%	15.8% → 7.7%	-0.75	-0.37
	Norway*	2003–2025	79% → 62%	11.4% → 5.4%	-0.78	-0.27
	Costa Rica	2009–2025	43% → 30%	0.3% → 0.1%	-0.80	-0.01
	Netherlands*	2003–2025	89% → 70%	25.5% → 12.8%	-0.86	-0.58
	France	2003–2025	83% → 64%	15.1% → 5.2%	-0.87	-0.45
	Slovenia	2006–2025	82% → 65%	13.7% → 5.9%	-0.90	-0.41
	Cyprus	2012–2025	58% → 44%	3.7% → 3.4%	-1.04	-0.03
	Iceland	2003–2025	85% → 62%	15.5% → 4.1%	-1.04	-0.52
	Finland	2003–2025	93% → 69%	23.4% → 6.7%	-1.10	-0.76
	Azerbaijan	2006–2025	89% → 48%	0.9% → 4.7%	-2.16	+0.20
	Median, the 25 similar systems				+0.19	-0.00
	Median, the other 55				-0.41	-0.09

Sources

Every source was downloaded and read. Each [download] link opens the archived copy for anyone signed in with an avantifellows.org account.

• Government of India

  • Ministry of Education, National Education Policy 2020. https://ncert.nic.in/pdf/nep/NEP_2020.pdf download
  • Ministry of Education, NIPUN Bharat guidelines (2021). https://static.pib.gov.in/WriteReadData/specificdocs/documents/2021/jul/doc20217531.pdf download
  • NITI Aayog, Viksit Bharat working paper (July 2024) download; Scenarios for Viksit Bharat and Net Zero, Vols 1–2 (February 2026). download
  • Ministry of Finance, Economic Survey 2025-26, Chapter XI. download
  • IMF (2026). World Economic Outlook, April 2026 (DataMapper: India GDP and GDP per head, current US$). download

• Growth and skills

  • Hanushek, E. and L. Woessmann (2012). “Do better schools lead to more growth?” Journal of Economic Growth 17: 267–321. download
  • Hanushek, E. and L. Woessmann (2011). “How much do educational outcomes matter in OECD countries?” Economic Policy 26(67). NBER w16515. download
  • Hanushek, E. and L. Woessmann (2021). “Education and economic growth.” Oxford Research Encyclopedia of Economics and Finance. download
  • Heller-Sahlgren, G. and H. Jordahl (2023). “Test scores and economic growth: update and extension.” Applied Economics Letters 31(11). https://eprints.lse.ac.uk/118145/ download
  • Gust, S., E. Hanushek and L. Woessmann (2024). “Global universal basic skills.” Journal of Development Economics 166: 103205. download
  • Bloom, N., C. Jones, J. Van Reenen and M. Webb (2020). “Are ideas getting harder to find?” American Economic Review 110(4): 1104–1144. https://doi.org/10.1257/aer.20180338 download
  • Castelló-Climent, A. and A. Mukhopadhyay (2013). “Mass education or a minority well educated elite in the process of growth: the case of India.” Journal of Development Economics 105: 303–320. download
  • Breton, T. (2014). “Higher test scores or more schooling?” Working paper; published in Journal of Human Capital (2015). download

• Top talent

  • Agarwal, R. and P. Gaulé (2020). “Invisible geniuses.” AER: Insights 2(4). Read as IZA DP 11977. https://docs.iza.org/dp11977.pdf download
  • Choudhury, P., I. Ganguli and P. Gaulé (2023). “Top talent, elite colleges, and migration: evidence from the Indian Institutes of Technology.” NBER w31308. download
  • Hsieh, C., E. Hurst, C. Jones and P. Klenow (2019). “The allocation of talent and U.S. economic growth.” Econometrica 87(5). download
  • Card, D. and L. Giuliano (2016). “Universal screening increases the representation of low-income and minority students in gifted education.” PNAS 113(48). Read as NBER w21519. download

• India learning levels

  • Das, J. and T. Zajonc (2010). “India shining and Bharat drowning.” Journal of Development Economics. Read as World Bank PRWP 4644 (2008). download
  • World Bank (2024). India Learning Poverty Brief. download
  • OECD (2026). PISA 2025 Results, Vol I, annex Tables I.B1.2a.11 and I.B1.2a.12. download

• Top-tier improvers (archived with this note; country reports in Turkish, Romanian, Portuguese and Arabic read in the original)

  • OECD (2026). PISA 2025 Results, Annex A2, Table I.A2.1 (coverage), and Table I.B1.2a.35 (low and top performers, 2003–2025). download
  • MEB (Türkiye Ministry of National Education). Millî Eğitim İstatistikleri: Örgün Eğitim 2010-11, 2017-18 and 2023-24; PISA 2022 Türkiye Raporu; LGS reports 2018 and 2025; Özel Yetenekli Bireyler Strateji ve Uygulama Planı 2013–2017; BİLSEM directive and 2025-26 identification guide. 2023-24 statistics PISA 2022 report gifted strategy
  • MEB (2026). PISA 2025 Türkiye Raporu (results by school type). download
  • OECD. PISA 2018 and 2022 student microdata, Türkiye, analysed by school type with plausible values and replicate weights (Avanti analysis, archived with this note).
  • Suna, H. E., H. Tanberkan, B. Gür, M. Perc and M. Özer (2020). “Science high schools in Türkiye” (Fen Lisesi intake and university outcomes). download
  • IEA. TIMSS 2019 and TIMSS 2023 International Results in Mathematics (Türkiye, Grade 8 benchmarks and exclusions).
  • OECD (2017). OECD Reviews of Evaluation and Assessment in Education: Romania. download
  • Pop-Eleches, C. and M. Urquiola (2013). “Going to a better school: effects and behavioral responses.” American Economic Review 103(4). Read as NBER w16886. download
  • UAE (2015). UAE School Inspection Framework 2015-16. download
  • KHDA (2017). Dubai PISA 2015. download
  • OECD (2019). PISA 2018 country notes: Qatar; United Arab Emirates. Qatar UAE
  • Crato, N., ed. (2021). Improving a Country’s Education: PISA 2018 Results in 10 Countries. Springer (open access). Chapters on Poland, Portugal and Chinese Taipei. download

• India’s national talent programmes

  • Lok Sabha Unstarred Question 1704 (10 March 2025), on NTSE. download
  • Department of Science and Technology. “INSPIRE-SHE: encourage talented students towards science research careers without KVPY examination” (July 2022). download
  • PIB (2025, 2026). INSPIRE scheme releases (PRID 2115859, 2201552); NMMSS 2026-27 (PRID 2305633). INSPIRE NMMSS
  • ETV Bharat (13 August 2026). Rajya Sabha reply on INSPIRE-SHE “under redesign” (press report). download
  • HBCSE. Olympiad funding page. download
  • Bhattacharji, P. and G. Kingdon (2017). The Great Indian Exam Debacle. ORF Issue Brief 191. download
  • NCERT (2022). Rationalisation of content in the textbooks, Class X booklet. download
  • CBSE. Class X and XII result press notes, 2023 and 2024 (no merit list, no divisions). download

• India’s residential schools

  • Avanti Fellows (2026). Selection-based government residential schools in India (census of 117 systems). Internal analysis, analysis/selection-residential-schools/.

• India learning levels (additional)

  • NCERT (2021). National Achievement Survey 2021, state and national results, Class X mathematics (Avanti’s warehouse copy).