Learning & Skills Brief — October 2026

Between 7 September and 6 October, the strongest evidence was about how learners are grouped and who teaches them, not about new curricula. A national evaluation in France found no learning gain from grouping pupils by level. A US experiment found that a tutor's experience inside a programme mattered more than their degrees. East African systems moved from training teachers and workers to certifying them. The main pooled fund for basic education ended its pledging event well short of its target.

Global education's pooled fund raised less than half its target — and the UK cut its pledge by 77%

• Source: Global Partnership for Education · 23 September 2026 · https://www.globalpartnership.org/news/new-financing-boosts-gpe-improve-education-750-million-children · corroborating: Save the Children UK · 24 September 2026 · https://www.savethechildren.org.uk/news/media-centre/press-releases/2026/un-general-assembly-slashing-global-partnership-education
• At a UN General Assembly event on 23 September, co-hosted by Italy and Nigeria, GPE announced $2.1 billion in confirmed and anticipated pledges to its 2026–2030 fund, against a $5 billion target. It also announced $1.4 billion in co-financing ($1 billion from development finance institutions, $400 million from philanthropic and private partners), against a $10 billion co-financing goal. Italy will double its contribution. The UK pledged £100 million, down from £430 million in 2021 (−77%). Partner countries announced $471 billion in domestic education financing commitments, a 32% increase in spending by 2030.
• Why it matters: For programmes running from 2027 to 2030 in GPE partner countries, the announced growth comes from national budgets, not donors. Proposals will increasingly have to fit ministry budget lines and priorities rather than donor funding windows. Pledging remains open, so the final total may rise.

Grouping pupils by level did not raise test scores in France's first national evaluation

• Source: DEPP (statistics directorate of the French Ministry of Education), working paper "Évaluation des organisations des enseignements de français et de mathématiques en sixième et cinquième" · September 2026 (relayed 25 September) · https://www.education.gouv.fr/depp/evaluation-des-organisations-des-enseignements-de-francais-et-de-mathematiques-en-sixieme-et-505810 · relay: OZP · https://ozp.fr/spip.php?article35618=
• Since September 2024, French lower-secondary schools have been required to regroup 6e–5e pupils (ages 11–13) in French and maths. A survey of a representative sample of 1,500 public lower-secondary schools found that 94% did so, in very different ways:
  • 21% formed mixed-ability groups.
  • 50% grouped by prior attainment, including 19% with homogeneous groups and 15% grouping struggling pupils.
  • 23% combined several approaches.
• The first causal analyses compared the 2024–25 cohort's national test scores at the start of 4e with earlier cohorts, and compared types of grouping. They found no significant effect in French or maths, including by initial level and by sex. Moving pupils between groups during the year, which was meant to keep groups flexible, rarely happened.
• Why it matters: Splitting learners by level is not in itself a remediation method. Here it used staff hours (other activities were cut to fund the groups) and produced no measurable gain. Anyone designing level-based cohorts should specify what is taught differently in each group and how learners move between groups, and budget for evaluating it. Caveat: the results cover one year of exposure, and the type of grouping is as reported by principals.

In online tutoring, experience in the programme predicted results; degrees and teaching experience did not

• Source: Huffaker, Levin Markel, Robinson & Loeb, EdWorkingPaper 26-1595, Annenberg Institute (Brown University) · September 2026 · https://edworkingpapers.com/ai26-1595
• The setting is a high-dosage online literacy tutoring programme for primary pupils (kindergarten to grade 5) in two US states, with tutors randomly assigned to groups. Groups had 3–4 pupils and met at least three times a week for a semester; the sample is 1,638 students in 2022–23.
• Tutors' bachelor's degrees and classroom teaching experience did not predict larger reading gains. In this virtual setting, Black pupils did not gain more with Black tutors. Tutors who had prior experience in the same programme were associated with about 0.2 standard deviations more literacy growth.
• Why it matters: For programmes that rely on large pools of tutors or facilitators, results seem to come from keeping and developing people inside a structured programme, more than from recruiting on credentials. Caveat: one US online provider, and the experience effect is an association across randomly assigned tutors, not a randomised test of training.

Rwanda moves from English training to English certification for teachers

• Source: Rwanda Basic Education Board (REB) communication to districts, 25 September 2026, reproduced by Afrocareers · 27 September 2026 · https://www.afrocareers.com/2026/09/reb-announcement-for-all-primary-and.html · context: The New Times · September 2025 · https://www.africa-press.net/rwanda/?p=110121
• REB announced that the computer-based English Proficiency Assessment, previously postponed, will start on 31 October 2026. It covers teachers who have completed the course content leading to B2 certification, starting with the City of Kigali and the Northern and Western provinces; other provinces will follow. Head teachers are asked to give teachers access to school computers to practise.
• Context: the 2024/25–2028/29 Education Sector Strategic Plan, cited by The New Times in September 2025, found that about 4% of primary teachers and 38% of secondary teachers met the B2 level. The target is B2 for all teachers and school administrators (over 116,000) by 2029, through an in-service programme under the World Bank–supported RQBE project.
• Why it matters: Once a course is followed by an external test, the effectiveness of in-service training becomes measurable, and so does its failure rate. A computer-based test also adds a digital-skills requirement to the language requirement; preparation for this cohort needs to cover both.

Uganda's TVET Act opens a route to certify skills learned on the job

• Source: The Independent (Kampala), via allAfrica · 23 September 2026 · https://allafrica.com/stories/202609230097.html · corroborating: ILO programme note (Lovguiden) · 21 May 2026 · https://lovguiden.dk/en/det-offentlige/den-internationale-arbejdsorganisation/2026-05-21-ilo-program-giver-flygtninge-i-uganda-formel-faglig-certificering
• The TVET Act 2025 provides for recognition of prior learning. Workers can have skills gained through experience assessed against occupational standards, and receive nationally recognised certificates aligned with Uganda's TVET qualifications framework. The Uganda Vocational and Technical Assessment Board (UVTAB) assesses and certifies. Candidates register through accredited assessment centres and may have to provide evidence of experience (letters, work samples).
• An ILO-supported programme has already adapted to this requirement. In March 2026 it assessed 70 candidates (40 refugees, 30 members of the host community) in trades such as welding, masonry and poultry farming.
• Why it matters: Where most skills are learned informally, certifying without full training becomes a design option: a short course to fill gaps, followed by assessment, instead of a full course. Volumes, fees and pass rates under the Act are not yet published. Cost per certificate is the figure to get before planning around this route.

OECD: 9% of teachers are not fully qualified, and shortages concentrate in maths, science and remote schools

• Source: OECD, Education at a Glance 2026 · 29 September 2026 · https://www.oecd.org/en/about/news/press-releases/2026/09/teacher-shortages-are-putting-pressure-on-education-systems-across-the-oecd.html · relay: IEU Monitoring · https://ieu-monitoring.com/editorial/oecd-cautions-that-teacher-shortages-are-putting-core-subjects-under-pressure/1251050
• Qualifications: in 2024/25, 9% of primary and secondary teachers across OECD countries were not fully qualified. The share exceeded 18% in Chile, Denmark, Estonia and Germany.
• Vacancies and shortages: unfilled posts stayed below 4% of the teaching workforce in all 20 systems with data. However, 11 of the 16 countries with data reported shortages of qualified secondary maths and science teachers.
• Age, pay and status: one secondary teacher in four is 55 or older. Starting salaries for lower-secondary teachers average 61% of the earnings of similarly educated workers. 85% of teachers are satisfied with their job, but only 22% think society values teaching.
• Why it matters: Even high-income systems now rely on a significant share of teachers who entered without full initial training. The question has shifted from avoiding unqualified teachers to training them while they teach: induction and in-service support become a permanent need, not an exception. The figures cover OECD and partner countries, mostly outside Africa (South Africa is among the partners).

Kenya trains its own teachers to write competency-based exam items

• Source: Kenyans.co.ke, citing a KNEC notice · 15 September 2026 · https://www.kenyans.co.ke/news/127065-knec-opens-training-senior-school-and-tvet-teachers-exam-item-writing
• The Kenya National Examinations Council opened applications (deadline 30 September) for online item-writing training for senior school and vocational-level teachers. The training runs from 12 to 30 October 2026 and focuses on writing test items for competency-based assessment; teachers who complete it will take part in assessment work. Applicants must currently teach the subject, have at least one year of experience and be registered with the Teachers Service Commission. Context: the first cohort under the competency-based curriculum entered senior school (Grade 10) in January 2026.
• Why it matters: In a competency-based system, the bottleneck moves from the curriculum to assessment: someone has to write tasks that test what learners can do. Building that capacity among serving teachers costs less than relying on a central team, but quality depends on how items are reviewed. The training content and the moderation process are what to watch.

AI models now outscore most students on university law exam questions

• Source: Alimardani, Porter & Gullett (University of Wollongong), The Conversation, republished by Phys.org · 30 September 2026 · https://phys.org/news/2026-09-ai-exams-universities.html
• The researchers ran nine AI models from five providers on real exam questions from two compulsory law subjects at the University of Wollongong (Australia). The models had internet access but no course materials. 18 papers were graded, 12 of them blind by tutors.
• In criminal law, the AI papers averaged 76.3%, scoring above 82.5% of students. In torts they averaged 66%, above 61% of students. 7 of the 18 papers ranked at or above the 90th percentile. In the team's 2023 test, the criminal-law average was 52.5%.
• Why it matters: Any certification that relies on unsupervised written work (take-home assignments, online exams, written case studies) can no longer reliably tell the candidate's work from the tool's. The authors propose combining three formats: supervised exams without AI, AI-assisted tasks graded on the process, and a "relay" format (supervised draft, refinement with AI, then a demonstration of independent verification). One subject and one university; the robust signal is how much the models improved in three years.