KALIMANI NEW REPORT 2026

View KALIMANI NEW REPORT 2026 flipbook.

  partnering with

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT ABOUT THIS DOCUMENT Recommended Citation Notes Reviewers Ndum, M., Kakoti, M., Neghestisumari, N., Ikonya, S., Odeny, T., Hwaga, D., Omare, M., & Mbah, O. P. (2026). Kalimani AI inclusive teacher model: Learning report. eBASE Africa, Jenga Hub, & The Action Foundation. This work was supported by the Gate foundation under the Daara development Academy. Pearl Boateng, Kiran Sahib, Rigobert Pambe A B O U T D A A R A Daara is a pan-African community of education leaders who come together to learn, innovate, and lead efforts to strengthen education systems across sub-Saharan Africa. Its first phase, launched in January 2024, brought together 11 organisations from eight countries, all working to advance foundational learning. A second 12-month programme began in June 2025, bringing together participants from the initial cohort alongside seven newly selected members. Across both phases, Daara has focused on strengthening the capacity of education leaders and organisations to improve foundational learning at scale. This work responds to a persistent global learning crisis, in which millions of children attend school but do not acquire foundational skills at the expected level. According to the World Bank, nine in ten children in sub-Saharan Africa are unable to read and understand a simple text by the age of 10. Foundational literacy and numeracy provide the basis for subsequent learning, making their development a critical priority for education systems. Daara partners share a common goal of improving children’s literacy and numeracy outcomes across sub-Saharan Africa. There is now a growing body of ev- idence on effective, age-appropriate approaches to teaching children to read and develop mathematical skills, and Daara provides a platform through which education leaders can learn from this evidence, adapt it to their contexts, and support its implementation. Further information is available on the Daara website

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT TABLE OF CONTENTS

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT LIST OF TABLES Table 1: Sample by country and Assessment point Table 2: Overall sample across countries per Arm Table 3: Acceptability and feasibility of the Kalimani model Table 4: Summary score of teacher compliances to inclusive classroom practices Table 5: CBE Band Distribution, Baseline to Endline by Group, Grade and Country Table 6: Learners Engagement and participation levels Table 7: Annex B Summary of pilot Table 8 Annex C: competency-based curriculum scoring rules Table 9: Annex D: Learner Numeracy Mean Score by School Table 10 Appendix E: Learner Numeracy CBC Band Distribution by School: Grade 2 Table 11 Appendix F: Learner Numeracy: CBC Band Distribution by School: Grade 3 Table 12 Annex G: Numeracy Components Table 13: Annex H: Numeracy Components Table 14 Annex I: Numeracy skills fluency and Problem-Solving Kenya Table 15 annex J: Learner Engagement School-Level Table 16 Annex K: Comparison Analysis

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT LIST OF FIGURES Figure 1: Overall sample across grade and groups Figure 2: Learner means score by arm, baseline to endline Figure 3: Numeracy skills Figure 4: Qualitative insights from Tanzania Figure 5: Qualitative Insights from Kenya Figure 6: theory of change Figure 7: Intervention framework

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT LIST OF ABBREVIATIONS AP Approaching Expectations BE Below Expectations CEMASTEA Centre for Mathematics, Science and Technology Education in Africa CHAVITA Chama cha Viziwi Tanzania (Tanzania Association of the Deaf) D/HH Deaf and Hard of Hearing EE Exceeding Expectations EGMA Early Grade Mathematics Assessment JKUAT ISERC Jomo Kenyatta University of Agriculture and Technology Institutional Scientific and Ethics Review Committee KICD Kenya Institute of Curriculum Development KII Key Informant Interview KJSEA Kenya Junior School Education Assessment KoBo Kobo Toolbox (digital data collection platform) KSL Kenyan Sign Language ME Meets Expectations MoE Ministry of Education MoEST Ministry of Education, Science and Technology (Tanzania) NRD Numeracy Research and Development (Expert Panel) TIE Tanzania Institute of Education TSL Tanzania Sign Language

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 1.EXECUTIVE SUMMARY The Kalimani project is an AI teacher-support model in Kenya and Tanzania aiming to boost numeracy for deaf and hard-of-hearing Grade 2 and 3 learners through sign-language visual demonstration. It is rooted in inclusive teacher professional development and coaching to strengthen teacher adoption of best practice in inclusive numeracy instruction. It reflects global evidence that teachers are the most important in-school factor for learner outcomes (World Bank, 2018), and that deaf learners’ mathematics achievement improves with visual, signed instruction, with sign language proficiency positively associated with numeracy performance (Henner et al., 2021). The Kalimani Model has three parts: Government support was mobilised from the start. Both governments went beyond endorsing the model; they contributed to resource development, quality assurance, coaching and feedback sessions, and alignment with national curricula and policies. This involvement strengthened classroom practice and ensured the training translated into consistent, inclusive teaching. By embedding coaching cycles within ministry structures, the intervention paved a credible pathway for systemic adoption and sustainability. All models built to work in the hands of a trained teacher. To strengthen implementation, teachers received a three-day inclusive-numeracy training and ministry-led coaching at Weeks 4 and 8. an avatar that teaches numeracy in Kenyan and Tanzanian Sign Language, a quiz on a tablet, remediation driven by data from the quiz The pilot ran over eight weeks across 12 purposely selected government schools, 8 intervention schools (4 in Kenya, 4 in Tanzania) and 4 control schools (2 in each country), spanning both special-unit and inclusive settings. Across the 12 schools, 309 learners were assessed at baseline and 319 at endline, with all analyses of learner change based on the matched cohort assessed at both points. 16 intervention teachers and 8 control teachers were observed across the 2 countries.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT The evidence on acceptability and feasibility of the Kalimani model, suggests the model was deliverable across several dimensions. Teacher professional development training produced measurable readiness: inclusive-pedagogy and formative-assessment knowledge rose from 30% to 90%, and digital confidence improved markedly, with reported digital-tool use in mathematics rising from 37% to 75%. In terms of teacher’s compliance with inclusive best practices, teacher instruction rose from baseline to endline in intervention schools from 42% to 85% (Grade 2) and 39% to 88% (Grade 3) in Tanzania, and from 57% to 74% and 56% to 88% in Kenya, while control schools barely moved. Another key finding within the pilot was that; overall, learner numeracy improved in both in- tervention and control groups over the pilot period. At baseline, most intervention learners were below expectations; by endline, a great number had moved into the Approaching and Meeting bands, with the largest gains among those who started at the lowest baseline. The control groups also improved (see Table 5; Figure 2). If we look deeper and consider specific skill types, fluency and problem-solving gains were strongest in Tanzania, though the applied problem-solving domains, where deaf and hard-of-hearing learners struggle most, showed the least movement and remained low throughout (Figure 3; full detail in Annex G). Both arms progressed; therefore, such gains are not attributed solely to the model; they reflect ordinary learner development and a full term of teaching, alongside the Kalimani model in intervention schools. Another teacher in Tanzania reported, ‘I have seen a 50% increase in engagement from when we started teaching using the Kalimani app, retention and collaboration also increased thanks to the Kalimani TV and tablets, now teaching is easier because there are a lot of teaching materials in the App’. This pattern is consistent with the strengthened practice being supported by the coaching around the model, and points to teacher acceptability as one of the pilot’s aims. The same uptake is visible in the app data. Active users were teachers using the app’s teaching resources and learners completing quizzes more than doubled over the eight-week implementation period, from (187 to 396). the training on digital skills and integration into my daily teaching really strengthened my skill and made my teaching very easy,” KENYAN TEACHER now students participate more in learning, and I now teach with confidence TANZANIAN TEACHER “ “ Also, learners’ engagement levels in the classroom increased strongly in intervention schools, reaching 82% to 90% across grades, with participation and responsiveness improving, while Control arms moved unevenly. Tanzania Grade 2 engagement increased (46% to 54%) and Grade 3 increased by (53% to 89%), while Kenya Grade 3 control weakened slightly (48% to 50% engagement, responsiveness down to 45%) This provides evidence to suggest that model strengthened engagement. As seen on table (Table 6). The pilot indicates that combining visualisation, demonstration and sign-supported instruction make numeracy more accessible for deaf and hard-of-hearing learners, consistent with evidence that these learners’ reason through visual-spatial representation (Blatto- Vallee & Kelly, 2007). Coaching delivered through ministry structures strengthened teachers’ inclusive practice, and in the intervention classrooms the model reliably built the participation and collaboration that research links to later learning (Fredricks, Blumen- feld & Paris, 2004). These findings point to Kalimani’s feasibility and acceptability, with coaching emerging as a critical lever for embedding inclusive pedagogy. While the pilot could not yet attribute learning gains to the model, these are early signals worth confirming in a larger, longer study with matched groups. Continued ministry-led coaching will be essential to hold practice gains and anchor the model within national systems. Together, the pilot offers credible early signals of model feasibility, aligning with SDG 4 and the Continental Education Strategy for Africa.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 2. INTRODUC TION Across sub-Saharan Africa, inclusive education is often neglected despite expanding policy commitments. Learners with disabilities, including deaf and hard-of-hearing learners, continue to face persistent barriers to foundational learning, especially in early numeracy (UNESCO, 2020; UNICEF, 2023). These barriers are not primarily a matter of learner ability but rather systemic gaps in teacher preparation, classroom communication, and access to appropriate instructional tools (Ojok et al., 2022; Uwineza, Uworwabayeho & Yokoyama, 2024). In both Tanzania and Kenya, hearing-related functional difficulties are a significant but under-addressed challenge in mainstream classrooms: national survey data indicate that a measurable proportion of children experience hearing difficulties that affect their ability to follow instruction (National Bureau of Statistics Tanzania, 2014; Kenya National Bureau of Statistics, 2019), and school-based screening in Tanzania confirms hearing impairment among a notable share of primary learners (Mkumbo et al., 2020). Yet most of these learners are taught by teachers with limited training in sign-supported instruction and few practical tools to adapt their teaching, particularly in numeracy, where abstract concepts rely heavily on spoken explanation (Ojok et al., 2022) 2.1 Background & Rationale 2.2 Why it matters for foundational learning Foundational numeracy is widely recognised as the stage where exclusion has the greatest impact, serving as the gateway skill for all later mathematics Studies indicate that deaf learner’s performance in mathematics is weaker than their hearing peers, particularly in foundational domains. This reflects limitations in the pedagog- ical approach, especially reliance on oral teaching methods, rather than reduced cognitive capacity Evidence suggests that the steepest difficulties arise when the curriculum shifts from concrete to abstract concepts such as place value, number operations, and applied problem-solv- ing, because these domains are conventionally taught through spoken expla- nation, which many deaf learners cannot fully access (World Bank, 2019). (UNESCO, 2020; Bavelier et al., 2006). (Werner & Hänel-Faulhaber, 2026). The reliance on oral methods at the very early stage where numeracy becomes abstract therefore magnifies exclusion, foreclosing pathways before they begin. This matters beyond the individual learner, since foundation- al numeracy is the base on which later STEM learning and workforce participation rest. A system that loses its most vulnerable learners at this stage not only undermines equity but also constrains national capacity for innovation and inclusive growth.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 01Teachers deliver sign-supported lessons 02learners’ complete tablet-based assessments, 03caregivers reinforce learning at home with WhatsApp reminders and progress updates, and teachers adapt instruction using the returned data. THE KALIMANI OPERATIONAL CIRCLE 2.3 The innovation Kalimani is an AI-enabled signing tool that demonstrates numeracy instructions in Kenyan and Tanzanian Sign Language, enhancing access for deaf and hard- of-hearing learners. It operates through a continuous cycle: This loop runs two to three times per week depending on teacher capacity. Training in remediation, formative assessment, explicit instruction, and differentiation enabled teachers to integrate the app to support class- room instruction and adapt pedagogy for deaf learners, reinforced by ministry coaching. The intervention on Annex M model’s links three inputs, teacher training, app integration, and coaching directly for outcomes: inclusive pedagogy in classrooms, accessible sign- supported numeracy content, and reinforced practice. Together these are expected to upskill teachers in inclusive practice, improve learner engagement, strengthen numeracy skills, and build teacher confidence, generating credible evidence for feasibility and scale-up. Details are available in Annex L -Theory of Change.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT The Kalimani Project was implemented over an 8-week pilot period, following an initial three-day teacher training those prepared teachers to integrate the Kalimani App into their numeracy lessons. During the pilot, teachers received ongoing support through double-dosage coaching and feedback sessions, reinforced by two ministry-led coaching visits at Week 4 and Week 8. The evaluation followed a mixed-methods, non-randomised feasibility design. Its purpose was to test whether the model could be delivered as intended under real-world conditions, and to gather early evidence of promise for a visual, sign-supported AI tool with deaf and hard-of-hearing learners. Data were collected electronically through Kobo Toolbox across two strands. 3. PROJEC T & DESIGN IMPLEMENTATION 3.1 Roles of Delivery Partners 4. METHODOLOGY » Implementation lead Tanzania. » School selection. » Teacher professional development. » Infrastructure set-up in Tanzania. » Adaptation and content development in Tanzania. » Development of TDP exclusive guide booklet. » Supported with Baseline and endline data collection. » Directed implementation. » Coaching week 4 and 8 in Kenyan schools. » Monitoring & Evaluation design. » Developed and administered data instruments. » Baseline and endline data collection. » Analysis and reporting across both countries. » Intergration and Oversight. » Role in selection of schools by Ministry of Education. » Working through Curriculum developers in Tanzania. » Coaching and evaluation. » Generate evidence basis. GOVERNMENTS QUANTITATIVE STRAND QUALITATIVE STRAND » The Grade 2 and Grade 3 Assessment Standardised numeracy assessments, at baseline and endline. » Recorded Teachers Reflection through the weekly Teacher Deftere a structured notebook teachers use to record classroom practice and their experience of using the intervention. It acted as a window for more tailored feedback and coaching.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT QUANTITATIVE STRAND QUALITATIVE STRAND » The teacher classroom observation Measuring compliance with inclusive practices, at baseline and endline. » Weekly Coaching Coaching notes from the Week 4 and Week 8 visits. » The learner engagement observation Measuring participation and engagement, at baseline and endline » Monitoring In addition to Kobo Toolbox monitoring data, endline key informant interviews were conducted with 24 stakeholders, including 12 teachers, six head teachers, one caregiver, and five Ministry representatives. KEY INFORMANT REVIEWS 12Teachers 06Head Teachers 01Caregiver 05Ministry Representative Evaluation Questions Was the intervention acceptable, delivered as designed, on schedule, with the planned dosage, to the intended population, and with the planned tools functioning effectively? Are teacher practices changing under Kalimani-enabled TPD, particularly in confidence, attitudes, and use of sign-supported and adaptive instruction? Are deaf and hard-of-hearing learners’ numeracy performance and engagement levels improving under Kalimani-supported practices? How do implementation and outcomes compare between Kenya and Tanzania, and what does this comparison reveal about conditions for scale? What is the average cost of implementing the Kalimani App per teacher trained and per learner reached, in Kenya and Tanzania?

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Schools were purposively selected by Jenga Hub and The Action Foundation, together with their Ministries of Education, against four criteria: 4.1 Study Sample School type and learner concentration: a special school for deaf and hard hearing children, or a mainstream school with an established integrated unit, enrolling enough deaf and hard-of- hearing learners in Grades 2 and 3 to sustain the sample. Active numeracy instruction: Grade 2 and Grade 3 delivered as clearly identifiable classes through- out the eight-week period. Basic Classroom Readiness: at least one classroom with a functioning TV or projection screen and a device able to run the Kalimani App, which operates offline. Institutional and teacher commitment: written agreement from the head teacher and at least one willing numeracy teacher per grade to support training, implementation and ministry coaching Country Number of pupils at baseline Number of pupils at endline Change Kenya 163 173 +10 Tanzania 146 146 0 Both countries 309 319 +10 Table 1: Sample by country and Assessment point Country Number of pupils in grade 2 Number of pupils in grade 3 Total Kenya (baseline/ endline) 70 93 163 Tanzania (baseline/ endline) 53 93 146 Total 123 186 309 Table 2: Overall sample across countries per Arm Overall sample by country and assessment point

4.3 Limitations Several limitations frame how far the Kalimani findings extend. The pilot ran for 8 weeks appropriate for testing acceptability and fea- sibility, but too short to attribute causality and sustained adoption to the model. The teacher sample was small, 24 teachers across (16 teachers in the intervention and 8 in the control) across 12 schools; statistical power was therefore limited, though sufficient for a feasibility study of this scope, where findings rest on consistent patterns across schools rather than on signifi- cance testing. Also, other constraints the pilot encountered, device shortages, uneven connectivity, sign variation, and the app’s early stage of development, are not flaws in the model but conditions the feasibility design was intended to surface. They are specific and addressable, and identifying them is part of what the pilot was designed to achieve; their implications for scale are taken up in the lessons and next steps. In addition, classroom observations and engagement checklists were scored by a single observer per visit, with no inter-rater reliability checks. Observed changes may therefore reflect some observer variation as well as actual practice change. Lastly In Kenya, data collection overlapped with school break, at the endline which likely affected learner focus and outcomes by dampening engagement. As a feasibility study, the pilot was designed to surface early sig- nals and conditions for a future evaluation, not to compare arms. Two design limitations should be noted. The arms were unequal in size 109 control learners against 200 interventions. Also, learn- ers in both arms had mixed abilities and disabilities that were not mapped beforehand, so progress reflects starting profile and dis- ability as much as teaching. These point to a design requirement for the next phase: identify and match learners by ability and dis- ability profile from the start. KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT At baseline, 309 deaf and hard-of-hearing learners were assessed across the two countries; at endline, 319 were assessed, the increase reflecting learners who joined the participating classes over the implementation period. The end- line sample comprised 173 learners in Kenya and 146 in Tanzania, spanning Grade 2 and Grade 3 across intervention and control schools (see Table 1). This spread allowed disaggregation by grade, gender, and country. 4.2 Data sources, management, and analysis All five data sources were captured electronically through Kobo Toolbox, with daily synchronisation and routine supervisor checks during fieldwork to assure quality and completeness. Quantitative data were analysed descriptively, mean scores, competency-band distributions, and percentage-point change by arm, disaggregated by grade, gender, and country. Qualitative data from the Teacher Deftere, coaching notes, and key informant interviews were analysed thematically to triangulate and explain the quantitative patterns. The learner numeracy was scored against a four-level, criterion-referenced competency scale (Below, Approaching, Meeting, and Exceeding Expectation) from Kenya’s Competency-Based Assessment Framework, applied consistently across both countries; the full scoring rules and band thresholds are set out in Annex 8. Numeracy Competency Scale Below Approaching Meeting Exceeding expectations

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 4.4 Ethical considerations The pilot met ethical and legal standards in both countries. Kenya clearance was granted by JKUAT ISERC on 31 March 2026 (ref: JKU/ISERC/02316/2010) and Tanzania clearance by COSTECH. Government authorisation was secured through formal country-entry meetings with Ministry of Education officials before any school-level engagement. Written informed consent was obtained from parents, caregivers, head teachers, and teachers, and child assent was delivered in Kenya sign Language or Tanzania sign Language before each assessment, with any hesitant response treated as a decline. A do-no-harm protocol applied throughout, and enumerators were trained to stop immediately at any sign of learner distress. All data were managed under the Kenya Data Protection Act (2019) and the Tanzania Personal Data Protection Act (2022), with learners identified by code only and all findings reported in aggregated, anonymised form. 5. FINDINGS This section presents the pilot’s findings across its four research questions, drawing on all 5 instruments, the learner numeracy assessment, the classroom observation checklist, the learner engagement checklist, Key informant interviews and the Teacher Deftere. 5.1 Acceptability and feasibility Table 3: Acceptability and feasibility of the Kalimani model Category Intended Pilot evidence Outcomes Teacher professional development 3-day inclusive training,16 teachers, pre-implementation 16 teachers, 4 Head teachers and 4 school support officers trained across 8 schools; inclusive-practice knowledge +48.7pp; digital confidence 37% to 75%; MC knowledge 81.5%; pedagogy >90% Readiness reached App pillars & usage Avatar signs, differentiation, remediation, formative assessment 30-day users 187 to 396 (+112%); 384 sessions; 300+ attempts; remediation loop worked (70 reattempted, 10 to remediation) Adopted; loop functioning Dosage Learner dose: App-supported numeracy 3 Kalimani lessons/week; delivery dose: sessions as planned; support dose: 2 coaching visits (Week 4, Week 8) 384 sessions recorded; 30-day active users 187 to 396, sustained weekly use consistent with the ≥3-lessons/ week target; both coaching dosages delivered Dose largely delivered

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Fidelity of practice Teachers deliver the full learning cycle as designed Compliance rose sharply (TZ 42to 85 / 39 to 88; KE 57 to 74 / 56to 88); control flat; gains followed the first coaching visit Fidelity improved under coaching Government engagement MoE, TIE, CEMASTEA, CHAVITA for approval and coaching Engaged full cycle, design, fieldwork, QA, data entry; anchored the model Exceeded intent The Kalimani model was delivered broadly as designed and accepted across teachers, learners and government: training-built readiness, the App was adopted and its assessment-remediation loop functioned, practice improved under coaching, and institutions anchored the work with the constraints observed being conditions The pilot demonstrated strong implementation fidelity. The intervention was delivered on schedule and reached the intended learners across participating schools. Although implementation took place within dynamic school environments characterized by calendar disruptions, fluctuating attendance, and infrastructure constraints, these required only minor operational adaptations and did not alter the core design, delivery model, or intended participant reach. Evidence also suggests that the model was both feasible and acceptable. The three-day inclusive teacher professional development program reached 24 teachers and headteachers, all of whom completed the training and demonstrated readiness to apply the model’s core practices. Objective knowledge remained high at 81.5%, while understanding of inclusive pedagogy and formative assessment exceeded 90%. Self-reported confidence in using digital tools reached approximately 79%, suggesting that participants felt capable of implementing the approach in their classrooms. Acceptability also rested on strong institutional ownership in both countries, with the government engaged throughout. As one Ministry representative from the Directorate of Special Needs Education put it, A potential reason for this acceptance is the model’s fit with how deaf and hard-of-hearing learners access mathematics. Beyond how it is taught, the model changed what was physically in the classroom. It introduced a TV, a tablet, and the Kalimani tool into rooms that previously had none of these, turning a bare numeracy lesson into one with visual, signed content children could see and interact with. This finding is important because it points to the value of making abstract mathematical concepts visible and tangible, a recognized challenge in early numeracy instruction more broadly. While such approaches benefit all learners, they may be especially valuable for deaf and hard-of-hearing children, whose learning relies more heavily on visual representation and communication (Blatto-Vallee & Kelly, 2007). The new resources drew learners in engagement, and in several schools’ children from other grades joined the lessons uninvited, pulled by the maths on the screen. For learners long taught through spoken explanation they could not fully access, simply making numeracy visible and present in the room was itself a form of inclusion.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT “building on these potentials, I believe this intervention is ready for scale across inclusive schools and special units.” MINISTRY REPRESENTATIVE “ 42% - 85% 57% - 74% 39% - 88% 187 - 396 38456% - 88% Grade 2 Grade 2 Grade 3 30 - day active users SessionsGrade 3 Tanzania Kenya Tanzania Kenya The pilot also suggests feasibility in a specific form: a trained teacher deliver- ing numeracy through the Kalimani app and sign-supported demonstration, supported by regular coaching. Observed teaching compliance in intervention schools rose from 42% to 85% (Grade 2) and 39% to 88% (Grade 3) in Tanza- nia, and in Kenya compliance also rose 57% to 74% (Grade 2) and 56% to 88% (Grade 3). while control schools changed little. Use was sustained rather than limited to demonstration: 30-day active users rose from 187 to 396, generating 384 sessions in total. Feasibility held where training, coaching and gradual adoption came together.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 5.2 Teacher outcomes 5.2. 1 Measuring compliance with inclusive practice Table 4: Summary score of teacher compliances to inclusive classroom practices Teacher practice was a key implementation outcome and was measured through structured classroom observations using the Kalimani Classroom Observa- tion Checklist. The tool captures 30 indicators across five dimensions associated with effective numeracy instruction: accessibility of materials, application integration, learning differentiation, formative assessment, and data-informed teaching. Observations were conducted at baseline and endline among 16 intervention teachers across Grades 2 and 3 and eight control teachers. By assessing the extent to which targeted pedagogical practices were evident in classroom delivery, the observation data provide a direct measure of intervention uptake and the mechanisms through which program impacts on student learning may have been realized. As a feasibility study, the pilot tested whether intensified cycles of observation, coaching and feedback could accelerate improvements in teachers’ inclusive instructional practices. Intervention teachers were observed at four time points, baseline, following coaching cycles at Weeks 4 and 8, and endline, while control teachers were observed only at baseline and endline. The interim observations served a dual purpose: generating evidence on changes in classroom practice over time and informing structured audit-and-feedback discussions with teachers. This design enabled assessment of whether repeated feedback cycles were associated with measurable improvements in observed teaching practice and provided insight into the extent to which the intervention was implemented as intended. The approach is grounded in evidence showing that audit and feedback are most effective when performance is assessed against explicit standards, feedback is delivered repeatedly, and baseline performance leaves scope for improvement (Ivers et al., 2012). Coaching complements this mechanism by supporting teachers to translate feedback into concrete instructional changes, thereby strengthening the likelihood that improvements in teaching practice are adopted and sustained over time (Kraft, Blazar & Hogan, 2018). Metric Tanzania (Intervention) Kenya (Intervention) Tanzania (Control) Kenya (Control) Number of schools 4 4 2 2 Grade 2 Baseline compliance (%) 42% 57% 43% 45% Endline compliance (%) 85% 74% 45% 37%

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Grade 3 Baseline compliance (%) 39 56 40 48 Endline compliance (%) 88 88 42 40 (%) change in compliance 49 32 2 −8 Most improved school SCH-TZ-4-I (+64) SCH-KE-1-I (+56) Least improved school SCH-TZ-3-I (+33) SCH-KE-2-I (+10) Teacher codes were assigned by country, grade, and group. For example: SCH-TZ-1-I (+53), SCH-KE-2-I (+50), SCH-TZ-2-I (+30), SCH-KE-3-I (−13). “SCH” denotes school, “TZ” Tanzania, “KE” Kenya, the number indicates grade, and “I” marks intervention teachers. Control teachers were coded with “C.” (%) change in compliance +43% +17% 2 −8 Most improved school SCH-TZ-1-I (+53) SCH-KE-2-I (+50) — — Least improved school SCH-TZ-2-I (+30) SCH-KE-3-I (−13) — — Teacher compliance with inclusive best practices shows that intervention schools in both Tanzania and Kenya improved from baseline to endline, with Tanzania rising from 42% to 85% in Grade 2 and 39% to 88% in Grade 3, and Kenya from 57% to 74% and 56% to 88% respectively. The control schools moved only slightly (from 43% to 45% in Tanzania, 43% to 378% in Kenya). The most improved schools showed improvements of +53 and +64 in Tanzania and +50 and +56 in Kenya, while least improved schools still showed positive results though the changes were smaller. This gave an indication that the intervention positively affected practice shifts compared to the control arms. Teachers reinforced this through the key informant interviews, other teachers also spoke to their own growth. As one put it, “now I feel I am growing in the professional space… this training helped me understand how to integrate technology and differentiation into my teaching.” The teacher diaries echo this from the teachers’ side, describing the app used alongside “questioning, group work, practical materials and repeated practice,” with other teachers in Kenya noting, “now I teach in a more participatory way… I now teach with confidence.”

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 5.3 Learners Outcomes As a pilot feasibility study, the aim was to capture early evidence of promise. Numeracy outcomes improved across intervention and control schools in both countries. In Tanzania, intervention mean scores increased from 16.13 to 26.43 in Grade 2 and from 5.56 to 23.93 in Grade 3. In Kenya, intervention mean scores increased from 12.80 to 18.59 in Grade 2 and from 9.74 to 12.33 in Grade 3. Progress was also observed among learners in control schools, where Grade 2 scores increased from 15.86 to 26.38 and Grade 3 scores from 5.20 to 24.00 in Tanzania, while in Kenya Grade 2 scores increased from 9.00 to 13.96 and Grade 3 scores from 0.94 to 4.74. These findings suggest that numeracy outcomes improved not only under the intervention but also under ordinary teaching conditions during the study period. While these patterns indicate encouraging learning progress, improvements across both study arms suggest that additional contextual, school-level, or instructional factors not examined within the scope of this feasibility study may also have contributed to learner out- comes. These findings suggest that the intervention was feasible to implement and associated with positive learning trajectories; however, stronger evidence would be required to establish its additional impact on learner outcome Baseline Tanzania Kenya Grade 2 - Mean Numeracy Score Endline Baseline Control Intervention Endline 15.86 16.13 26.38 13.96 9.00 26.48 18.59 12.80

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 5.56 24.00 0.94 23.93 9.74 5.20 Control Intervention Grade 3- Mean Numeracy Score Baseline Endline Baseline Endline 5.3.1 Numeracy skill type A further angle of the pilot examined learner performance by skill type, fluency and problem-solving. Fluency, the foundational, pro- cedural side of numeracy was defined as the quick, accurate recall of basic number skills, and in this study covered number identification, discrimination, missing number and place value. Problem-solving, the applied, reasoning-based side covered ad- dition, subtraction, measurement, shapes, time and money. This distinction is important because problem-solving is often the point at which early-grade numeracy becomes most challenging. Unlike fluency tasks, which rely primarily on the recall and application of learned procedures, problem-solving requires learners to reason, interpret information, and apply mathematical concepts in context. These demands present difficulties for many early-grade learners and may be particularly challenging for deaf and hard-of-hearing learners when instruction is delivered primarily through spoken language (Santos & Cordes, 2021; Pagliaro & Kritzer, 2013). While Figure 2 highlights broad improvements in fluency and problem-solving, examination of the individual numeracy compo- nents provides a more nuanced picture of learner progress. Across both countries, the largest gains often occurred in domains where baseline performance was extremely low, suggesting movement beyond simple number recognition towards more complex mathematical concepts and reasoning. Tanzania Kenya 4.74 12.33

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT But taking a deeper look, the more interesting gains were in the specific areas in (Annex H and I). In Tanzania, intervention learners demonstrated particu- larly strong progress in foundational concepts that underpin later mathematical understanding. Among Grade 2 learners, place value increased from 33% to 71%, missing number from 39% to 75%, and subtraction from 31% to 67%. Among Grade 3 learners, substantial gains were observed across several ap- plied domains, including shapes and geometry (2% to 60%), money (1% to 44%), measurement (4% to 43%) and time (0% to 31%). These findings indicate that learners were able to develop skills requiring interpretation, conceptual understanding and application, rather than solely procedural recall. Teachers described this shift in their own terms, linking the visual signed delivery to better grasp of the harder concepts. In Tanzania, teachers connected the app to “better recognition of place value and reduced fear of mathematics,” and a Kenyan teacher observed learners “working more confidently... showing increased understanding and independence” in the later weeks of the pilot A similar pattern was evident in Kenya; Kenya showed a different pattern. Although overall gains were smaller than those observed in Tanzania, intervention learners nevertheless demonstrated notable improvements in several areas. In Grade 2, performance increased in missing number (45% to 69%), addition level 2 (6% to 32%), and subtraction (20% to 45%). Grade 3 learners also improved in shapes and geometry (22% to 33%), money (19% to 27%), and number operations (34% to 42%). These gains indicate progress in learners’ ability to work with increasingly complex mathematical concepts. However, improvements were uneven across domains, and several skills remained persistently weak. In particular, place value showed little change in either Grade 2 (18% to 19%) or Grade 3 (10% to 15%), while time remained extremely low among Grade 3 learners, declining slightly from 18% to 16%. Therefore, findings suggest that while learners were able to strengthen foundational number knowledge during the pilot, the development of conceptual and applied mathematical skills remained more challenging. The continued gap between fluency and problem-solving scores at endline, combined with per- sistently low performance in areas such as place value, time, money and measurement, indicates that mathematical reasoning and application remain priority areas for future intervention. This pattern is consistent with evidence that deaf and hard-of-hearing learners often face greater challenges in mathematical tasks requiring interpretation, abstraction and language-mediated reasoning (Pagliaro & Kritzer, 2013; Santos & Cordes, 2021). “better recognition of place value and reduced fear of mathematics,” KENYAN TEACHER “working more confidently... showing increased understanding and independence” KENYAN TEACHER

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 5.3.2 Learners Assessment from baseline and Endline Table 5: CBE Band Distribution, Baseline to Endline by Group, Grade and Country Group N BE Below AE Approaching ME Meeting EE Exceeding Base End Base End Base End Base End Grade 2 Tanzania Intervention 31 23% 0% 63% 40% 10% 10% 3% 50% Tanzania Intervention 31 23% 0% 63% 40% 10% 10% 3% 50% Tanzania Control 22 24% 5% 62% 19% 14% 38% 0% 38% Kenya Intervention 46 43% 15% 50% 59% 4% 20% 2% 7% Kenya Control 24 75% 46% 25% 42% 0% 8% 0% 4% Grade 3 Tanzania Intervention 65 90% 3% 10% 71% 0% 24% 0% 2% Tanzania Control 28 86% 0% 14% 72% 0% 21% 0% 50% Kenya Intervention 58 64% 52% 34% 40% 2% 9% 0% 0% Kenya Control 35 100% 86% 0% 14% 0% 0% 0% 0% Gender TZ +KE Grade2 Intervention Girls 35 37% 14% 51% 51% 9% 14% 3% 20% Intervention Boys 42 34% 5% 59% 51% 5% 17% 2% 27%

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Control Girls 23 52% 35% 39% 26% 9% 30% 0% 9% Control Boy 23 50% 18% 45% 36% 5% 14% 0% 32% Grade 3 Intervention Girls 55 80% 26% 20% 50% 0% 24% 0% 2% Intervention Boys 68 75% 29% 24% 60% 2% 10% 0% 2% Control Girls 26 100% 50% 0% 37% 0% 10% 0% 3% Control Boys 37 89% 43% 11% 46% 0% 9% 0% 3% School types Special unit (4 sch) I 118 47% 7% 46% 53% 6% 29% 1% 11% Inclusive (4 sch) I 81 75% 34% 24% 55% 0% 3% 1% 8% Special unit(2sch) C 50 79% 44% 21% 38% 0% 10% 0% 8% Inclusive (2sch) C 60 74% 34% 21% 37% 5% 18% 0% 11% The table shows the percentage of learners in each CBC band at baseline and endline. Both intervention and control arms moved in the same direction, so change is reported as observed movement across groups. In terms of school types Special units are specialized classrooms for deaf and hard-of-hear- ing learners located within mainstream schools, while inclusive schools integrate deaf and hard-of-hearing learners into regular classrooms alongside their hearing peers with appropriate support.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Grade level appears to shape progress more strongly than treatment status The clearest pattern is that Grade 2 learners were more likely to transition into the higher competency bands than Grade 3 learners across both intervention and control groups. Grade 2 learners frequently reached the “Meeting” and “Exceeding Expectation” categories, whereas Grade 3 learners remained concentrated in the “Approaching Expectation” band. This suggests that foundational numeracy gaps may be easier to address earlier in the learning trajectory, before misconceptions and learning deficits become entrenched. School type appears to matter less than expected Although inclusive schools start- ed from a weaker baseline than special units, both school types showed movement towards higher competency bands by endline. This suggests that school placement alone may not be the primary driver of learner progress. Rather, improvements appear possible in both settings when learners receive consistent instructional support, indicating that instructional quality may be a more important determinant than school type. Gender differences were small compared to overall learning progress Both girls and boys demonstrated upward movement across competency bands, with no consistent evidence that one gender benefited substantially more than the other. The similarity of learning trajectories suggests that gender was not a major differentiating factor in learner progress, and that barriers to numeracy achievement among deaf and hard-of-hearing learners may be more strongly linked to instructional access than to gender. Progress was concentrated in movement towards competency rather than high achievement Across almost all groups, movement was strongest from “Below Expectation” into “Approaching” and “Meeting Expectations”, while movement into “Exceeding Expectations” remained relatively limited. This suggests that the primary challenge for these learners is achieving foundational competency. The findings therefore point to a system that is becoming more inclusive and effective at supporting basic numeracy, but where sustained support may be required before large numbers of learners reach the highest performance levels.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 5.4 Learners’ engagement and participation level Table 6: Learners Engagement and participation levels Alongside attainment, the pilot set out to measure the classroom engagement and participation of deaf and hard-of-hearing learners during lessons. The Learner Engagement Observation Checklist captured this through direct observation, with observers recording only what they saw. This section reports the three domains observed across all classrooms: engagement, participation, and responsiveness to feedback. The table shows the % of learners displaying each behaviour at baseline and endline. Engagement rose across all groups Group A. Engagement B. Participation C. Responsiveness to feedback Base End Base End Base End Grade 2 Tanzania Base End Base End Base End Tanzania Intervention 71% 90% 75% 83% 72% 84% Tanzania Control 46% 54% 52% 57% 49% 59% Kenya Intervention 72% 82% 67% 67% 79% 80% Kenya Control 39% 60% 53% 54% 58% 68% Grade 3 Tanzania Intervention 75% 86 70% 85% 64% 87% Tanzania Control 36% 54 33% 40% 53% 89% Kenya Intervention 65% 89 61% 61% 61% 74% Kenya Control 48% 50 60% 46 56% 45%

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Learner engagement, participation and responsiveness improved across most intervention classrooms, although the pattern varied by country and grade level (Table 7).In Tanzania, improvements were observed across all three dimensions, suggesting that learners became not only more attentive but also increasingly willing to participate and respond to teacher feedback. In Kenya, engagement and responsiveness improved consistently, although gains in active participation were more limited. Taken together, these findings suggest that learners became more involved in classroom activities over the course of the pilot, with the strongest improvements observed where instructional practices and learner support mechanisms appeared most established. Qualitative evidence helps explain how these shifts manifested in practice. Teachers frequently described greater learner interaction and peer-supported learning during numeracy activities. One Kenyan teacher noted that learners “independently helped each other complete a numeracy activity, discussed the answers using sign language, corrected one another, and completed the task successfully.” Similar observations emerged in Tanzania, where a teacher reported that “now they see, they touch, and retention span has greatly in- creased.” In Kenya, teachers also highlighted the social dimension of learning, reporting that learners “shared tablets and learned together...this improved how they think about Mathematics problem, this helped slow learners catch up.” These accounts suggest that engagement was not limited to attention or task completion but was accompanied by increased collaboration, confidence, and peer-to-peer support. The pattern observed is consistent with the instructional approach underpinning the intervention. The combination of visual demonstrations, sign-supported instruction, and digital learning materials provided deaf and hard-of-hearing learners with multiple entry points to mathematical concepts. Rather than relying primarily on spoken explanations, learners were able to interact with content through visual and signed modalities, an approach that research suggests is more accessible for early grade and deaf learners (Blatto-Vallee & Kelly, 2007). Ministry representatives conducting independent classroom visits similarly noted greater learner interaction and collaboration during lessons. While several control schools also recorded improvements in engagement-related measures, progress was generally less consistent across the three domains Tanzania Grade 2 control improved modestly (46% to 54% engagement; participation 52% to 57%), while Tanzania Grade 3 control rose from 36% to 54% engagement and responsiveness from 53% to 89%. In contrast, Kenya Grade 3 control weakened slightly (48% to 50% engagement; responsiveness fell to 45%) Although the pilot cannot attribute subsequent learning gains directly to these changes, the findings suggest that the Kalimani model helped create classroom conditions associated with active participation, collaborative learning, and sustained engagement in numeracy activities. 5.5 Unintended outcomes The pilot generated a spillover effect beyond its design. In several intervention schools, learners outside Grades 2 and 3 began joining Kalimani lessons unin- vited, drawn by the numeracy content delivered through television and tablets. As one teacher noted, “learners that were not Grade 2 and 3 started to rush in for the Kalimani maths classes” (TR4-TZ).

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 5.6 Cost effectiveness analysis for Kalimani Cost element Total USD Per school \ USD Per learner \ USD Recurrent, one 4-month cycle Coaching and feedback sessions 5,015.00 626.88 25.08 School monitoring visits 1,975.70 246.96 9.88 Local travel and logistics 1,840.36 230.05 9.20 Communication and airtime 1,082.81 135.35 5.41 Recurrent total 9,913.87 1,239.24 49.57 One-time, per new school Classroom equipment (TV, tablets) 8,470.31 1,058.79 42.35 Teacher training (2 teachers) 10,498.44 583.25 23.33 One-time total 18,968.75 1,642.04 65.68 One-time, national Country-specific adaptation 37,108.54 — — Shared across both countries 9,961.33 — — One-time total 47,069.87 — — Table 7 - Cost effectiveness analysis Cost analysis suggests that the Kalimani model is relatively affordable to operate once established, with most expenditure concentrated in upfront investments rather than ongoing delivery. Recurrent implementation costs were USD 9,913.87 per four-month cycle, equivalent to USD 49.57 per learner, covering coaching, monitoring, travel, and communication. Coaching and feedback sessions represented the largest recurrent cost component, reflecting the model’s emphasis on strengthening classroom practice through sustained support rather than one-off training. One-time school-level investments total led USD 18,968.75 (USD 65.68 per learner) and were primarily associated with teacher training and classroom equipment.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT While these start-up costs are substantial, they are not expected to recur at the same level once schools are established within the program. This cost profile has important implications for scale: as more schools and learners are reached, fixed investment costs are likely to be distributed across a larger beneficiary base, reducing average costs over time. The findings therefore suggest that Kalimani’s long-term affordability will depend less on technology investments and more on the ability of ministries and implementing partners to institutionalized coaching and support structures within existing education systems. 5.7 Discussion of the findings; An Implementation Science Perspective Building on the evidence presented, the next section distils the lessons learned from the pilot. It moves beyond reporting outcomes to examine what worked and what did not, the barriers and enablers encountered, and the adaptive strategies that emerged in practice. By triangulating insights from teachers, head teachers, and ministry representatives, the discussion highlights transferable lessons that can inform next steps across contexts. The strongest finding from this feasibility study relates to implementation rather than effectiveness. Evidence from classroom observations, coaching re- cords, teacher reflections, and platform analytics suggests that the Kalimani model was successfully integrated into routine classroom practice across both countries. This is important because educational innovations frequently fail during implementation rather than design. The convergence of evidence across multiple data sources suggests that teachers were not only exposed to the intervention but were able to adopt and sustain its core practices within everyday teaching. From an implementation science perspective, the pilot therefore provides evidence of implementation fidelity and adoption, two prerequisites for any future assessment of effectiveness (Proctor et al., 2011; Durlak & DuPre, 2008). » Implementation fidelity and adoption: The pilot reinforces a growing body of evidence that professional development alone rarely produces sustained changes in classroom practice. Rather, it is the combination of training, coaching, feedback, and opportunities for reflection that enables teachers to translate new knowledge into practice. One of the most important lessons from the pilot is that improvements in teacher practice were observed across contexts with very different levels of readiness. Tanzania entered the pilot with a more established signing environment, while Kenya was still developing aspects of sign-supported provision. Yet both settings demonstrated evidence of strengthened instructional practice. This suggests that implementation success may depend less on starting conditions and more on the availability of sustained professional support. Consistent with broader evidence, the combination of training, coaching, feedback, and structured reflection appears to have enabled teachers to translate new knowledge into classroom practice (Kraft, Blazar, & Hogan, 2018; UNICEF, 2020). From a scale-up perspective, these finding shifts attention away from context as a barrier and towards coaching as a critical enabling condition. Institutionalizing coaching within existing ministry support structures may therefore be more important than expanding training alone, particularly in systems where inclusive education capacity is still emerging. » Coaching may be the key mechanism enabling change across contexts:

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT A further finding relates to learner engagement and participation, which are widely recognized as important precursors to learning and achievement (Fredricks, Blumenfeld & Paris, 2004). While the pilot was not able to demonstrate a measurable intervention effect on numeracy outcomes, it did show more consistent improvements in engagement, participation, and responsiveness within intervention classrooms. This is noteworthy because these dimensions reflect the conditions through which learning occurs rather than learning outcomes themselves. Across both countries, teachers reported greater collaboration, sustained attention, and peer-to-peer support, particularly during activities that combined visual demonstrations, sign-supported instruction, and shared use of digital learning materials. The pattern suggests that the intervention may have strengthened learners’ access to and participation in classroom learning, even where changes in attainment were not yet detectable. From a feasibility perspective, these findings provide an early signal that the model is influencing important intermediate outcomes that merit further testing in a larger and longer-term effectiveness study. » Engagement provided the strongest early signal of change: The findings suggest that the model’s most distinctive contribution lies in improving access to instruction rather than directly improving learning outcomes within the pilot period. Through visualization, demonstration, and sign-supported instruction, the model provides alternative routes through which deaf and hard-of-hearing learners can access mathematical concepts. This interpretation is important given that learning gains were observed in both intervention and comparison schools, making it difficult to attribute improvements in attainment directly to the intervention. What appears more plausible is that the model strengthened the conditions that support learning by increasing accessibility, participation, and engagement. This interpretation aligns with evidence that deaf learners benefit from visual-spatial and sign-mediated approaches that reduce barriers created by language-dependent instruction (Blatto-Vallee & Kelly, 2007; Marschark & Hauser, 2012). » Instructional access as the likely mechanism of change: A key implementation finding is that similar patterns of uptake emerged across two systems with different levels of readiness. Although Tanzania had a more matured and established signing environment and Kenya required greater contextualization and sign harmonization, teachers in both settings were able to integrate the core components of the model into classroom practice. This suggests that the intervention’s core principles may be transferable across contexts. However, the pilot also demonstrates that successful transferability depends on adaptation. Sign standardization, curriculum, and contextual co- design emerged as implementation requirements rather than optional alignment enhancements. The implication for scale is that adaptation should be considered a core implementation strategy (Fixsen et al., 2005; Century & Cassata, 2016). » Context matters, but the model appears transferable: The pilot provides encouraging evidence that Kalimani is feasible, acceptable, and implementable under routine school conditions. However, it does not yet provide evidence that the intervention improves learning outcomes beyond those achieved through ordinary teaching conditions. Similar improvements in learner performance across intervention and comparison schools suggest that wider contextual factors, maturation, or regular teaching may also have con- tributed to observed gains. Consequently, the most defensible conclusion from this study is that Kalimani has demonstrated proof of implementation and ability transfer research evidence to practices. The next step should therefore be a larger and longer-term effectiveness study capable of isolating program effects while continuing to examine implementation quality. Such a study would establish whether improvements in instructional access, participation, and teacher practice translate into measurable learning gains over time (Proctor et al., 2011). » From proof of implementation to proof of impact:

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 7. Lesson learned A further formative finding concerns learning variability and sample mapping. Learner numeracy skills improved in both arms, consistent with ordinary pro- gression over a school term, but teacher practice improved only where the model was present. This asymmetry is the instructive result: the intervention’s clearer contribution lies in strengthening instruction, not in a learner-outcome effect the design can attribute. That distinction matters because these were mixed-ability learners, some with multiple disabilities, whose progression is heterogeneous and cannot be read cleanly against a single test score (NJCLD, 1998). Two features of the sample compound this: the control arm (109 learners) was roughly half the intervention arm (200), and in both arms ability and disability profiles were mixed rather than mapped so a learner’s movement between bands reflects starting profile and disability as much as instruction. The implica- tion is clear: in disability-inclusive evaluation, deliberate sample mapping and arm-matching are not refinements but preconditions for interpreting outcomes, and a next phase should build both in from the first assessment.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 109 200Learners - the control arm intervention arm The pilot showed that infrastructure shapes how well digital pedagogy holds. Where devices, power, and connectivity were reliable, Kalimani practices were sustained; where they were weak, app-dependent practices declined by endline. A workable device ratio, reliable electricity, and stable connectivity are therefore enabling conditions for inclusive, AI-supported pedagogy; without them, teacher training struggles to translate into consistent practice. Long-term sustainability will depend on resourcing infrastructure alongside professional development, embedding devices, connectivity, and power within education systems so the model can hold across diverse school types beyond the project cycle. Fix and co-design the product with teachers to bypass sign variations; The pilot named specific, fixable problems with the app itself: content and assessment pitched above the learners’ grade level, and sign variation that did not match the signs teachers use because teachers were not part of the design. The next step is to align the app content and in-app assessment to each grade, and to standardize the signed content by co-designing it with teachers of deaf and hard hearing children. These are product corrections within reach. 8. Sustainability and Next Steps Kalimani’s sustainability was built in from the start: ministries, CEMASTEA, and TIE co-designed, trained, observed, and coached, ensuring that the very systems meant to own the model have already shaped it. Endorsements are unanimous to move beyond the pilot, grounding expansion on formative lessons. Respondents tied durability to ministry-led coaching, verified delivery through platform records, and integration into curriculum and professional- development. Looking ahead, three levers anchor sustainability: commission a larger, baseline-matched evaluation with longer exposure and continuous follow-up; concentrate delivery on the teacher-facing widow where effects are strongest, paired with sustained coaching; and formalise government partnership with matched funding. The application remains at an early stage of development, and edtech in inclusive education is still a growing field. Sustainability will therefore depend on stakeholders strengthening the tool and scaling it together ministries, developers, and teachers of the deaf co-refining the content, signs, and design as the model expands. Scaling should follow the app’s maturity, beginning in special units and the strongest inclusive schools, and widening as the tool and the evidence base grow 01 03 02

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT 9. Annexes Annex A. Intervention Framework

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT PROBLEM Deaf and hard-of-hearing learners in Grade 2 and 3 attend school but show low attainment in foundational numeracy. Mathematics is taught mainly through oral instruction, which these learners cannot access. Teachers lack sign-supported tools and training to make mathematics visible, instruction is rarely adapted for learners who learn through sight, and professional development seldom carries into daily classroom practice. INPUTS Government and education partners. Kalimani App + KSL/TSL resources. Teachers, parents and learners. Pilot/control schools. TPD, data and MEL resources. ACTIVITIES Engage government and education offices Contextualised for teachers of deaf and hard-of-hearing learners, and its implementing and government partners Teacher professional development, on-site coaching and feedback sup- port, and participating government primary schools Monitor, assess and learn Disseminate findings and plan scale OUTPUTS Approvals and coordination established. Contextualized Kalimani App and teaching resources. 16 teachers trained Baseline/endline data. 8–12 weeks implementation. Formative assessments + remediation. Evidence, policy briefs and scale-up outputs. OUTCOMES Improved teacher and parent capacity. Increased use of sign-supported numeracy. Greater learner participation. Early improvements in numeracy. Evidence on feasibility, acceptability and effectiveness. Findings inform FLN priorities. IMPACT More inclusive education systems where all children, including deaf and hard-of-hearing learners, have the opportunity to develop strong foundational numeracy skills. Annex M: Theory of change Figure 6: theory of change

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Annex C. Data collection instruments Annex D: Learners’ outcomes Scoring rules Grade 2 Numeracy Assessment (marked out of 35 and grade 3 on 40), baseline and endline, all six schools. Classroom Teacher Observation Checklist (30 indicators, five domains), one observed lesson per teacher, baseline and endline, all six schools. Learner Engagement Observation Checklist, baseline and endline, all six schools. Teacher reflective diaries, kept at baseline, Week 4, and Week 8, intervention schools. Endline interviews with teachers in intervention schools; interviews and government stakeholders. 01 02 03 04 05 » Each item = 1 point (correct) or 0 (incorrect / no response), unless otherwise stated. » Accept self-corrections if the child corrects before you move to the next item. Write SC in notes. Award 1 point. CBC Level % (Raw score) Grade A to E Description EE (Exceeds) 28-30/30 A, Outstanding. Mastery of all Grade 2 competencies ME (Meets) 23 to 27to 30/ B Satisfactory. Meets most Grade 2 ex- pectations Within expectations 18 to 22/ 30 C Within expectation AE (Approaching) 12 to 17/ 30 D Needs support. Partial understanding BE (Below) 0 to11/ 30 E Intensive support needed. Significant gaps Table 8 Annex C: competency-based curriculum scoring rules

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Table 9: Annex D: Learner Numeracy Mean Score by School School Group Grade N Baseline % Endline % Change Tanzania SCH-TZ-1I Inte 2 13 54% 75% +21 SCH-TZ-2I Inte 2 5 58% 75% +17 SCH-TZ-3I Inte 2 4 20% 69% +49 SCH-TZ-4I Inte 2 8 39% 79% +40 SCH-TZ-5C Cont 2 12 60% 81% +21 SCH-TZ-6C Cont 2 9 25% 68% +43 Kenya SCH-KE-1I Inte 2 12 56% 65% +9 SCH-KE-2I Inte 2 5 47% 80% +33 SCH-KE-3I Inte 2 14 21% 48% +27 SCH-KE-4I Inte 2 15 32% 40% +8 SCH-KE-5C Cont 2 17 28% 46% +18 SCH-KE-6C Cont 2 7 21% 25% +4 SCH-TZ-1I Inte 3 35 14% 65% +51 SCH-TZ-2I Inte 3 2 4% 71% +67 SCH-TZ-3I Inte 3 7 4% 49% +45 SCH-TZ-4I Inte 3 15 20% 52% +32 SCH-TZ-5C Cont 3 16 3% 59% +56 SCH-TZ-6C Cont 3 13 26% 61% +35 SCH-KE-1I Inte 3 5 44% 30% +-14 SCH-KE-2I Inte 3 21 37% 46% +9

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT SCH-KE-3I Inte 3 19 6% 25% +19 SCH-KE-4I Inte 3 13 24% 15% +-9 SCH-KE-5C Cont 3 17 2% 9% +7 SCH-KE-6C Cont 3 18 2% 15% +13 Conte= Control, Inte = Intervention. Bands: BE Below, AE Approaching, ME Meeting, EE Exceeding. Each cell shows % of the school’s learners in that band at baseline (B) and endline (E). Across intervention and control, grade 2 and 3 School Group BE B BE E AE B AE E ME B ME E EE B EE E SCH-TZ-1I Inte 31% 0% 38% 38% 23% 15% 8% 46% SCH-TZ-2I Inte 0% 0% 100% 40% 0% 20% 0% 40% SCH-TZ-3I Inte 50% 0% 50% 50% 0% 0% 0% 50% Table 10 Appendix E: Learner Numeracy CBC Band Distribution by School: Grade 2 SCH-TZ-4I Inte 12% 0% 88% 38% 0% 0% 0% 62% SCH-TZ-5C Cont 0% 0% 75% 8% 25% 50% 0% 42% SCH-TZ-6C Cont 56% 11% 44% 33% 0% 22% 0% 33% SCH-KE-1I Inte 0% 0% 83% 58% 17% 25% 0% 17% SCH-KE-2I Inte 20% 0% 80% 0% 0% 100% 0% 0% SCH-KE-3I Inte 93% 21% 7% 64% 0% 7% 0% 7% SCH-KE-4I Inte 40% 27% 53% 73% 0% 0% 7% 0% SCH-KE-5C Cont 76% 29% 24% 53% 0% 12% 0% 6% SCH-KE-6C Cont 71% 86% 29% 14% 0% 0% 0% 0% Bands: BE Below, AE Approaching, ME Meeting, EE Exceeding. Each cell shows % of the school’s learners in that band at baseline (B) and endline (E).

SCH-KE-2I Inte 33% 19% 67% 71% 0% 10% 0% 0% SCH-KE-3I Inte 100% 58% 0% 32% 0% 11% 0% 0% SCH-KE-4I Inte 69% 92% 31% 8% 0% 0% 0% 0% SCH-KE-5C Cont 100% 94% 0% 6% 0% 0% 0% 0% SCH-KE-6C Cont 100% 78% 0% 22% 0% 0% 0% 0% Table 12 Annex G: Numeracy Components Table 10 Appendix E: Learner Numeracy CBC Band Distribution by School: Grade 2 KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT School Group BE B BE E AE B AE E ME B ME E EE B EE E SCH-TZ-1I Inte 86% 0% 14% 63% 0% 34% 0% 3% SCH-TZ-2I Inte 100% 0% 0% 0% 0% 100% 0% 0% SCH-TZ-3I Inte 100% 14% 0% 86% 0% 0% 0% 0% SCH-TZ-4I Inte 93% 7% 7% 93% 0% 0% 0% 0% SCH-TZ-5C Cont 100% 0% 0% 75% 0% 19% 0% 6% SCH-TZ-6C Cont 69% 0% 31% 71% 0% 21% 0% 7% SCH-KE-1I Inte 40% 60% 40% 20% 20% 20% 0% 0% Country Tanzania Numeracy subtask Intervention Control Base End Chg Base End Chg Grade 2 (score /35) Number Identification 81% 95% +14 78% 97% +19

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Number Discrimination 65% 87% +22 55% 98% +43 Missing Number 39% 75% +36 61% 77% +16 Addition (Level 1) 51% 75% +24 54% 79% +25 Addition (Level 2) 23% 57% +34 14% 61% +47 Subtraction 31% 67% +36 41% 69% +28 Place Value 33% 71% +38 13% 47% +34 Overall mean score 16.1 26.4 15.9 26.4 Grade 3 (score /40) Number Operations 24% 78% +54 27% 70% +43 Measurement 4% 43% +39 0% 49% +49 Shapes & Geometry 2% 60% +58 0% 45% +45 Time 0% 31% +31 0% 46% +46 Money 1% 44% +43 0% 59% +59 Place Value 32% 66% +34 28% 72% +44 Overall mean score 5.6 23.9 5.4 24.0 Average % correct per subtask. Baseline and Endline. Table 13: Annex H: Numeracy Components Country Kenya Numeracy subtask Intervention Control Base End Chg Base End Chg

Number Operations 34% 42% +8 5% 25% +20 Measurement 23% 28% +5 0% 3% +3 Shapes & Geometry 22% 33% +11 2% 3% +1 Time 18% 16% −2 2% 2% +0 Money 19% 27% +8 1% 11% +10 Place Value 10% 15% +5 0% 1% +1 Overall mean score 9.7 12.3 0.9 4.7 Table 14 Annex I: Numeracy skills fluency and Problem-Solving Kenya KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Grade 2 (score /35) (Intervention N=46, Control N=24) Number Identification 70% 90% +20 63% 76% +13 Number Discrimination 56% 63% +7 48% 72% +24 Missing Number 45% 69% +24 28% 39% +11 Addition (Level 1) 42% 54% +12 20% 38% +18 Addition (Level 2) 6% 32% +26 6% 22% +16 Subtraction 20% 45% +25 12% 29% +17 Place Value 18% 19% +1 4% 3% −1 Overall mean score 12.8 18.6 9.0 14.0 Grade Skill Int. Base Int. End Int. Chg Ctrl Base Ctrl End Ctrl Chg Tanzania Grade 2 Fluency 54% 82% +28 52% 80% +28

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Grade 2 Problem-solving 35% 66% +31 36% 70% +34 Grade 3 Fluency 28% 72% +44 28% 71% +43 Grade 3 Problem-solving 2% 44% +42 0% 50% +50 Kenya Grade 2 Fluency 47% 60% +13 36% 48% +12 Grade 2 Problem-solving 23% 44% +21 13% 30% +17 Grade 3 Fluency 22% 28% +6 2% 13% +11 Grade 3 Problem-solving 20% 26% +6 1% 5% +4 Note. Average % correct, baseline to endline, by country, grade and study arm (Intervention N: TZ G2=30, G3=59; KE G2=46, G3=58. Control N: TZ G2=21, G3=30; KE G2=24, G3=35). Fluency = number identification, discrimination, missing number and place value (Grade 2); number operations and place value (Grade 3). Problem-solving = addition and subtraction (Grade 2); measurement, shapes, time and money (Grade 3 Engagement and participation levels Percentage of learners displaying each engagement behaviour, baseline (B) and endline (E), by school. Five domains: A Engagement, B Participation, C Visual attention and sign engagement, D Digital skills, E Responsiveness to feedback. Digital skills (D) began at zero in every school, as the component did not exist before the intervention.

Kenya SCH-KE-1-I Intervention 61% 100% 60% 100% 10% 76% 0% 78% 72% 56% SCH-KE-2-I Intervention 100% 100% 100% 73% 15% 100% 0% 100% 100% 100% SCH-KE-3-I Intervention 58% 69% 60% 49% 10% 65% 0% 14% 61% 45% SCH-KE-4-I Intervention 67% 58v 47% 45% 25% 50% 0% 46% 83% 42% SCH-KE-5-C Control 67% 75% 35% 75% 0% 0% 0% 0% 33% 61% SCH-KE-6-C Control 44% 67% 70% 33% 0% 0% 0% 0% 83% 75% KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Table 15 annex J: Learner Engagement School-Level Grade 2 School Group A Engagement B Participation C Visual/Sign% D Digital E Responsiveness Base End Base End Base End Base End Base End Tanzania SCH-TZ-1-I Intervention 67% 83% 60% 80% 10% 86% 0% 100% 33% 67% SCH-TZ-2-I Intervention 100% 100% 70% 100% 10% 57% 0% 83% 70% 100% SCH-TZ-3-I Intervention 67% 83% 100% 60% 15% 100% 0% 10% 100% 100% SCH-TZ-4-I Intervention 50% 33% 70% 40% 25% 71% 0% 67% 75% 0% SCH-TZ-5-C Control 33% 40% 33% 40% 0% 0% 0% 0% 33% 33% SCH-TZ-6-C Control 38% 40% 0% 40% 0% 0% 0% 0% 45% 40%

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Grade 3 School Group A Engagement B Participation C Visual/Sign% D Digital E Responsiveness Base End Base End Base End Base End Base End Tanzania SCH-TZ-1-I Intervention 64% 83% 54% 80% 5% 71% 0% 100% 55% 100% SCH-TZ-2-I Intervention 54% 83% 55% 80% 10% 86% 0% 83% 25% 100% SCH-TZ-3-I Intervention 83% 100% 70% 100% 25% 100% 0% 63% 75% 83% SCH-TZ-4-I Intervention 100% 100% 100% 80% 25% 82% 0% 63% 100% 83% SCH-TZ-5-C Control 33% 43% 33% 50% 0% 0% 0% 0% 33% 10% SCH-TZ-6-C Control 34% 49% 33% 40% 0% 0% 0% 0% 75% 19% Kenya SCH-KE-1-I Intervention 100% 100% 100% 100% % 76% 0% 100% 100% 100% SCH-KE-2-I Intervention 54% 100% 45% 65% 25% 100% 0% 88% 45% 83% SCH-KE-3-I Intervention 67% 78% 60% 57% 25% 60% 0% 17% 60% 58% SCH-KE-4-I Intervention 39% 78% 40% 80% 10% 65% 0% 33% 85% 56% SCH-KE-5-C Control 50% 40% 55% 56% 0% 0% 0% 0% 45% 57% SCH-KE-6-C Control 46% 40% 35% 36% 0% 0% 0% o% 35% 33%

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Table 15 annex J: Learner Engagement School-Level

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Cross country comparison Table 16 Annex K: Comparison Analysis Index What the evidence shows Tanzania (established platform) and Kenya (establishing platform) Formative learning, what the pilot revealed for scale Teacher livelihood Both: teachers gained durable digital and inclusive-pedagogy capability; several now act as Inclusive Pedagogy Leaders in their own staffrooms (“growing in the professional space”). The model builds transferable professional capital that stays with teachers beyond the app. At scale this positions teacher development as a retained, this conforms the hypothesis that say when teacher capacity improves ite has the potential to accelerating learning for learners Government engagement Both: government engaged across the full cycle, design, fieldwork, quality assurance and data entry; both Ministries asked to continue. Government ownership held identically across two different systems, making it the model’s most transferable asset and the credible pathway to continuity. Scale should formalize this through matched funding and ministry-led coaching. Feasibility Tanzania: deliverable on an existing platform; compliance rose 42→85 (G2) and 39→88 (G3). Kenya: deliverable even while numeracy content and signs were built for the first time; 57→74 and 56→88. The model is deliverable whether a system is established or- establishing; deliverability is not the binding constraint on scale. The question shifts from whether it can be delivered to whether the conditions that sustain it are resourced. Acceptability Tanzania: teachers and the Institute of Education endorsed the model; uninvited learners joined lessons(spillover). Kenya: Ministry opened and closed the training; the Director of Special Needs judged it “a national standard … ready for scale.” Acceptance held across two systems and rested on the model’s fit with how deaf learners access learning through sign and demonstration indicating demand is a property of the approach, not of a favourable setting.

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Index What the evidence shows Tanzania (established platform) and Kenya (establishing platform) Formative learning, what the pilot revealed for scale Dosage Both: identical dose, a 3-day training, two coaching visits (Week 4, Week 8), and 384 recorded sessions with sustained weekly use. Signal: higher-order practice rose while coaching was present and tapered once it withdrew. The coaching dose is a core mechanism of the model, not a supplementary input, and must be cost as continuous at scale. Learning outcome In both countries, learners moved out of Below Expectation, and the movement concentrated in the lower bands where most began. Mean scores rose in Tanzania from 16.1 to 26.4 (Grade 2) and 5.6 to 23.9 (Grade 3), and in Kenya from 12.8 to 18.6 and 9.7 to 12.3, where levels remained lower and applied domains flatter. The control groups moved in the same direction. Read against their small samples and the maturation expected over a school term, the gains indicate broad progress across all groups Score gains appeared in intervention and control alike, consistent with learning maturation over a school term. Tanzania’s higher endpoint partly reflects a two-year class cycle against Kenya’s single year. Access & inclusion channels Tanzania & Kenya: visual and sign engagement reached 76– 80% and digital-tool use 62–68% in intervention classrooms, against near-zero in control, where these channels did not exist. Opening numeracy through sign, visual and digital channels is the model’s defining contribution and the one signal maturation cannot produce, it emerged even in the establishing system. For scale, this is the property to protect first.

Figure 6: Qualitative insights from Tanzania KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Teacher Diaries week 1 to 8 Tanzania/Kenya Over 8weeks of sharing their experience of using the intervention, this is the progress of tree of their experiences Tanzania Interactive Word Tree

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Intervention Kalimani, an AI-enabled inclusive numeracy model for deaf and hard-of-hearing learners; a sign-language avatar (KSL/TSL) delivers visual, demonstrated numeracy, with teacher professional development and coaching, formative assessment, and targeted remediation Location 12 government primary schools’ special units and inclusive schools across Kenya and Tanzania Design 8 intervention and four control schools, spanning special-unit and inclusive settings; teacher practice observed in both arms, learners assessed at baseline and endline Period April to July 2026, approximately eight weeks of classroom use Learners assessed 309 at baseline and then we had 319 at the endline deaf and hard-of-hearing learners across Grades 2 and 3 (200 intervention, 109 control) and 10 learners that added only at endline Teachers trained & observed 24 teachers 16 in intervention schools, trained and coached to deliver Kalimani, and 8 in control schools who continued standard practice; observed at baseline, Week 4, Week 8, and endline (intervention) and at baseline and endline (control) Platform delivery 384usage and 300+ learner assessment attempts; 30-day active users rose from 187 to 396 over the period Results Table 17: Summary of pilot

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT References Bavelier, D., Dye, M. W. G., & Hauser, P. C. (2006). Do deaf individuals see better? Trends in Cognitive Sciences, 10(11), 512–518. https://doi.org/10.1016/ j.tics.2006.09.006 Black, P., & Wiliam, D. (2009). Developing the theory of formative assessment. Educational Assessment, Evaluation and Accountability, 21(1), 5–31. https:// doi.org/10.1007/s11092-008-9068-5 Blatto-Vallee, G., Kelly, R. R., Gaustad, M. G., Porter, J., & Fonzi, J. (2007). Visual–spatial representation in mathematical problem solving by deaf and hearing students. Journal of Deaf Studies and Deaf Education, 12(4), 432–448. https://doi.org/10.1093/deafed/enm022 Century, J., & Cassata, A. (2016). Implementation research: Finding common ground on what, how, why, where, and who. Review of Research in Education, 40(1), 169–215. https://doi.org/10.3102/0091732X16665332 Durlak, J. A., & DuPre, E. P. (2008). Implementation matters: A review of research on the influence of implementation on program outcomes and the factors affecting implementation. American Journal of Community Psychology, 41(3–4), 327–350. https://doi.org/10.1007/s10464-008-9165-0 Fixsen, D. L., Naoom, S. F., Blase, K. A., Friedman, R. M., & Wallace, F. (2005). Implementation research: A synthesis of the literature (FMHI Publication No. 231). University of South Florida, National Implementation Research Network. Fredricks, J. A., Blumenfeld, P. C., & Paris, A. H. (2004). School engagement: Potential of the concept, state of the evidence. Review of Educational Research, 74(1), 59–109. https://doi.org/10.3102/00346543074001059 Henner, J., Pagliaro, C., Sullivan, S., & Hoffmeister, R. (2021). Counting differently: Assessing mathematics achievement of signing deaf and hard of hearing children through a unique lens. American Annals of the Deaf, 166(3), 318–341. Heritage, M. (2010). Formative assessment: Making it happen in the classroom. Corwin. Ivers, N., Jamtvedt, G., Flottorp, S., Young, J. M., Odgaard-Jensen, J., French, S. D., O’Brien, M. A., Johansen, M., Grimshaw, J., & Oxman, A. D. (2012). Audit and feedback: Effects on professional practice and healthcare outcomes. Cochrane Database of Systematic Reviews, (6), CD000259. https://doi. org/10.1002/14651858.CD000259.pub3 Kraft, M. A., Blazar, D., & Hogan, D. (2018). The effect of teacher coaching on instruction and achievement: A meta-analysis of the causal evidence. Review of Educational Research, 88(4), 547–588. https://doi.org/10.3102/0034654318759268 Marschark, M., & Hauser, P. C. (2012). How deaf children learn: What parents and teachers need to know. Oxford University Press. National Joint Committee on Learning Disabilities. (1998). Operationalizing the NJCLD definition of learning disabilities for ongoing assessment in schools. https://www.asha.org/policy/rp1998-00130/

KALIMANI INCLUSIVE NUMERACY | LEARNING REPORT Pagliaro, C. M., & Kritzer, K. L. (2013). The math gap: A description of the mathematics performance of preschool-aged deaf/hard-of-hearing children. Journal of Deaf Studies and Deaf Education, 18(2), 139–160. https://doi.org/10.1093/deafed/ens070 Personal Data Protection Act, No. 11 of 2022, United Republic of Tanzania. (2022). Proctor, E., Silmere, H., Raghavan, R., Hovmand, P., Aarons, G., Bunger, A., Griffey, R., & Hensley, M. (2011). Outcomes for implementation research: Concep- tual distinctions, measurement challenges, and research agenda. Administration and Policy in Mental Health and Mental Health Services Research, 38(2), 65–76. https://doi.org/10.1007/s10488-010-0319-7 Santos, S., & Cordes, S. (2021). Math abilities in deaf and hard of hearing children: The role of language in developing number concepts. Psychological Review. Advance online publication. https://doi.org/10.1037/rev0000303 UNESCO. (2020). Global education monitoring report 2020: Inclusion and education: All means all. UNESCO. https://doi.org/10.54676/JJNK6989 Uwineza, I., Uworwabayeho, A., & Yokoyama, K. (2024). Classroom practices in primary schools’ mathematics teaching supported by the Interactive Mathematics Software for Rwanda. International Journal of Evaluation and Research in Education, 13(1), 511–522. https://doi.org/10.11591/ijere. v13i1.25244 World Bank. (2018). World development report 2018: Learning to realize education’s promise. World Bank. https://doi.org/10.1596/978-1-4648-1096-1 World Bank. (2019). Ending learning poverty: What will it take? World Bank. http://hdl.handle.net/10986/32553