EGRA, EGMA, ICAN, TIP, ASER, ELANA, DIBELS, MELQO, Literacy Boost, Pakistan incumbents, AI readers
Purpose: give the operator a one-screen-per-instrument view of what each major early-grade assessment tests, how it is run, and what it reports, so we can adjudicate which items go into our 5-minute, coach-administered, AI-scored battery (English + Urdu + maths, Pakistani government primary, grades 1–5). Prepared: 2026-09-15. Companion to 01_EGRA_EGMA_structure.md (which holds the full EGRA/EGMA detail and is only summarised here). Claims not confirmed against a primary source are marked unverified.
Primary sources read directly: RTI EGRA Toolkit 2nd ed. and EGMA Toolkit (via 01_); PAL Network ICAN 2020 Report (https://neqmap.bangkok.unesco.org/wp-content/uploads/2020/07/2020_PAL-Network_ICAN-Report_EN-1-compressed.pdf) and ICAN 2022-SSA Report (https://palnetwork.org/wp-content/uploads/2023/06/ICAN-2022-SSA-Report.pdf); ASER Pakistan ICAN page (https://aserpakistan.org/International-Common-Assessment-of-Numeracy-ICAN), FAQ 2023 (https://aserpakistan.org/document/2023/9/Frequently_Asked_Questions_ASER.pdf) and ITA PSL Guide & Training Manual (https://itacec.org/document/2019/december/Guide_and_Training_Manual_PSL.pdf); PAL ELANA pages (https://palnetwork.org/elana/, https://palnetwork.org/pal-elana-advancing-early-learning-assessment-across-12-countries/); ACER Assessment GEMS: Uwezo (https://www.acer.org/files/AssessGEMs_Uwezo.pdf); U. Oregon DIBELS 8 Administration & Scoring Guide (https://dibels.uoregon.edu/sites/default/files/2024-01/dibels8_admin_scoring_guide.pdf) and Benchmark Goals (https://dibels.uoregon.edu/sites/default/files/2024-01/dibels8_benchmark_goals.pdf); Frontiers in Public Health 2021 MELQO item paper (https://pmc.ncbi.nlm.nih.gov/articles/PMC7888256/); Room to Read Sepedi 2017 Endline (https://www.roomtoread.org/media/z2gaoxvy/za-sepedi-2017-endline-evaluation-technical-report.pdf); PIE NAT 2023 Findings Report (https://pie.gov.pk/SiteImage/Downloads/NAT%202023%20Findings%20Report%2006.03.2024%20-Final-%20v6%20.pdf); Andrabi et al. Free to Choose (TIP, https://tahir-andrabi.sites.pomona.edu/wp-content/uploads/2025/04/Free-to-choose.pdf); CERP TIP page (https://www.cerp.org.pk/targeted-instruction-program/); Amira ISIP Assess page (https://amiralearning.com/amira-isip-assess) and Oklahoma SDE Amira guidance (https://oklahoma.gov/content/dam/ok/en/osde/documents/services/literacy-policy-and-programs/strong-readers-for-schools/24-25%20Amira%20Screening%20Technical%20Guidance%20Amira.pdf).
Owner / reach: RTI International for USAID (2006–), 70+ countries; Pakistan versions under PRP (Urdu, Sindhi, Pashto), SRP and Sindh SELECT 2024 (7,248 G2 pupils on Tangerine). Target: grades 1–3. Purpose: system monitoring and impact evaluation; diagnostic only at aggregate level. Administration: 1:1 oral, trained enumerator, paper or Tangerine tablet; 15–20 min (Toolkit), 10–15 min for 5-subtask forms. Every timed grid is 60 s with an auto-stop if nothing is correct in the first line; 3-s hesitation = move on; self-corrections count. Score: rate per minute via score / (60 − time_remaining) × 60; comprehension as correct/asked. Reported as cwpm distributions, % zero scores and % meeting benchmarks (Urdu G2 "meets" = 60–90 cwpm; Sindh 2024 Urdu-medium mean 19 cwpm, 45% zero). Urdu: yes (PRP/SRP/SELECT); open design questions are letter-name vs letter-sound, aerab on/off, RTL first-line stop logic. Reliability: enumerators must hit 90% agreement with a scripted gold standard; field IRR 88–98%.
| Subtask | Items | Timed | Stop rule | Measures |
|---|---|---|---|---|
| Listening comprehension | 3–5 Q on a ~30-word read-aloud | no | none | oral language |
| Letter sound (or name) ID | 100-letter grid | 60 s | 0 correct in first 10 | alphabetic knowledge, clspm |
| Initial sound ID | 10 words | no | 0 in first 5 | phonemic awareness |
| Nonword reading | 50 | 60 s | 0 in first 5 | decoding, cnwpm |
| Familiar words | 50 | 60 s | 0 in first 5 | sight/word reading, cwpm |
| Oral reading fluency | ~60-word passage | 60 s | 0 in first line | cwpm |
| Reading comprehension | 3–5 Q (only text reached) | no | not asked past point read | comprehension |
| Dictation / vocabulary / maze | optional | — | — | weakly validated |
Good at: continuous, comparable rates with known benchmarks; component skills. Bad at: too long for 5 min; rates are meaningless if the 60-s window is shortened without re-norming.
Owner / reach: RTI for USAID (2008–; Core EGMA 2014). Target: grades 1–3. Purpose: system monitoring / impact. Administration: 1:1 oral, ~20 min (plan 25); the child must say every answer (pointing "is insufficient evidence"); 5-s no-response prompt; untimed subtests stop after 4 successive errors. Pre-2014 versions had counting, one-to-one correspondence, shapes and patterns; they were cut for length. Score: timed subtests as rate/min, untimed as proportion correct; no criterion bands in the Toolkit (countries set their own, e.g. Tanzania A/S-L2 80%). Urdu/Pakistan: SRP adapted EGMA for Sindhi (grades 3 and 5); no published Urdu EGMA norms found (unverified).
| Subtest | Items | Timed | Stop rule | Measures |
|---|---|---|---|---|
| Number identification | 20 (0–999) | 60 s | none | numeral reading |
| Number discrimination | 10 pairs | no | 4 errors in a row | magnitude |
| Missing number | 10 | no | 4 errors | patterns / number line |
| Addition L1 / Subtraction L1 | 20 + 20 (sums ≤19) | 60 s each | none | fact fluency |
| Addition L2 / Subtraction L2 | 5 + 5 (two-digit) | no | 4 errors | procedures; paper allowed |
| Word problems | 6, read aloud | no | 4 errors | application |
Good at: fact fluency and procedure separately; oral so reading does not confound. Bad at: 20 minutes; no geometry/measurement/data.
Owner / reach: PAL Network (17 citizen-led members); ICAN 2019 in 13 countries, one rural district each, incl. Pakistan (ITA/ASER: Toba Tek Singh, 1,616 children age 5–16, 1,198 households); ICAN 2022-SSA repeated in Kenya, Mozambique, Nigeria. Open source, 11 languages incl. Urdu. Target: one tool for ages 5–16, items pitched at grade 2–3 (GPF-aligned). Purpose: system monitoring for SDG 4.1.1(a), household-based so out-of-school children are included; not a classroom diagnostic. Administration: oral, 1:1, in the home, by trained volunteers; paper and pencil allowed; average 15 min per child (20–30 min per household with the questionnaire). Structure: 26 items (2022 report; PAL's site now says 36 in the current version), half number knowledge, half geometry + measurement + data display, arranged in three sets: Set 1 (counting, comparing groups, shapes, position/direction, length/capacity, time/calendar, data chart) and Set 2 (number recognition, add/sub/mult/div "Task 1") go to every child; Set 3 (harder operations "Task 2" and subtraction/division word problems) only to children who got the corresponding Set 2 task right — a gated, two-tier design, not adaptive item selection. Number recognition uses the ASER rule: 4 of 5 correct. Scoring: total correct; policy-linking set the Minimum Proficiency Level at 17/26 for grade 2 and 21/26 for grade 3; reports show % of class 2–3 children who can do "at least 1 task each on spatial orientation, shape, measurement, number recognition and ≥3 number operations" (Pakistan 2019: 32.3%). Reliability: no psychometrics published in the two reports read (unverified); PAL cites ASER-style volunteer training and recheck protocols.
| Domain (share) | Tasks (Set) |
|---|---|
| Number knowledge (50%) | count objects, compare groups (S1); recognise 1-digit and 2-digit numbers, 4/5 rule (S2); +, −, ×, ÷ without and with carry/borrow/remainder (S2 → S3); word problems (S3) |
| Geometry | position/direction, shape recognition (S1) |
| Measurement | length/capacity, clock time, calendar day/date (S1) |
| Data display | read a bar chart, "how many more" (S1) |
Good at: broad domain coverage (shapes, time, data) that EGMA lacks, in 15 min, comparable across countries. Bad at: no fluency/speed, few items per domain, one score with two cut-points.
We could not open the Drive folder, so this is an identification exercise. Candidates found, with evidence:
| Candidate | What it is | Child test? | Pakistan link | Likelihood |
|---|---|---|---|---|
| TIP — Targeted Instruction Program / "Targeted Instruction in Pakistan" (LEAPS-Harvard, CERP, EdTech Hub, KP govt) | 40-day TaRL-style remediation, grades 1–5, maths + Urdu + English. Starts with a written diagnostic covering Class 1–3 curriculum (4 equivalent forms, calibrated), children who fail Class-1 items get an oral KG-level assessment; pupils sorted into KG / C1 / C2 / C3 groups; teachers grade with a smartphone "tech tool" (offline-capable); quizzes every 5–7 lessons. ICT pilot 2022–23 (560 schools, ~91k pupils, 2,200 teachers); KP Peshawar + Mardan (1,250 schools, target 250k pupils). Gains 0.19 SD (optional tech). | Yes | Yes — federal + KP govt, Islamabad schools | High — a Pakistani grade 1–5 Urdu/English/maths placement test is exactly the shape of our brief; "TIPS" is plausibly "TIP assessment" |
| TIPPS — Teacher Instructional Practices and Processes System (NYU Global TIES) | Classroom observation rubric of teacher behaviours; used in DRC, Uganda, Ghana, India; Pakistan roll-out "planned" | No | Planned only | Low for a student test; medium if the folder is about coaching |
| TIPS — Teacher Induction Programmes (PIE / NAT 2023 glossary) | Teacher-side programme name that appears in the NAT 2023 report's abbreviation list | No | Yes | Low |
| Pratham "TIPS", "Tool for Identifying Pupils' Skills" | No such instrument found in any source | — | — | None found |
Verdict: treat the folder as the TIP diagnostic (LEAPS/CERP) until the operator confirms; ask for the folder to be shared with the service account. If it is TIP, its relevance is the level ladder (KG/C1/C2/C3 in three subjects) and the fact that it is written and group-administered — the opposite of what we are building, but the natural comparator for "what does the government already accept".
Owner / reach: ITA (Pakistan, 2008–), Pratham/ASER Centre (India, 2005–); 2023 Pakistan round 151 rural districts + 123 urban centres, 272,370 children. Target: ages 5–16, tool pitched at Class 1–2 curriculum. Purpose: system monitoring (district estimates) and TaRL grouping. Administration: household, 1:1 oral, volunteers with 3 days' training, untimed; reading + arithmetic ≈ 10 min (Vagh 2012, India). The tester ticks the highest level reached, testing from the middle of the ladder and moving up or down. Languages: Urdu, Sindhi, Pashto, English; separate arithmetic and general-knowledge sheets; two sample forms per tool.
| Tool | Ladder (bottom → top) | Criterion |
|---|---|---|
| Urdu/Sindhi/Pashto reading | beginner → letters → words → sentences (Class 1 text) → story (Class 2 text) | 4 of 5 letters; 4 of 5 words; sentences read "as sentences, not strings of words"; story "fluently with ease"; bonus comprehension Q at story level |
| English | beginner → capital letters → small letters → words → sentences | 4 of 5 at each list level; 2 of 4 sentences |
| Arithmetic | beginner → number recognition 1–9 → 10–99 (PSL variant adds 100–200) → subtraction (2 sums, both correct, borrowing) → division (1 sum, right method + answer) | 4 of 5 numbers; start at subtraction |
| General knowledge | picture/"largest" items, 60-s items | new in 2019 PSL guide |
Score: an ordinal level per child; reported as % at each level (e.g. % of grade-5 children who can read a Class-2 story). Reliability: Hindi validation r = .90–.94 with EGRA-style fluency battery; test–retest κ .76; no Urdu mapping published (unverified). Relation to EGRA: untimed, criterion-referenced, 5-item sampling; "story level" ≈ 44+ wpm on grade-2 text in Hindi.
Good at: 10 minutes, any volunteer, one ladder the whole country already understands. Bad at: no rate, ceiling at Class 2, 5 items per rung so a single child's level is noisy.
Uwezo (Twaweza/PAL; Kenya, Tanzania, Uganda; ages 6–16; household; volunteers; grade-2 level) is the East-African ASER: literacy ladder letter/syllable (5 of 10) → words (5 of 10) → paragraph (4 sentences) → story (6–14 sentences) → 2 comprehension questions; numeracy: count objects 1–9 → recognise 10–99 → greater-than (Uganda) → 2-digit addition → subtraction → 1×1-digit multiplication → division (3 of 8 items each). Same virtues and limits as ASER; no Urdu.
PAL-ELANA (Early Language & Literacy and Numeracy Assessment, 2021–; 12 countries incl. Pakistan via ITA; 9 languages incl. Urdu): ages 4–10; tablet-based, oral 1:1 (no child reading/writing needed); a three-stage multi-stage adaptive design — performance on the first stage routes the child to easier or harder blocks, ending in low/medium/high competency bands mapped to the GPF and ACER learning progressions. Literacy tasks cover oral language → decoding → reading comprehension in five levels; numeracy covers numbers, geometry, data, measurement, patterns in five levels. Item counts, minutes per child and psychometrics are not published on the pages read (unverified); rollout was 180 communities per country. This is the closest existing thing to "adaptive, digital, Urdu, 4–10" and the tool ITA will compare us to.
Owner / reach: University of Oregon (DIBELS 8, 2018–; Amplify distributes) and Acadience Learning (Acadience Reading K-6, formerly DIBELS Next); the US universal-screening standard, K–8, three times a year. Purpose: classroom screening and progress monitoring (not system monitoring). Administration: 1:1, teacher, five 60-second subtests plus a 3-minute group Maze; whole benchmark ≈ 5–8 min in K–1; 3-s hesitation rule; self-correct within 3 s counts; discontinue if 0 correct in the first line (LNF) or first 5 (PSF/NWF/WRF); "gating" rules skip harder subtests for a child who scores 0 on an easier one. Score: raw count per minute per subtest plus a weighted composite; each score falls in a band — blue (negligible risk), green (core), yellow (strategic), red (intensive) — derived so that ~80% of children above/below the 40th percentile on an external reading test land in the right band. Reliability/validity: ORF concurrent validity with Iowa Total Reading .75 (fall G1) to .91 (winter/spring G1); accuracy alternate-form reliability median .79 (Technical Manual). Urdu: none; Latin-script only.
| Subtest | Grades | Time | Measures | G1 end-of-year "core" goal (DIBELS 8) |
|---|---|---|---|---|
| LNF letter naming | K–1 | 60 s | alphabet automaticity | 59+ |
| PSF phoneme segmentation | K–1 | 60 s | phonemic awareness | 61+ |
| NWF nonsense words (CLS / WRC) | K–3 | 60 s | decoding | 87+ letter-sounds / 28+ whole words |
| WRF real words | K–3 | 60 s | word reading | 50+ |
| ORF passage (words correct + accuracy) | 1–8 | 60 s | fluency | 76+ cwpm; 91%+ accuracy |
| Maze (every 7th word blanked) | 2–8 | 3 min, group | comprehension | G2 end 18.0+ (adjusted) |
Acadience end-of-year ORF goals: G1 47, G2 87, G3 100 cwpm; NWF-CLS G1 fall 27 / winter 43 / spring 58.
Why it matters for us: it proves a complete K–1 profile in ~5 minutes is a solved design problem — one skill, one minute, one risk band — and that gating on the easiest subtest is standard practice. Bad at: no oral language, no maths, English-only norms.
Target: ages 4–6 (48–83 months), school readiness. Purpose: system monitoring of pre-primary; adaptable per country. Administration: 1:1 direct assessment by trained enumerator, untimed, ~20–30 min, plus a caregiver/teacher report. Structure (cross-country robust core, 12 countries, 24,533 children): 16 tasks / 84 items — early literacy 27 items (letter ID 17, listening comprehension 4, initial sound 3, letter sound 2, name writing 1); early numeracy 29 (number ID 10, addition/subtraction 5, spatial position 4, set production 3, mental transformation 3, shape naming 3, number comparison 1); executive function 27 (Head-Toes-Knees-Shoulders 12, pencil tap 12, digit span 3); social-emotional 1. Score: domain scores, no fluency, no proficiency bands (countries set them). Urdu/Pakistan: not found (unverified); Bangladesh, Tanzania, Lao versions exist. Good at: what a pre-reader can do (EF, oral language, spatial). Bad at: ceiling by grade 1; 30 min.
Literacy Boost (2009–; 30+ countries incl. Pakistan — Allai 2010 baseline, PEPAS/ICT 2014 with Numeracy Boost; Urdu forms exist): programme-evaluation tool for grades 1–3, 1:1 oral, enumerator; subtests are concepts about print, letter knowledge (letters known out of the alphabet), most-used/single-word reading, then passage fluency (wpm), accuracy (%), reading comprehension, with listening comprehension for non-readers; the headline indicator is "reading with comprehension" = ≥75% of passage comprehension questions correct. Item counts and timings not confirmed in sources read (unverified). Reports % readers / % reading with comprehension. Useful mainly as precedent that an NGO tool with a Pakistan/Urdu form treats accuracy and fluency as separate scores.
Room to Read (Literacy Program endlines, 20 countries; SurveyCTO tablets): a 4-task EGRA subset — letter sounding (60 s, clpm), nonword reading (60 s), oral passage (~60 words, up to 3 minutes, cwpm) and 5 comprehension questions. Bands: fluent reader ≥45 cwpm, emergent 20–45, non-reader below; comprehension "met" = 4 of 5; grade-1 letter-sound benchmark 45 lpm. Sepedi 2017: project G2 mean 31 cwpm, 2/5 comprehension. No Urdu. Precedent for a four-task EGRA that funders accept.
| Instrument | Owner | Who / when | What is tested | Format & time | Output | Verdict for us |
|---|---|---|---|---|---|---|
| LND (Literacy & Numeracy Drive), Punjab 2015– | PITB + SED/PMIU | Grade 3, monthly; MEA picks 7 pupils per school visit; 47k schools, ~329k pupils/month | Urdu, English, maths SLOs: comprehension, sentence completion, 2–3-digit add/sub, ×, ÷ | 7 MCQs on the MEA's tablet, child taps; minutes | % correct per school; feeds school ranking (3 marks) | Recognition-only, no oral reading, no fluency; teachers "teach to the LND app". Not a model, but it is the incumbent accountability metric in Punjab |
| NAT 2023 (NEAS/PIE) | Pakistan Institute of Education | Grade 4 and 8, sample of 1,304 schools, 23,456 pupils, 18–19 May 2023, every ~2 years | G4: Urdu/Sindhi reading (52 items), English (48), maths (48) + a 15-item Foundational Literacy test of letter and word recognition; G8 maths, science | Paper, group, 4-option MCQ; ~1 test booklet per subject | % items correct (G4: FL 88%, Urdu/Sindhi 68%, English 56%, maths 49%; 1 in 6 at guessing level in maths) | System monitoring only; no oral component; grade 4 is above our floor |
| PEC/PECTAA SBA, Punjab 2019– | PEC (defunct) → PECTAA | Grades 1–8 school-based assessment replaced the G5/G8 board exams under the 2019 Assessment Policy Framework | Written papers per subject; 2025 SBA adds listening/speaking/reading rubrics on separate days | Paper, whole class | Marks | Written; irrelevant to a 5-min oral screen except as the "grade level" reference |
| Sindh SELECT EGRA 2024 (World Bank) | RSU Sindh | Grade 2, 7,248 pupils, 604 schools, Tangerine | 5 subtasks: initial sound (10), nonwords (60 s), expressive vocabulary (10 pictures), 60-word passage (60 s) + 5 Q, listening (3 Q); Urdu + Sindhi | 1:1 oral, ~10 min | cwpm, % zero, % meeting SELD standards | The Urdu template that already exists and is government-owned — see 01_ §4 |
| ASER Pakistan | ITA | see §5 | ||||
| TIP diagnostic | LEAPS/CERP | see §4 | written placement, oral for KG level | KG/C1/C2/C3 | ||
| Urdu phonics / phonological tools | academic | U-PASS (Frontiers 2025): 10 PA/PM/RAN subtests, Urdu nonwords built on the mohmil convention, Nastaliq | tele-assessment | research | Only published Urdu nonword construction rule; no school screener found (unverified that any exists) | |
| SDIP / EG-EQUIP tools | — | Not found in any accessible source; the World Bank Sindh project's grade-3 EGRA (baseline Apr 2023, province-wide Apr 2024) is the nearest match (unverified) |
| Product | What one session measures | Length | Languages | Evidence |
|---|---|---|---|---|
| Amira Learning / ISIP Assess (HMH) | Child reads aloud to the app; ASR scores ORF (wcpm), accuracy, plus RAN (2 min), word manipulation/PA (3–5), letter names+sounds (2), word ID fluency (2–3), nonsense words (2–3), vocabulary (2), spelling (5), reading comprehension (2–3); dyslexia-risk flag; Lexile and mastery scores | ≤20 min, self-administered on a device | English, Spanish | NCII "highest" ratings; sensitivity/specificity >.80 (vendor); sole state screener in 5 US states |
| Google Read Along (Bolo) | Child reads a levelled story to "Diya"; on-device ASR gives accuracy, speed (wpm), comprehension (embedded Qs), phonics skills, per story and last-5 trend in Classroom | one story, ~3–10 min; no fixed test | Urdu, Hindi, English + 7 others | 64% of India-pilot children improved on ASER; internal "<45 cwpm beginner" metric; no ASR-vs-human accuracy published |
| Microsoft Reading Progress / Coach (Teams) | Teacher assigns any passage; child records; ASR reports accuracy rate, correct wpm, mispronunciations, omissions, insertions, repetitions, self-corrections; Coach adds practice and attempts | passage length set by teacher; 1–5 min | 100+ passage languages, Coach fewer; Urdu unverified | 8-week study +22 wpm vs +10.5 control (Microsoft blog) |
| Curious Learning (Feed the Monster / Curious Reader) | Instructional game (letter sounds, phonics, sight words); no published in-app assessment; outcomes measured externally with EGRA | — | 50+ languages | World Bank Nigeria 9,000-household study cites app use second only to tutoring |
Read-across: every AI-native product converges on the same three numbers per read-aloud — wcpm, accuracy %, error taxonomy — and adds comprehension questions; only Amira attempts the sub-word skills (letters, nonwords, PA) by machine.
| Assessment | Reading domains | Maths domains | Min / child | Administration | Outcome metric | Urdu / Pakistan |
|---|---|---|---|---|---|---|
| EGRA | listening, letters, PA, nonwords, words, ORF, comprehension | — | 10–20 | 1:1 oral, enumerator, tablet | cwpm rates, % zero, benchmark bands | Yes (PRP, SRP, SELECT) |
| EGMA | — | number ID, discrimination, missing number, ± L1/L2, word problems | 20 | 1:1 oral | rates + % correct | Sindhi; Urdu unverified |
| ICAN | — | number, operations, word problems, geometry, measurement, data | 15 | 1:1 oral, household, gated sets | total/26, MPL 17 (G2) / 21 (G3) | Yes (ITA, Urdu) |
| ELANA | oral language, decoding, comprehension (5 levels) | numbers, geometry, data, measurement, patterns | n/p | tablet, 1:1, 3-stage adaptive | low/med/high band | Yes (ITA, Urdu) |
| ASER PK | letters→words→sentences→story; English ladder | numbers 1–9, 10–99, subtraction, division | ~10 (both) | 1:1 oral, household, volunteers | ordinal level | Yes (Urdu/Sindhi/Pashto) |
| Uwezo | letter→word→paragraph→story+2 Q | count→recognise→+→−→×→÷ | ~10 | household, volunteers | ordinal level | No |
| DIBELS 8 / Acadience | LNF, PSF, NWF, WRF, ORF, Maze | — | 5–8 (K–1) | 1:1, teacher, 60 s each | counts/min, 4 risk bands, composite | No |
| MELQO MODEL | letters, sounds, listening, name writing | number ID, ±, shapes, spatial, sets | 20–30 | 1:1, enumerator | domain scores | Unverified |
| Literacy Boost | print concepts, letters, words, fluency, accuracy, comprehension, listening | (Numeracy Boost) | n/p | 1:1 oral | % readers; ≥75% comprehension | Yes (Urdu forms) |
| Room to Read | letter sounds, nonwords, ORF, comprehension | — | ~8 | 1:1, tablet | cwpm; fluent ≥45 / emergent 20–45 / non-reader | No |
| LND Punjab | comprehension, sentence MCQs | ± ,×,÷ MCQs | ~3 | child taps MEA's tablet | % correct per school | Yes (Urdu, English) |
| NAT 2023 | G4 reading MCQ + 15-item letter/word FL | G4 maths MCQ | ~60 | paper, group | % correct | Yes (Urdu/Sindhi) |
| TIP diagnostic | Urdu, English written, oral at KG | Class 1–3 written | n/p | written, teacher-graded via app | KG/C1/C2/C3 group | Yes (ICT, KP) |
| Amira | ORF, accuracy, PA, letters, nonwords, vocab, spelling, comprehension | — | ≤20 | self, ASR | wcpm, risk, Lexile | No |
| Read Along | accuracy, wpm, comprehension, phonics | — | per story | self, on-device ASR | per-story stats | Yes (Urdu) |
| Reading Progress | accuracy, cwpm, error types | — | per passage | self, ASR | per-passage stats | Urdu unverified |
The invariants — components in nearly every battery 1. Letter knowledge (names or sounds) — EGRA, DIBELS, ASER, Uwezo, MELQO, Literacy Boost, Room to Read, NAT-FL, Amira. The universal floor item; the only disagreement is name vs sound. 2. Decoding via nonwords or single words — EGRA, DIBELS (NWF/WRF), ASER/Uwezo (word rung), Room to Read, Amira. Every serious tool separates "can decode" from "has memorised". 3. Connected-text oral reading with a rate — EGRA, DIBELS, Room to Read, Literacy Boost, and all three AI readers; ASER/Uwezo have the rung but not the rate. cwpm is the one number every funder and ministry already understands (Urdu G2 60–90; Sindh 2024 mean 19). 4. Comprehension questions on that text — universal; 3–5 literal + 1 inferential; ≥75–80% is the common "meets" bar. 5. Number recognition — EGMA, ICAN, ASER, Uwezo, MELQO; and basic operations (addition/subtraction; ASER's subtraction-with-borrowing rung; ICAN gates ×/÷ behind them). Word problems appear in EGMA and ICAN only.
Unique to one family (drop unless we have a reason): phoneme segmentation (DIBELS PSF); maze (DIBELS); RAN, spelling, executive function (Amira, MELQO); shapes/time/calendar/data display (ICAN/ELANA/MELQO); dictation (EGRA optional, "hard to score consistently"); listening comprehension (EGRA, MELQO, Literacy Boost — valuable but slow; Sindh 2024 kept 3 questions).
What the 5-minute budget forces: the timed EGRA/DIBELS grids are 60 s each plus instructions; five of them alone exceed 5 minutes. Every short instrument in the table solves this the same two ways — gating (ASER/ICAN/DIBELS: stop or skip on the easy rung) and subsetting (Room to Read 4 tasks, Sindh 2024 5 tasks, Tanzania 5). A defensible 5-minute core is therefore: letter-sound grid (gated) → nonword or word grid → 60-word passage + 3 questions → number recognition → one timed addition/subtraction strip, with ASER-style 4-of-5 stop rules, and everything else optional per grade.
What AI scoring buys that paper never could 1. Rate on every item, not just the passage: with audio timestamps, ASER's ordinal ladder gains a cwpm for free, and the 5-item rungs can grow to 10 without adding tester time. 2. Accuracy and an error taxonomy (omissions, substitutions, self-corrections) as separate scores — the AI-reader convention; paper enumerators only tick right/wrong. 3. Comprehension answers in any language, LLM-graded (Henkel et al. QWK .92) — lets us ask more inferential questions. 4. Automatic time-remaining and stop-rule enforcement — the two things Tangerine had to build UI for and where human IRR is lowest. 5. Oral maths as first-class: EGMA insists the child says the answer; audio makes that the record, and strategy (fingers, paper) can be photographed. 6. Adaptive routing across grades 1–5 (ELANA's three-stage idea) without a bigger paper kit.
What AI does not fix: letter- and nonword-level ASR for children in a low-resource script is the hard case (Xhosa 6.9% off-the-shelf; ~91% only after per-item fine-tuning with 200–300 labelled clips), and ASR over-flags errors on weak readers. The coach's tap remains the record of truth per subtask until the AI meets the 90%-agreement enumerator bar (01_ §6).
Open decisions for the operator (the ones this walkthrough cannot make): letter names vs sounds for Urdu; aerab on/off in nonwords and passages; whether to keep ASER's ordinal levels as the reported outcome (government-legible) with cwpm as the secondary metric, or the reverse; whether maths gets 90 s (number ID + one operations strip) or 2 min (add ICAN's shapes/time item); and whether the TIPS folder is the TIP diagnostic — if so, its KG/C1/C2/C3 ladder is the grouping the ICT and KP governments already use and we should map onto it.