Run a visit yourself

Partly. The Grade 3 stories and all of Set A's maths are the May 2026 instrument, tried on 874 children, so their difficulty is known. Not calibrated yet: the Grade 5 story continuations and all of Set B (no child has read them), and the way questions are asked (EGRA asks only about what the child read; the bot asks all three).

What it costs: comprehension that punishes slow readers twice; Grade 3 sums 3–4 that only 6% get right and Grade 5 number pairs most get all right; no listening task, so a low English score can't be split into reading and spoken English; scores rough for one child (maths ±2 of 10). The full table, and the three cheapest fixes.

One grade per visit (Grade 3 or Grade 5, picked by the bot). The list is grouped by classroom, with the class teacher, in the roster's own words: "Grade 3 - A · Teacher: Saima Bibi". From the NIETE roster (924 Grade 3/5 classes, 31,826 children): every child has a full name, 92% of classes have a section letter, 99% a reachable class teacher; two in three visits need only one classroom. One button sends each class teacher only their own room's children, in order.

Done. No roll number in any message, the form or the portal. Two classmates with the same name (2% of children) get the father's name when the roster has it, or "2 in this class" when it doesn't.

There is no rotation any more. EGRA gives every child in a round the same form; forms rotate between children only in pilots and group tests. So each coach carries one card set per grade for the whole term, colour-coded (Grade 3 green, Grade 5 blue) with "Set A · Oct–Dec 2026" printed on every side, and the bot names the card in every step. The print pack is A4, laminated, one task per side, as EGRA's are: the print pack above, PDF sent to you.

EGRA's initial-sound task: the adult says a word ("machhli") and the child says its first sound ("m"). It is a Grade 1–2 task, Pakistan's Grade 3 tests drop it, and our AI could only guess at it. The "coach's copy" was a 5-page paper sheet the bot never mentioned. Both are gone.

Your sandbox account had been switched to Urdu for the first demo recording and never switched back; it is back to English. Every message now follows the coach's language (try both buttons above). Only the words the coach says to the child stay in Urdu, and English for the English story. Each step is now numbered actions: which card, what to say, when to stop, when to send.

The old design had the next child write maths while the current child read, to save time; that child wasn't even in the room. Removed. One child at a time, start to finish; maths is spoken, no strip, no photo, no child number.

The child reads the four number pairs and the four sums on one maths card and says the answers; the coach points to each and says "which is bigger?" / "tell me the answer". The coach reads the two word problems aloud with the card face down. Try the maths step above: the table shows both.

That whole arrangement is gone: no strip, no second page, nothing handed to another child. The child in front of the coach does Urdu, English, maths; then the next child comes.

Reads it aloud, slowly, once, as EGMA does (EGMA reads word problems to the child so a weak reader isn't marked down in maths; the child may use paper and pencil; no explaining). The two problems are in the maths message and on the coach card.

Yes: Pakistan's Urdu EGRA asks 5 questions; we ask 3. In EGRA the questions are never on the child's card; the assessor asks them aloud after taking the passage away. The old message never told you to ask them. Now step 4 of the Urdu message lists them, the coach turns the card face down and asks; the table above shows them with translations.

They don't count. The "key" was the test child's answer sheet in my demo; real coaches have none. EGRA assessors train 5+ days to mark every word live and still disagree. Here the recording is the score sheet: the AI counts from it, and on the May Rawalpindi children it agreed with the enumerators more often than a second trained person did (Urdu 74% vs 58% within ±5 words). The coach only answers the few things the recording couldn't settle, at the end.

Yes: the EGRA/EGMA assessor holds their own sheet (paper or tablet) with the questions and accepted answers and the word problems to read; only the child's card is bare. In v2 the phone message is that sheet, and a printed coach card per grade ("don't show the child": questions with accepted answers, word problems, sum answers) is in the print pack (pages 5 and 7), as the paper backup.

No. In a data-collection round every child in every school reads the same form; that is what makes schools comparable. Forms change between rounds (baseline, endline) using parallel forms checked to be equally hard. So: one set for every school this term, Set B next term. EGRA rounds last weeks, our term three months, so the cards stay with the coach (never left at a school or photographed), the questions are never printed for children, and we can watch for leakage: a passage whose scores climb through the term while letters and words don't.

Two things had been called rotation: two card forms in one visit, and the next child writing maths while the current one read. Both are gone. The time came back from elsewhere: the coach no longer counts or fills a form per child (about 2 minutes each), first sounds, made-up words and the 60-second quick sums are dropped, and maths is one short spoken set. Measured on real WhatsApp: about 5 minutes per child, about 27 minutes for five plus a 1-minute review.