Alignment call, 15 Sep 2026

Decisions with timestamps; how the transcript was recovered

On Student Data within Taleemabad (Across Regions) | Alignment Call — 15 Sep 2026, 08:15–10:04 UTC

Meeting id: d7ee93ac-3a82-4136-9f07-8e38a495e1a2 · bot bot_H3Bulsz0fzzAe9aa · row is restricted (hidden from the meetings_team_ro role) and its recording_jobs row was stuck at processing with no audio key, so the pipeline never produced a transcript. Pulled the recording via the Attendee API and transcribed with Soniox: probe2_soniox_alignment_call.mp3.md (272 diarised turns, 1 h 49 min). Speaker ids: spk1 = Haroon, spk4 = Mashhood (chairing), spk2 = Ahwaz (impact), spk3 = Ali Sipra (inferred from content; verify before quoting by name). Participants (calendar): Mashhood, Ali Sipra, Amena, Sabeena, Haroon, Hataf, Ahwaz, Sameer.

What was decided (verbatim-anchored)

  1. "I want to invest in a better test. I'm not happy with ASER. More granularity is the need of the hour." (Haroon, 25:00). ASER is a crude yearly/twice-yearly instrument; the goal is to see learning levels move on faster intervals, which is a function of the instrument's granularity as much as of sampling frequency (23:44). Fluency changes faster than paragraph-with-meaning; decoding/encoding/blending (what the phonics-heavy lesson plans teach) are the "minute aspects" the instrument should pick up (33:25).
  2. The promise is 30 minutes per coach per day, on top of the regular observation flow; 5 children at ~5 minutes each; child goes back to class after. Not released until it is proven under 30 minutes. Islamabad first, a few coaches tested; port to other regions ~14–20 days later; Sumaiya and Mara to confirm the 30-minute ask holds for AEOs (54:36, 61:02). "We don't go live till we can get down to this time period" (61:02).
  3. Complete battery, all three subjects — "we need to have our finger on all three" (Urdu, English, Maths), Urdu prioritised (56:52). Comprehension is built into EGRA (1:16:41).
  4. Randomised, pre-selected roster; the coach does not choose the children. The impact team's algorithm names the children and class; the coach pulls them out to test (52:07). Sampling debate: 5 new children per visit initially to cover the roster and establish levels, repeat tests next term (Ali, 20:47); vs 3 new + 2 old per visit (Haroon, 29:30). Impact team's standing recommendation for outcome measurement: EGRA every 3 months, population-size sample (Ahwaz, 1:10:38; 1:14:25).
  5. Instrument: EGRA-derived, not ASER, not LEAPS/TIP unless someone shows LEAPS is better (1:14:30). Ahwaz: fitting EGRA into 30 min for 5 children "would be very difficult" (1:15:44). Haroon: same abstraction as FICO — a big form the AI pre-fills so the coach's job gets faster; measure time aggressively and keep iterating until it is under 5 min per child, as with the roster exercise (20 → 18 → 10 min) (56:52).
  6. Disaggregated EGRA as the formative layer (Luminos Fund pattern): break EGRA/EGMA competencies into small weekly quizzes (letter ID, phonics, …) and use them to predict the quarterly FLN result (Ahwaz, 1:20:46; Mashhood: "change the quiz to be a disaggregated EGRA assessment", 1:22:00).
  7. SLO-level visualisation for coaches in the night-before pre-visit brief (1:27:51, 1:28:27).
  8. Automate enumerator/coach QA from the audio — the QA officers' brief for the May enumerator recordings goes into the doc so we can decide whether to automate it "similar to how Rumi does FICO: feed it the assessment-aligned indicators and the flow, and it does the QA check from the audio" (Mashhood + Ahwaz, 1:38:13–1:40:14). Otherwise QA enumerators on retainer for rolling QA.

Open questions raised in the call

Also discussed (first 20 min): the student quiz

Video/WhatsApp quiz reach is biased to the top 15–20% with smartphones; nudges lifting completion; mandate discussed; SLO-level cuts already show fractions/division gaps.