OverviewActive

high · Leslie asked on 2026-04-01 whether the shared question files are enough for all grades or if Essai should still purchase textbooks offline. Eric proved the EdCity scraper works on 2026-04-02: 3,117 math MCQs captured, but about 95% are image-based.

high · Mar 28 check-in set this as a P0. 107 skills already exist, but they are not useful to Leslie's AI chat until moved into DB tables.

high · Explicit ask from the Mar 28 check-in. Needed so Leslie can reason about how Trap/context data feeds the AI chat.

high · Leslie's Apr 8 meeting notes reframed the marketing thread into a concrete question: which features should be shown next, rather than just 'make more marketing assets'.

high · Leslie meeting notes 2026-04-08 and WA: after showcase integration, next step is EdCity inside AI Chat with backend on Essai GitHub. OQB CLI already exists at tools/edcity-client.js.

high · Leslie's latest meeting notes added a new packaging/product priority: revise the UI design to make the question bank more central, alongside the ongoing math QA and EdCity integration threads.

critical · Eric

high · Eric

critical · Leslie, Renee

high · Eric

high · sub-e1 (2026-04-21) ran but hit packages 13-16 unfiltered; pulled 8,282 records that turned out to be Chinese+English language papers. sub-e5 (2026-04-22) discovered: pkg 13=star_chi, 14=star_eng, 15=star_math_zh, 16=star_math_en. Re-pull restricted to [15, 16]. Output: workdir/math/corpus/edcity-star/ (canonical/raw/images); prior pull at wrong-subject-quarantine/.

high · sub-e6 (2026-04-22) confirmed 14,597 P1-S3 math questions live across books 1A/1B/2A/2B/3A/3B (sc=SECJRMATH, JSMIA series). 100% image-based stems — must mirror assets. List pull is ~3 min; image asset mirror is multi-hour. Recommended priority: 1A→1B→3B→2A→3A→2B (ascending Q count, largest book last).

high · Leslie flagged 中英夹雜 in some math questions and shared a full rules pack: 5 anti-patterns, English→香港書面語 term map, trad punctuation/style rules, LLM checklist + 4 regex lints, Apr 2026 audit snapshot row counts. Paths (in sygy1920-code/essai monorepo): docs/playbooks/math-question-generation-language-rules.md, api/scripts/investigate_mixed_lang_math.js, api/scripts/fix_mixed_lang_math.js. Eric to decide how much lands in workdir/math CLIs vs stays in essai-api and how to align with existing chinese.js / trap phrase hygiene.

high · After reviewing the agentic OQB report, Leslie said the direction is good and asked for a Word document teachers can download for the showcase.

medium · Leslie's Apr 8 notes make teaching material a live workstream again: more school feedback and format discovery are needed. Current source pile still spans WA-only assets: teacher logins, Pearson/Oxford/mock-paper files, Mar 30 Sayo screenshots, and related competitor/reference material.

medium · Apr 8 WA + Leslie notes: Renee surfaced Cloud SAMs / school case material, while Leslie framed the broader narrative around privacy plus agentic consolidation across fragmented school systems.

medium · STAR records use mixed grade strings: P1/P2/.../P6 (English medium pkg 16), 小一/小二/.../小六 (Chinese medium pkg 15), 'Key Stage 3 (S1 - S3)' / '第三學習階段﹙中一至中三﹚' (KS3 in both). TSA records use S3/P6 cleanly. EDB uses KS1/KS2/KS3/curriculum. Need a canonical mapper to {P1..P6, S1..S3} for downstream competency mapping + UI filtering. Pure normalization — no re-extraction.

low · sub-e2 (2026-04-21) processed 54/55 docx mocks → 1,411 records + 1,721 PNGs but parser bundled multiple Qs per record, filled topic_native with header text, left answer_native = null. All S6 DSE mocks (not P1-S3 critical). Output: workdir/math/corpus/teacher-mocks/.

medium · sub-e4 (2026-04-22) parsed 20/20 TSA P6 PDFs → 765 records, but only 2.6% got answer_native populated. MS files for 12/16 papers absent or alignment broken. Need paper-specific patterns + keyed 1./(1) alignment. Output: workdir/math/corpus/tsa-p6/.

medium · Audit reframed to P1-S3 active workstream (DSE/S6 already covered by Essai generators). Subagents sub-e4/e5/e6 results folded in.

low · UPEP AIQ generator returns errorCode 1002 across 13 endpoint variants because account roles array is empty. Not technical — partner-context blocked. Small (~580 Qs). Defer until UPEP QB pull lands and we have momentum to point to.

2026-04-07

2026-04-07

2026-04-03

2026-04-02

2026-03-31