Math Question Bank — P1-S3 Coverage

Source-by-source breakdown of what we have across all shared P1-S3 question banks. S4-S6 (DSE) generators are already live and serving the product; this report covers the primary and junior secondary expansion.

Prepared 22 Apr 2026 · essai × talentcoop

29,174
P1-S3 questions indexed
35,721
question images mirrored
4 / 4
P1-S3 sources complete
1
source pending access

1. Summary

We have indexed the full P1-S3 surface from every shared source. Four question banks, 29,174 question records, 35,721 images — all tagged by grade (P1–S3) and ready to power the product. Every record from S4-S6 (DSE) is already live in the app via Essai's own generators.

Per source:

  1. UPEP OnlineQB — complete: 14,597 questions across all 6 books (1A–3B = S1–S3). Per-book grade split: S1 3,703 · S2 5,637 · S3 5,257. All questions have answer data and images mirrored.
  2. EdCity STAR (Math packages) — complete: 510 papers / 11,826 questions covering P1–P6 and S1-S3, both Chinese-medium and English-medium. 100% with answer data.
  3. TSA papers (P6 + S3) + EDB curriculum — complete: 452 TSA S3 questions (84.9% with answer), 760 TSA P6 questions (96.9% with answer), and 1,539 EDB curriculum sections (used for topic calibration).
  4. Grade tagging — unified: every record carries a consistent grade label (P1–P6, S1–S3) across all four sources.

One source pending: UPEP AI Generator (~580 Qs) — account access request needed (see §5).

2. The reframed source scoreboard

The original "15 sources" list bundled everything regardless of grade. Re-cut by P1-S3 relevance and current state:

2a. Sources that serve the P1-S3 workstream

Source Grade Native Qs On disk now Status Effort to land
UPEP OnlineQB
teacher account, 6 books, JSMIA series
S1-S3 14,597 14,597 Qs / 29,188 images · grade-tagged per book · ~1% images pending complete
EdCity STAR — math packages
Chinese-medium + English-medium
P1-S3 11,826 510 papers / 11,826 Qs / 6,533 images · 100% with answer complete
TSA S3 (2024 + 2025)
24 PDFs, 9ME papers
S3 452 452 Qs · 84.9% with answer complete
TSA P6 (2019-2024)
20 PDFs, 6ME papers
P6 760 760 Qs · 96.9% with answer complete
TSA P3
parsed text from prior round
P3 60 60 Qs · parsed complete
EDB Curriculum (KS1-KS3)
6 PDFs — KS1/KS2/KS3 explanatory notes + content + KLA guide
P1-S3 1,539 sections curriculum text (no answers — used for topic calibration) calibration only
UPEP AI Generator
role-gated — access request pending
P1-S3 (partial) ~580 not yet pulled access pending ~10 min once access granted

2a-bis. Per-source × canonical grade matrix (P1-S3 corpus, on disk now)

Every record across the four P1-S3 sources now carries a canonical grade_canon tag in {P1..P6, S1..S3} or a range bundle (P1-P3, P4-P6, P1-P6, S1-S3, P1-S3). Counts below are actual on-disk record counts (jq over canonical JSON), not estimates.

Source P1 P2 P3 P4 P5 P6 S1 S2 S3 P1-P3 P4-P6 P1-P6 S1-S3 P1-S3 Total
UPEP OnlineQB 3,7035,6375,257 14,597
EdCity STAR 1,4001,3381,4981,6261,0781,372 158 128823,146 11,826
EDB Curriculum 126100252303758 1,539
TSA P6 760 760
TSA S3 452 452
TOTAL 1,4001,3381,4981,6261,0782,132 3,7035,6375,867 2541822523,449758 29,174

Reading the matrix: single-grade columns (P1..S3) are clean grade labels; range columns are bundled labels that the source provides without finer resolution (e.g. STAR's "Key Stage 3 (S1-S3)" papers, EDB's curriculum guides keyed to whole stages). Single-grade S1-S3 totals are S1: 3,703 · S2: 5,637 · S3: 5,867 (UPEP single-grade plus 452 TSA S3 plus 158 STAR S3); plus 3,449 records bundled under S1-S3. Single-grade P1-P6 spread is balanced (1,078-1,626 per grade) — STAR alone covers it. Bundled "P1-S3" records (758) are entirely EDB curriculum and span the whole P1-S3 surface.

2b. Sources that are S4-S6 / DSE — already-covered ground

Source Grade Native Qs On disk now Net-new value for P1-S3?
EdCity OQB Math
14 NSS strands
S4-S6 3,117 indexed No — feeds the S4-S6 generators already live in the product
DSE MC-Bank
2023 Paper 2 etc.
DSE 585 indexed No
DSE bytopic PDFs
38 of ~66 parsed
DSE 2,220 indexed No
Teacher mock papers (.docx)
55 mocks
DSE 1,411 indexed No — DSE mock papers
EdCity OQB Chem / Bio / ICT NSS ~1,700 parsed No (cross-subject, not math)

2c. Reference-only

SourceUseFilesOn disk
HKAT / PSLE / UK SATsCalibration / cross-jurisdiction reference46yes (defer)

3. Recommended next moves

Source indexing is done. The next moves are downstream: wiring the corpus into the product and closing the one outstanding access item.

  1. Knowledge-graph alignment. Tag the 29,174 P1-S3 records against the topic taxonomy (P1–S6 knowledge structure). EdCity STAR and UPEP both carry structured topic codes — most of the alignment can be automated. ~4-8 dev hours for the full P1-S3 surface.
  2. Generator-seed pipeline for P1-S3. Pick 2-3 high-frequency P1-S3 topics (suggested: P3 Number sense, P5 Fractions, S1 Linear equations, S2 Pythagoras). Build question templates using UPEP and STAR records as gold examples per topic.
  3. Direct-delivery pipeline for long-tail topics. For topics where generation doesn't add value, deliver curated UPEP/STAR questions directly — with topic tag and difficulty proxy. Cheaper and faster for thin-coverage areas.
  4. UPEP AI Generator access — see §5 for the outstanding request.

4. Open items

5. Version notes