Roadmap

Milestone 1: Foundation

Status: complete for the current starter.

Status: complete for CSEE Mathematics 2023 syllabus topics.

Register all 44 current syllabus topic entries with stable IDs and slugs.
Create one leaf wiki topic page per syllabus topic.
Keep hub pages as navigation and grouping pages.
Connect topics to subject, form, source, competence, and hub records in graph data.

Status: first five-year Basic Mathematics pilot complete.

Flattened 2021-2025 Basic Mathematics Paper 1 JSON extracts into 191 answerable leaf-question records.
Mapped 147 records to 2023 syllabus topic IDs.
Preserved 44 unmapped records for review.
Added 125 conservative review-queue records for low-confidence, multi-topic, figure/table-dependent, missing-mark, or uncertain cases.
Used the 2022 CSEE examination-format crosswalk to compare actual exam-question patterns against expected table-of-specification groups.
Current pilot should be treated as unreviewed exam intelligence until checked against original papers.

Status: recommended next before adding more subjects.

Introduce data/curricula/{level}/{subject_slug}/{year}.json as the canonical home for official curriculum versions.
Add data/curricula/index.json with current curriculum pointers, historical versions, canonical subject IDs, subject aliases, language notes, source IDs, and review status.
Keep data/curriculum_map.json as a temporary Mathematics compatibility view until existing question mappings and page generators understand the versioned layout.
Adopt append-only curriculum versioning: add future syllabuses, such as 2028, as new curriculum records instead of overwriting 2023.
Namespace new form, competence, hub, and topic IDs by level, subject, and year to avoid collisions across subjects.
Separate curriculum topics from durable concepts: topic-csee-physics-2023-measurement is an official syllabus placement, while concept-measurement is the reusable idea.
Model legacy and multilingual names as aliases. For example, Civics aliases should support search and redirects without rewriting historical records into the current Historia ya Tanzania na Maadili identity.

Status: planned.

Expand source records with checksum, file size, source URL where known, retrieval date, extraction date, extraction tool/version, rights note, public/private availability policy, and review status.
Treat syllabus PDFs under raw/ as private evidence artifacts and structured curriculum files as the operational rules.
Add extraction audit records linking source IDs to extraction runs, extracted curriculum records, and review outcomes.
Confirm a public build can ship wiki pages, structured curriculum data, graph projections, citations, and checksums without requiring raw/**/*.pdf.

Status: planned after architecture and provenance milestones.

Add versioned 2023 curriculum records for Physics, Chemistry, Biology, Geography, History, Historia ya Tanzania na Maadili, English Language, and Kiswahili.
Create subject, source, form, hub, and topic stub pages from the versioned curriculum records.
Keep the first pass to official curriculum spine stubs; do not add exam mappings or full learner chapters until the multi-subject graph validates cleanly.
Flag duplicate sources, ambiguous topic boundaries, translation uncertainty, and syllabus-table extraction issues in review queues.

Status: recommended next.

Review data/review_queue_question_mapping.jsonl.
Correct or confirm question-to-topic mappings.
Reduce unmapped question count where the original PDF supports a clear mapping.
Upgrade high-signal topic pages with learner-facing explanations, examples, common mistakes, and past-question signals.
Start with exponents, logarithms, matrices, rates and variations, similarity and congruence, sets and Venn diagrams, ratios and proportions, and approximations.

Status: planned.

Define a sample-exam specification using topic frequency, form coverage, hub coverage, and exam-format groups.
Decide whether sample questions are selected, adapted, or newly generated.
Add difficulty, marks, topic IDs, source policy, and review status to each sample question.
Keep generated sample exams separate from official past papers.

Status: planned.

Add official marking schemes where available and source-verified.
Add textbook-backed explanations and examples.
Add teacher notes and multimedia resources with citations.
Keep all new content mapped to the stable topic registry.
Treat NECTA and official Tanzania Institute of Education sources as the constitutional layer for topic names, topic boundaries, competences, form placement, and exam coverage.
Use non-NECTA sources only as enrichment: aliases, background context, diagrams, prerequisite hints, examples, and cross-subject bridges.
Keep external enrichment separate from official curriculum records until reviewed.
Add external enrichment candidates with source, confidence, rationale, and review status before promoting them into the main graph.

Status: planned.

Use Wikidata as the preferred structured external graph for canonical IDs, aliases, multilingual labels, and same-as links.
Use Wikipedia for learner-readable background context and disambiguation, with citation and licensing discipline.
Use Google Knowledge Graph only as an optional entity-disambiguation helper, not as a bulk curriculum authority.
Add typed enrichment edges such as topic_has_external_alias, topic_has_external_reference, topic_has_candidate_prerequisite, topic_has_reviewed_prerequisite, and topic_has_cross_subject_bridge.
Prevent node-edge congestion by capping weak external fanout, keeping broad hubs navigational, and promoting only reviewed high-signal links.
Store rejected or uncertain external matches in a review queue rather than deleting the audit trail.

Status: planned.

Build a local search/retrieval index over wiki pages and structured data.
Test local model workflows with source-grounded retrieval before any fine-tuning.
Add answer-generation rules that cite wiki pages and source files.
Use human feedback to improve retrieval quality.