Skip to content

Skill Graph Production Pipeline

Generated artifact. Do not edit manually.

Generated by: scripts/build_compendia.py Regenerate with: python3 scripts/build_compendia.py --compendium production-pipelines --web-and-tex Source of truth: docs/production-pipelines/source/

Language / Sprache: English (EN) · Deutsch (DE) | PDF: English PDF · Deutsches PDF

Compendium: §1 Skill Graph · §2 Package Provisioning · §3 Subject Export Package · §4 MEM/FWU Roundtrip · §5 Exam Example · §6 Sek-I Math Exam

1. Skill Graph Production Pipeline

This chapter defines the general, repeatable, auditable blueprint for turning any authoritative or institutional curriculum source into a clean SkillLandscape JSON graph under curricula/<...>/json/<file>.json (or curricula/DE/Gymnasium/canonical/; for exam authoring, see 5).

1.1. Inputs and Steps 0–2: Extraction, Outline, and JSON Generation

Inputs comprise (1) a normalized plain-text source extract capturing topics, outcomes, prerequisites, grading scope, and source URL/locator, (2) a human-curated outline separating root, clusters, and measurable atomic goals, plus optional (3) sibling landscapes for consistent naming and ID patterns:

  • Step 0 (Extract source): Normalize the official curriculum into clean plain text (no HTML) with an explicit URL/page reference.
  • Step 1 (Structured outline): Create a root node, group phases/topic fields into clusters, and cut competencies into measurable atomic goals (1–3 tasks, phrased as "The learner can ..."). Include motivation anchors and phase-local autonomy clusters of ordinary exam character (Übungen E-Phase, Übungen Q1–Q4, plus optional Übungen Prozesskompetenzen), keeping global Abitur/final-exam branches separate as an additional assessment layer.
  • Step 2 (Encode JSON):
  • Stable IDs: Project-namespaced UUID v5 or deterministic mapping from (landscapeId, shortKey) (shortKey is optional, unique within landscapeId; repeating the same (goalId, shortKey) pair across locale files is valid, reusing one shortKey for distinct goalIds is forbidden).
  • Required fields: Top-level landscapeId, title, description, locale, subject, goals; goal-level id, title, description, weight, requires, contains, dimensionTags (at least phase; prefer area, topicCode, demandLevel, processCompetencies, guidingIdeas; optional tags, core, examples, sourceRef, resourceLinks).
  • Localization, assets & links: Multilingual landscapes keep separate meaning-aligned locale JSONs; co-located images use relative Markdown paths. sourceRef holds canonical provenance, while resourceLinks holds ordered learning links (broad overviews on root/clusters, goal-specific practice/visualization links on atomic nodes).

1.2. Steps 3–7: Relations, Reference Implementations, Validation, and Registration

  • Step 3 (contains hierarchy): Form an acyclic DAG (cluster \(\to\) subgoals) with shallow, meaningful grouping.
  • Step 4 (requires sequencing): Author atomic-to-atomic prerequisites without transitive redundancy (Graph Definition, Graph Validation Rules); inherited cluster requires remain a transition mechanism only. Every learner-facing atomic goal should lie on at least one teachable route from a motivation anchor to a phase-local autonomy goal. Two retained Hessen snapshots under curricula/DE/Gymnasium/input/HE/upper-secondary/source-json/ serve as practical authoring benchmarks: DE_HES_S_GYM_2_PHYSIK.de.json.snapshot and DE_HES_S_GYM_2_MATHEMATIK.de.json.snapshot (local Übungen E-Phase/Q1–Q4 plus Übungen Prozesskompetenzen, cleanly separated from global Abitur containers; live runtime landscapes reside under curricula/DE/Gymnasium/canonical/).
  • Step 5 (Validation & manual reviews):
cd app
npm run validate:graph

Complement structural checks with manual atomicity (Atomic Review) and prerequisite (Requires Review) passes.

  • Step 6 (Register curriculum): Add root curricula to curricula/curriculum_manifest.json or attach sub-curricula via parent contains.
  • Step 7 (QA checklist): Clear titles/descriptions testable in 1–3 tasks, canonical atomic requires, acyclic contains/requires, closed motivation-to-local-autonomy routes per phase, stable IDs, consistent metadata (phase, area, level, core, weight), and aligned localization.