Skip to content

MEM/FWU Roundtrip Plan

Status: implemented reference roundtrip Scope: SkillPilot curriculum landscapes as a practical compatibility probe for MEM-compatible RDF/OWL representations based on the FWU Lehrplan Ontology

Purpose

SkillPilot should treat Metadata for Educational Media (MEM) and the FWU-DE/lehrplan-ontologie project as an adjacent semantic schema, metadata, and provenance layer, not as a competing model.

The practical question is:

How can schema-level curriculum metadata and concrete curricular data become usable competence landscapes from which a rule-bound navigation logic can derive the next didactically justified learning step?

SkillPilot can serve as an open-source reference implementation and test bench for this question.

Reference concept paper:

https://aifyer.com/ki-lernbegleitung/versions/v1.0/schulisch-verantwortete-ki-lernbegleitung-v1.0.pdf

Framing

MEM and the FWU Lehrplan Ontology are understood as important infrastructure for:

  • machine-readable curriculum references
  • educational standards
  • validity scopes such as Bundesland, Schulart, Fach, Jahrgangsstufe, Niveau, and Bildungsgang
  • source and provenance records
  • cross-state comparability

SkillPilot starts from the adjacent runtime need:

  • concrete competence landscapes
  • explicit requires dependencies
  • learner-facing composition views
  • validation rules for DAG integrity and route quality
  • frontier logic for the next sensible learning steps

The integration work should therefore ask what a MEM-compatible representation must carry so that AI-supported learning guidance can do more than identify curriculum references: it must also derive next learning steps in a traceable way.

Planned Test

The core validation mechanism is a roundtrip test:

SkillPilot curricular competence landscape
    ->
MEM-compatible RDF/OWL representation based on FWU Lehrplan Ontology
    ->
reconstructed SkillPilot curricular competence landscape
    ->
SkillPilot validation

The second transformation must reconstruct the SkillPilot landscape from the semantic bundle without reading or relying on the original SkillPilot ZIP input. The bundle consists of RDF/OWL plus explicitly referenced binary sidecars such as goal visualizations; binary files are not embedded as opaque Base64 literals in RDF.

First Quality Goal

The first quality goal is a semantically lossless MEM/FWU roundtrip for the mathematics Gymnasium publication package:

skillpilot-de-gymnasium-mathematik-v0.1.0.zip
    ->
FWU/MEM-first semantic bundle with a minimal SkillPilot profile and referenced assets
    ->
reconstructed SkillPilot knowledge-landscape data
    ->
semantic validation

The goal is not to reconstruct the ZIP byte-for-byte. The goal is to reconstruct all content that is required to use, validate, and independently inspect the SkillPilot knowledge landscape.

This makes the roundtrip a concrete quality gate and a concrete MEM/FWU compatibility probe. If the remaining SkillPilot profile terms point to reusable ontology gaps, the result should be turned into a focused issue and, where feasible, a small PR in FWU-DE/lehrplan-ontologie.

Initial Scope

The first test scope should be conservative:

  • German Gymnasium landscapes
  • initially canonical Gymnasium data in SkillPilot
  • later additional German school forms
  • later broader school and curriculum scopes if the first roundtrip is successful

The first export package should be a simple ZIP containing:

  • SkillPilot landscape JSON files
  • JSON schemas
  • composition views
  • mapping and provenance files
  • validation configuration or reports

Individual learner state and personal mastery data are out of scope for the first public roundtrip.

Required Semantic Coverage

The roundtrip is meaningful only if the RDF/OWL representation can preserve all SkillPilot-required data, including:

  • goal identity: id, shortKey, title, description
  • goal classification: atomic/cluster, node kind, core/extension where applicable
  • graph structure: contains
  • didactic dependencies: requires
  • progression metadata: weight, phase/topic/course-level metadata where relevant
  • program structure: programUnits
  • placements: goalPlacements
  • competency axes: competencyCatalog, competencyRefs
  • learner-facing structure: composition views
  • source and provenance references
  • goal-visualization references, metadata, package paths, and byte-identical image assets
  • validation-relevant metadata

The RDF/OWL representation must not simply embed the original SkillPilot JSON as an opaque blob. It must express the data semantically enough that an independent reverse transformation can reconstruct the package.

Expected Ontology Profile

The current FWU Lehrplan Ontology already covers many curriculum metadata concepts, but it does not directly define all SkillPilot runtime semantics or a media-package format.

The implemented roundtrip uses a small SkillPilot profile aligned with the FWU ontology for the semantics that the core intentionally does not cover.

The first profile focus remains the two graph relations that are explicit in the concept paper's learning-goal graph:

  • requires / didactic prerequisite
  • contains / learning-goal composition

Mappings and learner-facing composition views remain a separate application layer and are included in the semantic bundle without being recast as curricular parthood.

Since FWU-DE/lehrplan-ontologie#9, didactic prerequisites have a dedicated core pattern. A prerequisite is a reified LP_0000554 resource, attached to the source through LP_0030071 (hat Verweis) and pointing to the prerequisite goal through LP_0030072 (verweist auf). This replaces the earlier skillpilot:didacticRequires proposal for the canonical roundtrip.

Strict curricular composition uses BFO_0000051 (hat Teil) directly. Every authored direct edge is additionally retained as skillpilot:containsGoal, because BFO parthood is transitive and a reasoner may materialize indirect descendants. The application assertion is therefore the lossless direct-edge anchor; mixed runtime, practice, assessment, memory, orientation, or navigation edges use only that assertion.

Candidate application terms that still remain useful:

skillpilot:LearningGoal
skillpilot:AtomicGoal
skillpilot:ClusterGoal
skillpilot:containsGoal
skillpilot:weight
skillpilot:core
skillpilot:shortKey
skillpilot:ProgramUnit
skillpilot:GoalPlacement
skillpilot:CompositionView
skillpilot:GoalVisualizationReference

The referenced media target uses the standard schema:ImageObject class rather than a new SkillPilot asset class.

Important modeling constraint:

A generic FWU CE-Verweis must not be treated as a prerequisite. Only the specialized LP_0000554 pattern carries SkillPilot requires semantics.

A generic curriculum reference is not automatically a didactic prerequisite. The specialized core class now provides the explicit semantics needed for frontier and learning-path logic while retaining a first-class edge resource for rationale, source, scope, version, and review metadata.

Likewise, plain parthood must be checked for domain fit. SkillPilot contains is additionally represented as BFO_0000051 / hat Teil only when the relation is a genuine curricular part-whole relation: a cluster or goal comprises subgoals that are semantic parts of that cluster or goal. It must not be used for UI placement, ordering, loose topic association, state visibility, memorization/practice nodes, or view mappings. skillpilot:containsGoal preserves the authored direct graph independently of BFO inference.

Goal visualizations reuse the core CE-Verweis pattern from a goal to an external resource. The target image is represented with standard media vocabulary and shipped as a hashed binary sidecar. Package path, checksum, byte length, review state, and exact SkillPilot link reconstruction remain legitimate application-level packaging metadata because the FWU core does not define a binary media package model.

Deliverables

  1. SkillPilot ZIP export format for the selected Gymnasium scope.
  2. JSON schemas and SkillPilot validation command for the ZIP.
  3. Transformation project A: SkillPilot ZIP to RDF/OWL.
  4. Transformation project B: RDF/OWL to SkillPilot ZIP.
  5. RDF/OWL profile documentation for the SkillPilot-specific terms.
  6. Binary resource index and byte-integrity checks for goal visualizations.
  7. Roundtrip comparison report.
  8. List of ontology gaps or candidate relation additions for discussion in the FWU GitHub project.
  9. Draft issue or PR text backed by the current roundtrip evidence, if the test reveals a generally useful MEM/FWU modeling improvement.

Success Criteria

The first roundtrip is successful when:

  • the reconstructed ZIP validates with the existing SkillPilot validators
  • graph invariants are preserved:
  • no requires cycles
  • no contains cycles
  • referential integrity
  • valid composition views
  • all SkillPilot-required fields are reconstructed deterministically
  • every active goal-visualization reference resolves to exactly one packaged image and the reconstructed sidecar is byte-identical by SHA-256
  • canonicalized source and reconstructed JSON are equivalent for required data
  • RDF/OWL core-pattern validation passes for the intermediate representation; until the upstream release artifacts and auto-shapes contain LP_0000554, the roundtrip applies an equivalent local source/reference/target structural check and records the pinned core module
  • the transformation does not depend on hidden copies of the original JSON

Collaboration Channel

Findings that concern modeling, interfaces, validation, or missing relations should be contributed through the open channels of the FWU project:

https://github.com/FWU-DE/lehrplan-ontologie

The intended stance is collaborative:

SkillPilot is not a counter-model to MEM. It is a concrete test bench for checking which information a MEM-compatible structure must carry so that curriculum-aware AI learning coaching can derive traceable next learning steps.