Learning Map

How the map was built

The pipeline, the numbers, and the honest limitations — so you can judge for yourself how much weight to put on any alignment.

The skills and their connections

The skills come from the open Marble Skill Taxonomy (© Generative Spark, Inc., CC BY-SA 4.0 / ODbL 1.0): 1,590 topics across 8 subjects and 54 domains, connected by 3,221 prerequisite links. Each link is typed — hard (must come first) or soft (helps) — and carries a written reason. Each skill carries a plain-English description, evidence of mastery, an assessment prompt and a typical age range. We render this data as-is; we don’t edit skills or connections.

The Australian Curriculum data

No machine-readable version of the Australian Curriculum v9 F–6 content descriptions existed, so we made one: all 614 content descriptions (Foundation to Year 6) parsed from ACARA’s official DOCX files, with code, learning area, strand, year level and full text (© ACARA, CC BY 4.0) — English 189, Mathematics 139, HASS 84, Science 77, Health & Physical Education 59, Digital Technologies 38, Design & Technologies 28. One parsing bug — dropped fraction glyphs in AC9M3N02 — was found and fixed against the official source; a sweep confirmed it was the only affected record. The dataset lives at github.com/curiousape-ai/acara-v9-f6-json and is downloadable here.

One version note: the candidate alignments below were computed against the 475-record v0.1 parse of this dataset. The 139 sub-strand descriptions added in the current 614-record set have no candidate matches yet — their pages simply show the official text. A v0.2 re-match against the full set is planned.

The matching

Each Marble skill was matched to candidate curriculum codes by semantic similarity: a sentence-transformer model (all-MiniLM-L6-v2) over skill and code descriptions, boosted by exact-term TF-IDF, and constrained so matches stay within a compatible subject and a plausible year range (age ≈ year + 5, ±1 year). The same pipeline ran against 164 NSW syllabus outcome codes across the five K–6 syllabuses.

Every match gets a confidence band from its similarity score: high (score ≥ ~0.62–0.66 depending on subject), medium (≥ ~0.50–0.53) and low (≥ ~0.40–0.42). Below that, no match is shown. The bands are honest about uncertainty — a “low” match is a lead worth checking, not a mapping.

The numbers

Australian Curriculum v9 (F–6)NSW syllabuses (K–6)
Skills with candidate matches1,007 of 1,590795 of 1,590
— best match high confidence233183
— best match medium confidence461350
— best match low confidence313262
No F–6 / K–6 match445657
Beyond Year 6120120
General capabilities1818

Unmatched skills are often finer-grained than the curriculum describes, sit beyond Year 6, or align to ACARA’s General Capabilities rather than a learning area. Their pages say so honestly.

The limitations

  • Every alignment is a candidate — machine-suggested and unreviewed (v0.1). None has been verified by a teacher or curriculum authority. Never rely on them for registration or compliance.
  • Semantic matching rewards similar wording; it can miss conceptually equivalent pairs phrased differently, and over-match superficially similar ones.
  • Ages on the map are the taxonomy’s typical ranges, not Australian year-level prescriptions.
  • NSW outcome descriptions aren’t reproduced here (they’re not openly licensed), so NSW matches show codes and stage labels only — always read the official outcome at curriculum.nsw.edu.au before acting on a match.

What’s next

A community review workflow is planned: reviewed matches will graduate from “candidate” to “community-reviewed” in a future dataset version. The scores in v0.1 stay frozen — this site renders the data, it never re-matches.

Want the raw material? Everything is on the data page, openly licensed.