robworks-software/k12-science-standards

[!WARNING] Deprecated - use [`k12-science-standards-expanded`](https://huggingface.co/datasets/robworks-software/k12-science-standards-expanded) instead. This dataset is superseded: every instruction in this set also appears there, plus 1,123 more and nine additional metadata columns. Nothing here is unique to it. It stays online so existing references keep resolving, but it will not be updated. New work should point at robworks-software/k12-science-standards-expanded.

K-12 Science Standards (generated instruction data)

6,787 instruction/input/output records for science, generated around a K-12 standards taxonomy for instruction-tuning and educational-content experiments.

How this was built (read this first)

These are programmatically generated training examples, not curriculum written by educators and not the text of any official standard. A generator combined standards metadata - codes, grade levels, domains, Bloom's and DOK levels - into instruction and output text using a fixed set of templates.

Of 4,750 training rows, only 664 instructions (14.0%) and 572 outputs (12.0%) are distinct. Text is assembled by concatenating NGSS taxonomy fields into a frame, so one output recurs 26 times verbatim.

Structured metadata fields (standard codes, grade levels, domain names) come from the source taxonomy and are reliable. The generated prose is not.

Loading

from datasets import load_dataset

ds = load_dataset("robworks-software/k12-science-standards")

Splits

Appropriate use

• —Synthetic instruction-tuning data where fluent, on-topic, low-variety text is acceptable.
• —Format and schema experiments: task routing, metadata-conditioned generation.
• —Structural analysis over the standards taxonomy itself.

Inappropriate use

• —Classroom or student-facing material. Nothing here was reviewed by an educator.
• —A source of official standard text. Go to the issuing body.
• —Benchmarking model knowledge of standards. Low text variety lets a model score well by learning templates rather than content.
• —Training a model whose output will reach students without human review.

Limitations

• —12-14% distinct text - among the most repetitive datasets in this collection.
• —Not NGSS performance expectation text. NGSS publishes specific performance expectations; this generates prose around NGSS codes and dimension names. Official NGSS material is at nextgenscience.org.
• —"Three-dimensional learning" flags are structural, set because the generator attached a practice and a crosscutting concept - not because the item was designed or reviewed against the three-dimensional framework.
• —`metadata_estimated_time` is generated, not measured or piloted.
• —No educator or subject-matter review at any stage.
• —`quality_score`, `difficulty`, `bloom_level` and `dok_level` are generator assignments, not expert annotations. A high quality score reflects a rule firing, not a judgment about instructional value.
• —Sparse metadata. Many optional metadata columns are nan for most rows.
• —US-centric and English-only.

Source taxonomy

Next Generation Science Standards structure - disciplinary core ideas, science and engineering practices, crosscutting concepts. NGSS codes and dimension names are public; generated text is original output of this project and is not endorsed by the NGSS developers.

License

CC0-1.0 for this compilation and its generated text.

Citation

@dataset{k12_science_standards,
  title  = {K-12 Science Standards (generated instruction data)},
  author = {Robworks Software},
  year   = {2025},
  publisher = {Hugging Face},
  note   = {Programmatically generated instruction-tuning data; not official standard text},
  url    = {https://huggingface.co/datasets/robworks-software/k12-science-standards}
}