A lesson draft can use a curriculum term fluently while changing its intended meaning. Test vocabulary within the actual source and teaching objective rather than judging whether the model recognises the word in isolation.
Build a small reviewed set of terms, definitions, examples, and confusable alternatives. Keep the source edition and page context so a disagreement can be traced to evidence rather than preference.
Define meaning and acceptable explanation
For each term, record the source definition, required conditions, acceptable paraphrases, common confusions, and learner context. Separate a simpler explanation from an oversimplification that removes a necessary distinction.
Include domain-specific names, abbreviations, notation, and language mixtures relevant to the curriculum. Do not assume broad model language support proves reliable use of a particular textbook's terminology.
Test a fictional comparison vocabulary set
Suppose a chapter distinguishes two similarly named processes. Ask for a short explanation, a contrast, an example constrained to the passage, and a quiz item that requires choosing between them.
Create controlled fixtures with the correct term, a near-neighbour spelling, an OCR error, and an ambiguous pronoun. Check whether the draft preserves the distinction, flags uncertainty, or invents a meaning. Review distractors and answer keys separately.
Use authorised source material and teacher judgement for the reference answers. This is a proposed evaluation matrix, not a claim that a named curriculum or model has been tested successfully.
Separate extraction from generation failures
If recognition changed the term before the model saw it, retain that evidence. A prompt change cannot reliably repair missing source meaning. Where the source is correct but the draft substitutes a more familiar term, classify the generation defect separately.
Keep approved vocabulary available to the task in a bounded way, with source references and clear scope. A glossary can help but should not override the actual passage or become an unsupported authority for unrelated questions.
The tradeoff is vocabulary support versus review and maintenance. A detailed glossary can improve consistency while requiring updates across editions and lessons. Start with terms whose confusion materially changes the learning objective.
After model, prompt, or extraction changes, rerun the same reviewed cases before attributing an improvement. A fluent new phrasing may still lose the distinction the test was designed to protect.
The ChalkLens case study documents source-based lesson generation and OCR. This vocabulary exercise is fictional and does not publish curriculum-specific accuracy or learning outcomes.
Questions about curriculum vocabulary tests
Is a dictionary definition enough?
Not always. The curriculum's context, scope, and required distinctions may differ from a broad definition.
Should simpler wording be rejected?
No. Accept clear paraphrases that preserve required meaning. Reject simplifications that remove conditions or merge distinct concepts.
What should a pilot review inspect first?
Inspect a small set of high-consequence confusable terms across extraction, explanation, quiz, and key. Bring recurring substitutions to an education evaluation review.