Student help is more controllable when it uses the lesson currently being taught. The app should know which source range and teacher-approved material are active rather than treating every imported document as equally relevant.
Grounding is an evidence relationship, not a promise of perfect answers. Define what the assistant may explain, when it should ask for clarification, and when the lesson does not support an answer.
Keep lesson context explicit
Record the active lesson identity, source pages, approved explanations, relevant vocabulary, and any teacher-defined constraints. Keep private teaching notes and student information out of the help context unless an authorised design explicitly requires them.
Select evidence relevant to the question and preserve its source reference. Avoid silently switching to another chapter because a similar phrase ranks well in retrieval.
Test a fictional classroom explanation
Suppose the lesson compares two materials using a short authorised passage. A student asks why one example behaves differently. The assistant should explain within the source and taught scope, identify missing evidence where needed, and avoid inventing an experiment result.
Test a direct question, a paraphrase, an ambiguous pronoun, a question from the next chapter, and a question not answered in the material. Add a source passage containing instruction-like text and confirm it is treated as material to analyse rather than authority to change the application's rules.
Review responses for factual support, age-appropriate wording, preserved conditions, and useful uncertainty. A fluent answer outside scope should not pass simply because it sounds educational.
Give teachers practical control
Make the active lesson visible and allow the teacher to stop or redirect the interaction. Preserve the original source and distinguish generated help from teacher-approved content.
If interactions are stored, define access and retention deliberately. Do not collect identifiable student details merely to make an example or log easier to diagnose. The help workflow should not become an unreviewed student-record system.
The tradeoff is narrow relevance versus breadth. A lesson-bound assistant may decline questions it could answer generally, but that limit can make classroom use easier to review. Broader help requires its own scope, evidence, and teacher expectations.
The ChalkLens portfolio account documents Student Help tied to lesson context. The material-comparison example here is fictional and does not claim measured student outcomes or perfect grounding.
Questions about lesson-grounded help
Does selecting the lesson guarantee supported answers?
No. Evaluate retrieved evidence and generated claims. Scope selection is one boundary, not complete accuracy proof.
Should private teacher notes be included automatically?
No. Separate teacher-only material from student-facing context and test that boundary explicitly.
What should a pilot review test first?
Ask one supported question, one ambiguous question, and one outside-scope question from the same lesson screen. Bring unsupported answers to an education assistant review.