Skip to main content

​
Model selection is managed by LearnTerms

Question Studio currently uses a centrally managed LearnTerms generation model. There is no model picker in the curator workflow. This keeps the source, planning, drafting, checking, usage, and telemetry contracts consistent across cohorts. LearnTerms can update the underlying provider configuration without asking curators to understand vendor model names.

​
What you control

Instead of selecting a model, you control the parts that most directly shape the questions:
  • Pages or Topics — the evidence the agent may use
  • Question count — from 1 to 30 per run
  • Question mix — Learn, Clinical, and Critical thinking
  • Focus notes — priorities, misconceptions, or boundaries for the run
  • Destination — the semester, class, and module for saved drafts
  • Curation — editing, selection, saving, and eventual publishing

​
Why question mix matters

Each question type defines an evidence and reasoning contract.
  • Learn asks for direct reinforcement of a source-supported concept.
  • Clinical requires source evidence that supports a relevant clinical application.
  • Critical thinking requires at least two supported relationships that must be combined.
When the source lacks support for a requested type, Question Studio should abstain or cut the candidate.

​
Why source quality matters more than a model label

Missing or garbled extraction produces missing evidence for the generation model. Before tuning focus notes or increasing the batch size:
  1. Check the Content Library preview.
  2. Compare extracted text with the original PDF.
  3. Narrow the page or topic selection.
  4. Match the question mix to the evidence.

​
Operational telemetry

Question Studio records run-level usage such as calls, tokens, duration, and spend when available. These values help monitor reliability and cost. Factual correctness and educational quality require source review.