"Question sourcing" selector with three modes Manually created — you write every question; no AI generation anywhere. Both (recommended, default) — your authored questions run the formal exam, while t...
Upload a cover image in Admin Settings > System Manuals to replace this training placeholder.
"Question sourcing" selector with three modes
Manually created — you write every question; no AI generation anywhere.
Both (recommended, default) — your authored questions run the formal exam, while the AI keeps a shared practice pool topped up from your lesson topics. Generation only fires when a learner's suitable difficulty band runs dry — either they're struggling (need easier items than exist) or too experienced (outgrew/exhausted the set) — with a 10-minute per-course cooldown and a 120-questions-per-category cap.
Fully AI-generated — the AI builds the whole question bank from course content, and each formal attempt randomly samples a fresh set (you choose how many per attempt, default 20, category-balanced). Grading covers exactly the served sample including skipped items, and the results breakdown reviews only what that attempt saw.
How the pool economics work, per your design: generated questions are stored once and shared across all learners — one generation amortizes over the whole cohort, and pool generation is recorded as NON_BILLABLE usage (tracked in the admin AI usage dashboard under feature practice_pool_generation, but never debits a learner's credits). Pool questions inherit the entire adaptive machinery — IRT difficulty calibration from real answers, spaced repetition, the AI tutor, and the weak-spots archive — and the exam list page shows each exam's pool size next to its authored count.
To see it in action: open any exam in the builder, leave sourcing on "Both", and run a practice session as a learner — once you exhaust the authored questions near your level, the pool generates its first batch automatically.