AssessmentNo gimmicks. No gaming.

Practice and assessment are not afterthoughts.

Practice and assessment are not afterthoughts. Every lesson concludes with a generated practice set across five question formats, governed by explicit anti-gaming rules in the generation prompt and graded server-side. Distractors are misconception-probing; option lengths are normalised; explanations are 60–120 words covering both why the correct answer is right and why each distractor is wrong.

Five question formats

What the generator can build.

  • Multiple choice

    Single correct answer chosen from 4–5 plausible options.

  • True / false

    Single-step claim with explanation behind both poles.

  • Fill in the blank

    Constrained free-text gap-fill with accepted-answer normalisation.

  • Multi-select

    Multiple correct answers chosen from a list — partial credit configurable per item.

  • Matching

    Left-column items paired against right-column targets; order randomised.

Five rules against gaming

What the generation prompt forbids.

  • Length-normalised options

    All options within 5–10 characters of each other. The longest-option-wins heuristic dies on contact with this rule.

  • Correct-answer placement randomised

    The correct answer is not consistently the longest, not consistently in position A, not consistently in position D.

  • Misconception-probing distractors

    Distractors are written to surface specific misunderstandings — common conflations, off-by-one errors, inverse relationships — not nonsensical filler.

  • Bloom-aware difficulty mix

    A practice set covers a spread of Bloom’s-taxonomy levels: remember, understand, apply, analyse. The mix is configurable per pack.

  • Server-authoritative grading

    Client-submitted answers are graded against the canonical key on the server. Client scoring is for UX only.

Explanations do the heavy lifting.

Every question generates a 60–120-word explanation. The explanation has three jobs: establish why the correct answer is right, explain why each incorrect option is wrong, and reinforce the underlying concept with a concrete example. That is the moment of learning — not the answer-submission moment. The system enforces the structure in the generation prompt.

Practice accuracy is computed at the lesson level and persisted as the mastery signal that feeds tomorrow’s lesson generation. The loop is closed.

Code is truth
  1. 01

    Five formats supported: multiple-choice (MCQ), true/false, fill-blank, multi-select, matching.

  2. 02

    MCQ generation rules in the prompt: ALL options NEARLY IDENTICAL LENGTH (within 5–10 characters); the correct answer must NOT be consistently the longest option; distractors must probe common misconceptions, not be nonsensical.

  3. 03

    Explanation discipline: every question generates a 60–120-word explanation covering (1) why the correct answer is right, (2) why each incorrect option is wrong, (3) the underlying concept reinforced with a concrete example.

  4. 04

    Server-authoritative grading: client-submitted answers are graded on the server against the canonical key; client cannot manipulate scoring.

  5. 05

    Practice analytics: per-lesson accuracy is computed on final answers and persisted to the lesson record; that accuracy is the mastery signal that adjusts the next day’s generation.