generate question types for each lo and manually choose question type…

[graceQUE03]

Bloom-level question type selection

Lets instructors control, per Bloom's Taxonomy level on a granular learning objective, which question type(s) get generated and how many of each, instead of the previous single "total question count" per objective with the type resolved implicitly from a course-wide Bloom→type preference.

Summary

On the Question Generation → Create Objectives step, each granular objective's Bloom chips (Remember, Understand, Apply, Analyze, Evaluate, Create) now expand into a per-type breakdown when clicked: Multiple Choice, Fill-in-the-Blank, Calculation, Open-Ended, each with its own -/count/+ stepper. The AI objective-generation pipeline also proposes an initial breakdown per Bloom level, which the instructor can review and adjust before generating questions.

UI changes

• client/src/pages/question-generation/BloomTypePanel.jsx (new): the per-Bloom-level type breakdown panel. Renders the four question types horizontally, each with its own count stepper (max 3 per type).
• ObjectiveGroupCard.jsx: clicking a selected Bloom chip now opens/ closes its BloomTypePanel instead of deselecting the level. Multiple panels can be open at once, stacking in the order they were opened. A small × on the chip removes the level (replacing the old click-to-deselect). Once an item has any per-type counts, its total "Questions" count becomes a derived, read-only badge instead of the old manual stepper.
• ObjectivesStep.jsx: new removeBloomChip (level removal, prunes its type counts) and changeTypeCount (per bloomLevel/questionType count, recomputes the item's derived total) state functions. Loads/saves the new questionTypes field alongside bloomTaxonomies wherever granular objectives are read from or written to the API.
• QuestionGeneration.jsx: validateStep1 now also blocks advancing if a selected Bloom level has a type breakdown with a zero total.
• AIGenerateModal.jsx: passes the AI's suggested questionTypes through when saving a generated objective.

Backward compatible by default: granular objectives with no `questionTypes (anything created before this change) keep rendering with the original single "Questions" stepper.

Data model

New field questionTypes on a granular objective, added alongside the existing bloomTaxonomies array (not replacing it) — a flat array of { bloomLevel, questionType, count } triples:

{
  "text": "Explain the role of enzymes in catalysis.",
  "bloomTaxonomies": ["Understand", "Apply"],
  "questionTypes": [
    { "bloomLevel": "Understand", "questionType": "multiple-choice", "count": 2 },
    { "bloomLevel": "Apply", "questionType": "calculation", "count": 1 }
  ]
}

A flat array (rather than an object keyed by Bloom level) was chosen because it maps directly onto JSON Schema structured output and is trivial to validate/group client-side.

Backend changes

• src/constants/llm-schemas.js: OBJECTIVES_SCHEMA gains questionTypes on each granular objective (enum-constrained bloomLevel and questionType, count 1–3).
• src/constants/app-constants.js — both objective-generation prompts (OBJECTIVE_GENERATION_AUTO_PROMPT, OBJECTIVE_GENERATION_MANUAL_PROMPT) gained an instruction + guidance table asking the model to recommend question type(s) and counts per Bloom level, plus a matching self-check.
• src/controllers/rag-llm.js (generateLearningObjectivesHandler): cleans/validates the LLM's questionTypes output (drops entries with an invalid bloom level or type; any Bloom level left with none gets a default type via the existing DEFAULT_BLOOM_TYPE_PREFERENCES, so the UI never shows an empty breakdown for a selected level).
• src/services/objective.js: createObjective/updateObjective persist questionTypes on the granular objective document (Mongo, grasp_objective collection), same treatment as bloomTaxonomies.

Wiring into actual question generation

generateQuestionsWithRagHandler (POST /api/rag-llm/generate-questions-with-rag) now accepts an optional questionTypes array. When present, it expands the triples into an explicit (bloomLevel, questionType) work list, and replaces the old "round-robin Bloom levels for N questions, resolve type via course preference" logic for that request. When absent, generation falls back to the previous behavior unchanged.

client/src/pages/question-generation/generationApi.js (generateQuestions) passes a granular item's questionTypes through when present.

sample response from LLM

  "materialIsRelevant": true,
  "relevanceReason": "The provided course materials cover various topics in calculus, including sequences and series, Taylor's theorem, and Maclaurin series.",
  "objectives": [
    {
      "name": "Understand the concept of sequences and series",
      "granularObjectives": [
        { "text": "Students will be able to define and distinguish between a sequence and a series.", "bloomTaxonomies": ["Remember", "Understand"] ,
          "questionTypes": [ {"bloomLevel": "Remember", "questionType": "multiple-choice", "count": 1}, {"bloomLevel": "Understand", "questionType": "fill-in-the-blank", "count": 2} ]},
        { "text": "Students will be able to identify the terms and notation used in sequences and series.", "bloomTaxonomies": ["Remember", "Understand"] ,
          "questionTypes": [ {"bloomLevel": "Remember", "questionType": "multiple-choice", "count": 1}, {"bloomLevel": "Understand", "questionType": "fill-in-the-blank", "count": 2} ]},
        { "t...```