Skip to main content
Diagnostic assessments are Sprout’s way of understanding what you already know before teaching you. They’re not tests — they’re conversations that help Sprout personalize your learning path.

​
How Diagnostics Work

Every concept in your learning path has a diagnostic assessment that measures your existing knowledge. Based on your answers, Sprout adapts the subconcept graph to fill gaps and skip material you’ve already mastered.

​
The Diagnostic Flow

1

Diagnostic questions are generated

When the Subconcept Bootstrap Agent runs for a concept, it automatically creates 5-10 diagnostic questions (or 2-3 in small mode).Questions are strategically designed with:
  • Mixed formats: Multiple choice (MCQ) and open-ended questions
  • Mixed difficulty: Ranges from 1 (basic) to 5 (advanced)
  • Targeted coverage: Each question maps to specific knowledge gaps
2

You answer the questions

Submit your answers via the API:
POST /api/assessments/:assessmentId/answers
{
  "userId": "your-user-id",
  "questionId": "question-id",
  "answerText": "Your written response",     // For open-ended
  "selectedOption": "B",                      // For MCQ
}
The system tracks:
  • Your answer text or selected option
  • Timestamp of submission
  • Auto-completion when all questions are answered
3

Answers are graded

The Concept Refinement Agent uses a specialized grading sub-agent to evaluate your responses:
  • MCQ questions: Automatically compared to the correct answer
  • Open-ended questions: Evaluated semantically by Claude for conceptual understanding
  • Partial credit: Scored on a 0-1 scale (not just right/wrong)
  • Detailed feedback: One-sentence explanation for each answer
4

Learning path adapts

Based on your diagnostic results, the Concept Refinement Agent personalizes your subconcept graph using the Observe-Reason-Act-Verify loop.

​
Question Formats

​
Multiple Choice Questions (MCQ)

MCQ questions test specific factual knowledge: 
{
  "format": "mcq",
  "prompt": "Which operation preserves the solution set of a linear system?",
  "options": ["A) Multiplying a row by zero", "B) Swapping two rows", "C) Adding a variable"],
  "correctAnswer": "B",
  "difficulty": 2
}
Grading is automatic: your selectedOption is compared to correctAnswer.

​
Open-Ended Questions

Open-ended questions evaluate deeper understanding: 
{
  "format": "open_ended",
  "prompt": "Explain why row operations don't change the solution set of a linear system.",
  "correctAnswer": "Row operations create equivalent systems by applying reversible transformations...",
  "gradingRubric": {
    "key_concepts": ["reversibility", "equivalent systems", "same solution set"],
    "partial_credit": true
  },
  "difficulty": 3
}
Grading uses Claude to:
  • Compare your answer semantically to the expected answer
  • Check for key concepts (doesn’t require word-for-word match)
  • Assign partial credit (e.g., 0.3 if you identified the right concept but confused details)
  • Provide constructive feedback
The grading agent is designed to be generous with partial credit. Students don’t need to match expected answers word-for-word — they just need to demonstrate understanding.

​
The Observe-Reason-Act-Verify Loop

After you complete a diagnostic, the Concept Refinement Agent runs through a structured process:

​
1. GRADE

Call grade_student_answers to grade all responses using the grading sub-agent.

​
2. OBSERVE

  • Call get_current_subconcepts to view the existing subconcept DAG
  • Call check_student_history to analyze your cross-concept performance, mastery scores, and overall level

​
3. REASON

Claude analyzes:
  • Knowledge gaps from low-scoring diagnostic questions
  • Misconceptions revealed in feedback
  • Mastery patterns (e.g., strong fundamentals but weak on applications)

​
4. ACT

The agent modifies your learning graph:
ToolPurpose
add_subconceptInsert bridge/remedial subconcepts for knowledge gaps
remove_subconceptRemove subconcepts you’ve already mastered (reconnects edges automatically)
add_prerequisite_conceptInsert an entire concept BEFORE the current one if you’re missing foundations
add_followup_conceptInsert a concept AFTER for reinforcement or enrichment

​
5. VERIFY

Call validate_graph to check for:
  • Orphaned nodes (no incoming or outgoing edges)
  • Broken edges (missing nodes)
  • Unreachable nodes (not connected to the main path)

​
6. VERIFY AGAIN

Confirm the graph is clean and ready for learning.
This is Sprout’s star feature — autonomous agents that observe, reason, act, and verify their own changes to create a personalized learning path.

​
Diagnostic Summary Data

After grading, the system generates a diagnostic summary: 
{
  assessmentId: string,
  answeredCount: number,
  totalQuestions: number,
  overallScore: number | null,  // Average score 0-1
  strengths: string[],          // Prompts you scored ≥0.7 on
  gaps: string[],               // Prompts you scored <0.7 on
  items: [
    {
      questionId: string,
      prompt: string,
      format: "mcq" | "open_ended",
      difficulty: number,
      studentAnswer: string | null,
      correctAnswer: string | null,
      isCorrect: boolean | null,
      score: number | null,  // Normalized 0-1
      feedback: string | null
    }
  ]
}
This data is used by the Concept Refinement Agent to make intelligent decisions about your learning path.

​
Viewing Your Diagnostic

Retrieve diagnostic status for a concept: 
GET /api/agents/concepts/:conceptNodeId/diagnostic?userId=your-user-id
Response includes:
  • Assessment metadata
  • All questions
  • How many you’ve answered
  • Whether the assessment is complete

​
Triggering Refinement

Once you’ve answered diagnostic questions, trigger the Concept Refinement Agent: 
POST /api/agents/concepts/:conceptNodeId/run
{
  "userId": "your-user-id"
}
The agent will:
  • Bootstrap subconcepts if they don’t exist yet
  • Grade your diagnostic answers (even if only partial)
  • Refine the subconcept graph based on your performance
  • Stream all changes via SSE events
You can run refinement even with zero answers. The agent will work with whatever data is available, though it’s most effective when you’ve completed the diagnostic.

​
Next Steps

After your diagnostic is graded and your learning path is refined:
  1. Review the updated subconcept graph to see what was added or removed
  2. Start tutoring sessions for each subconcept
  3. Track your mastery scores as you progress

Build docs developers (and LLMs) love

Get started for freeTalk to us