PrepAmigoHow Prepamigo Scores Your Writing
Trillion-Parameter LLM + Grammar Analysis: A Dual-Engine Approach to Professional Writing Assessment
Quick Navigation
Writing is one of the most challenging skills to self-assess. How do you know if your essay truly meets CELPIP standards? Without professional feedback, you're left guessing about your real level. Prepamigo's writing scoring system was built to solve this problem—we use a trillion-parameter large language model combined with a specialized grammar analysis model to deliver examiner-quality scoring and actionable feedback.
Our Philosophy
Accuracy First: We're Responsible to Every Test-Taker
Many AI writing tools promise instant results—but those fast scores are often inflated, giving you a false sense of confidence. When you practice with unrealistic feedback, you walk into the real exam unprepared, wondering why your actual score is lower than expected.
We made a deliberate choice: accuracy over speed. Our scoring takes longer because we use more sophisticated models that genuinely understand the nuances of language. We believe every test-taker deserves to know their true level—not a feel-good number that sets them up for disappointment.
What We Avoid
- • Inflated scores that don't reflect reality
- • Generic feedback that doesn't help you improve
- • False confidence that leads to exam-day surprises
- • Shortcuts that sacrifice accuracy for speed
What We Deliver
- • Honest scores aligned with CELPIP standards
- • Specific, actionable improvement suggestions
- • Clear understanding of your real ability level
- • Reliable preparation you can trust
What We Evaluate: Four Scoring Dimensions
Our scoring dimensions are strictly based on the official CELPIP Score Comparison Chart. This ensures your practice scores accurately reflect how real CELPIP examiners evaluate written responses.
Each writing response is evaluated across four core dimensions, scored from 4-12 (CELPIP Writing does not include Level 3):
Content & Coherence
Does your response address the prompt completely? Are your ideas logically organized with clear transitions? Do your paragraphs flow naturally from one to the next?
Vocabulary
Are you using varied and precise vocabulary? Can you employ advanced expressions and collocations appropriately? Is your word choice accurate and natural?
Readability
A comprehensive evaluation of grammar accuracy, sentence structure variety, punctuation, and spelling. This is where our grammar analysis model plays a critical role—detecting errors that affect how easily your writing can be understood.
Task Fulfillment
Have you addressed all parts of the task? Is your tone appropriate for the context (formal email vs. informal message)? Have you met the word count requirements?
Our Dual-Engine Scoring Architecture
Just like our speaking assessment, writing scoring requires multiple specialized engines to achieve examiner-level accuracy:
Trillion-Parameter LLM
Handles semantic understanding, content analysis, coherence evaluation, vocabulary assessment, and task fulfillment checking.
Responsibilities:
- • Content relevance and completeness
- • Logical flow and coherence
- • Vocabulary sophistication
- • Task requirement fulfillment
Grammar Analysis Model
A specialized model dedicated to detecting grammatical errors, spelling mistakes, punctuation issues, and sentence structure problems.
Responsibilities:
- • Grammar error detection
- • Spelling accuracy checking
- • Punctuation analysis
- • Sentence-level corrections
💡 Why Both Engines Are Necessary
A large language model excels at understanding meaning and context, but it can sometimes overlook mechanical errors. The grammar analysis model is purpose-built for precision in error detection. By combining both, we catch both content issues AND technical mistakes—the same comprehensive evaluation a professional examiner would provide.
How Your Writing Gets Scored
Dual-Engine Analysis
Your response is simultaneously processed by both the trillion-parameter LLM (for content) and the grammar analysis model (for technical accuracy).
Dimension Scoring
Each of the four dimensions receives an individual score. Grammar analysis results directly influence the Readability dimension.
Penalty Rules Applied
Score caps are applied for off-topic responses, incomplete tasks, or insufficient word count—just like real examiners would.
Final Score & Feedback
You receive a standardized 4-12 score plus detailed feedback: grammar corrections, vocabulary suggestions, and personalized improvement recommendations.
Frequently Asked Questions
❓"I thought I wrote well, but why did I only get a 7?"
This is one of the most common questions we receive. There are several reasons why your score might be capped even when you feel your writing was good:
Task Requirements Not Fully Met
If the prompt asked you to address specific points (e.g., "explain why AND suggest alternatives") and you only covered part of it, the AI will detect this as incomplete task fulfillment—capping your score at 7.
Off-Topic Response
If your response doesn't address the actual prompt, even beautiful English will be capped at 5.
Insufficient Word Count
Writing significantly fewer words than required (less than 75% of the target) indicates incomplete development and will limit your score.
Hidden Grammar Issues
Sometimes errors feel natural to you but are technically incorrect. Our grammar analysis model catches these subtle mistakes that affect readability.
Remember: A score of 7 is still above average! Check your detailed feedback to understand exactly what to focus on for improvement.
⏱️"Why does scoring take so long?"
We understand the wait can feel long. Here's why our scoring takes more time than some alternatives:
Trillion-Parameter Model
Processing through a massive language model requires more computation than lightweight alternatives—but this is what enables deep semantic understanding.
Dual-Engine Processing
Running both content analysis AND grammar checking simultaneously takes additional time—but delivers comprehensive results.
Detailed Feedback Generation
We don't just produce a number—we generate line-by-line corrections, suggestions, and an improved version of your response.
Quality Over Speed
We could use faster, smaller models—but they produce inflated scores. We chose accuracy over convenience.
Think of it this way: A real CELPIP examiner doesn't grade your essay in 2 seconds. Our system takes time because it's doing what a careful human evaluator would do—analyzing content, checking grammar, and preparing helpful feedback.
⚖️"Other tools score faster—why should I wait?"
Fast doesn't mean better. Here's what happens with lightweight models (GPT-3.5 class and similar):
Inflated scores typically 1-2 levels higher than your actual ability
Missed errors especially in complex sentences and nuanced vocabulary usage
Generic feedback that doesn't help you understand your specific weaknesses
False confidence leading to exam-day disappointment
Would you rather feel good today and struggle on exam day? Or would you prefer honest feedback now that helps you genuinely improve?
Help Us Improve
Your Feedback Makes Us Better
Our scoring system is constantly evolving. We analyze feedback from thousands of users to improve our accuracy and provide better insights. If you notice something that doesn't seem right—a score that feels off, feedback that's unclear, or an error we missed—we want to know about it.
Every piece of feedback helps us train our models to better serve all CELPIP test-takers. You're not just practicing for yourself—you're helping improve the experience for future users.
Ways to share your feedback:
- 📧Email us at support@prepamigo.com
- 💡Use the feedback button after receiving your score
- 🎯Include your response and tell us what you expected—this helps us understand the gap