How we score your writing
Honest scoring is the whole product. Here is exactly what happens to your text and how far you can trust the number.
1. Word limits first
Your text is counted the way the paper counts words. Outside the limits, the exam gives « A1 non atteint », so we tell you and use no credit.
2. Five criteria on the CEFR scale
An AI examiner, instructed with the official criteria and the CEFR descriptors, rates task fit, coherence, development, vocabulary, and grammar & spelling from A1 to C2, each with its reason.
3. The mark is computed, not guessed
We average the five levels and place the result in France Éducation international's bands: A2 = 2–5, B1 = 6–9, B2 = 10–13, C1 = 14–17, C2 = 18–20. Then the IRCC table gives the NCLC level.
4. Every correction is checked
Each correction must quote your exact words; anything that doesn't match your text is dropped before you see it.
The five criteria
- Task fit
- Every element the instructions ask for, the right text type and register (tu/vous, greeting, sign-off).
- Coherence
- Organisation, paragraphs, linking words, pronouns and punctuation that guide the reader.
- Development
- Ideas, examples, explanation and argument, with nuance for the task's level.
- Vocabulary
- Range, precision and fit of words and expressions; repetition and anglicisms.
- Grammar & spelling
- Agreement, verb forms, tenses and moods, prepositions, word order, spelling and accents.
Mark out of 20 → CEFR → NCLC
| Mark /20 | CEFR | NCLC |
|---|---|---|
| 2 | A2 | < 4 |
| 4 | A2 | 4 |
| 5 | A2 | 4 |
| 6 | B1 | 5 |
| 7 | B1 | 6 |
| 9 | B1 | 6 |
| 10 | B2 | 7 |
| 11 | B2 | 7 |
| 12 | B2 | 8 |
| 13 | B2 | 8 |
| 14 | C1 | 9 |
| 15 | C1 | 9 |
| 16 | C1 | 10 |
| 18 | C2 | 10 |
| 20 | C2 | 10 |
What an estimate cannot do
- It is not an official result. Only France Éducation international marks the TCF.
- Two human examiners can differ by a band; so can we. Watch your trend over several texts.
- Handwriting, the exam room and time pressure are not in the simulator.
We test the grader every month: our B2 model answers must land between 10 and 13, and weaker versions of the same texts lower. Results are logged with the model used.