How Faust marks your writing and speaking
A language model reads your answer and reports facts only: corrections, task points done, moves made. It never gives a band or a point. Fixed code does that, from the official exam criteria. The same answer gets the same score.
Updated 17 October 2026, Faust Academy team
The method
- The model observes
- For writing it corrects each sentence and rates each task point as done, partly done or missing. For speaking it reports which required moves you made and what each of your turns does: suggest, react, ask, answer, decide.
- Code judges
- Fixed rules turn the facts into a band from A to E for each criterion. The rules follow the official criteria. Where the official text is vague, Faust sets its own rule. A is 100 percent of the criterion, B 75, C 50, D 25 and E 0. A change to the rules is a new version.
- Several reads, one vote
- Writing is read 3 times and speaking 5 times, each at temperature 0. A fact stands when most reads agree on it.
- Every point has a reason
- Each lost point links to a rule and, where one exists, to the words that cost it. Each point you earn links to its evidence.
Writing is read by DeepSeek V4.1 Flash. Speaking is read by Gemini 3.8 Flash. Versions: faust-writing-doctrine/2, faust-fact-marker/4, goethe-speaking-facts/2. Every answer is marked by the same fixed rules.
A marked writing sample, Goethe B1 Teil 1
Sample text written by the Faust team. Marked on 16 October 2026 by the live writing marker.
Task
Text
| Criterion | Band | Points |
|---|---|---|
| Aufgabenerfüllung (task fulfilment) | A | 10 of 10 |
| Kohärenz (coherence) | A | 10 of 10 |
| Wortschatz (vocabulary) | A | 10 of 10 |
| Strukturen (structures) | B | 7.5 of 10 |
| Total | 37.5 of 40 |
- Aufgabenerfüllung (task fulfilment): Coverage 100 percent: 3 task points done in full, 0 partly, 3 in all
- Kohärenz (coherence): Linking score 3 (A from 2.5, B from 1.5, C from 0.5)
- Wortschatz (vocabulary): 0 weighted errors per 100 words (A up to 1, B up to 2.5, C up to 5, D up to 8)
- Strukturen (structures): 2.7 weighted errors per 100 words gives B; 9 of 12 sentences without an error gives A
Why points are missing
Strukturen (structures)
- Spelling “Wochenede” → WochenendeWochenende needs the letter n.0.5 points
- Word form “deine” → deinermit takes dative: mit deiner Mutter.1 point
- Word order “Wir könnten am Samstag ins Museum gehen, danach ich koche für dich” → Wir könnten am Samstag ins Museum gehen, danach koche ich für dich.After danach, the verb comes second.1 point
A marked speaking sample, Goethe B1 Teil 1
Sample transcript written by the Faust team. Marked on 16 October 2026 by the live speaking marker.
Task
Transcript
- Partner: Hallo! Wir wollen ja ein Abschiedsfest für Katrin planen. Hast du schon eine Idee, wann wir feiern könnten?
- Candidate: Hallo. Ich denke, am Freitagabend ist gut, weil alle Kollegen dann früher gehen können. Am Wochenende haben viele schon andere Pläne.
- Partner: Gute Idee. Und wo wollen wir feiern? Im Büro oder lieber woanders?
- Candidate: Ich schlage vor, wir feiern im Büro, weil es billig ist. Dann wir brauchen keine Reservierung.
- Partner: Gut, im Büro. Wen laden wir eigentlich ein?
- Candidate: Ja, das ist okay. Wir laden alle Kollegen ein und auch den Chef, wenn er Zeit hat.
- Partner: Das stimmt. Wer kümmert sich um das Essen und die Getränke?
- Candidate: Ich kann einen Kuchen backen. Du kannst Salat und Getränke mitbringen.
- Partner: Gern. Ich mache zwei Salate. Und was schenken wir Katrin?
- Candidate: Ich habe keine Idee.
- Partner: Eine Pflanze wäre schön. Also: Freitagabend, im Büro, ich bringe Salate und Getränke, du Kuchen.
- Candidate: Hmm, mal sehen.
Required moves
- Wann soll das Fest sein?done
- Wo feiert ihr?done
- Wen ladet ihr ein?done
- Wer kümmert sich um Essen und Getränke?done
- Was schenkt ihr der Kollegin?partly done
Interaction
The band asks for 2 own suggestions, 2 reactions, 1 question and 1 joint decision. This transcript has 4 own suggestions, 2 reactions, 0 questions, 0 joint decisions. That meets 4 of 6.
Bands and points
| Criterion | Band | Points |
|---|---|---|
| Aufgabenerfüllung (task fulfilment) | B | 6 of 8 |
| Interaktion (interaction) | B | 3 of 4 |
| Wortschatz (vocabulary) | A | 8 of 8 |
| Strukturen (structures) | A | 8 of 8 |
| Total | 25 of 28 |
Why points are missing
Aufgabenerfüllung (task fulfilment)
- You did not make or discuss a suggestion for the gift. “Ich habe keine Idee”2 points
Interaktion (interaction)
- Ask your partner a question.0.5 points
- Agree on a decision together.0.5 points
Same answer, same score
We marked the same answer many times. The table measures consistency, not agreement with an examiner. Measured on 11 October 2026 on three real practice sessions.
| Case | Score | Runs with the same score |
|---|---|---|
| Strong B1 speaking exam, also with punctuation and capitals removed | 79 of 84 | 10 of 10 |
| Weak B1 part with English words and broken sentences | 10 of 28 | 13 of 152 runs gave 11 |
| A2 speaking exam | 8 of 20 | 11 of 11 |
The earlier flow, where one call picked a descriptor per criterion, moved by up to 3 points when punctuation and capitals were removed.
Agreement with official points, C1 writing
The Goethe-Institut materials we found hold four scored C1 writing samples. Two are in the current modular format. Faust marked both. The points of the examiners stand next to the Faust points.
| Criterion | Examiners | Faust |
|---|---|---|
| Aufgabenerfüllung (task fulfilment) | B, 10.5 of 14 | C, 7 of 14 |
| Kohärenz (coherence) | A, 14 of 14 | A, 14 of 14 |
| Wortschatz (vocabulary) | B, 12 of 16 | C, 8 of 16 |
| Strukturen (structures) | A, 16 of 16 | C, 8 of 16 |
| Total | 52.5 of 60 | 37 of 60 |
| Criterion | Examiners | Faust |
|---|---|---|
| Aufgabenerfüllung (task fulfilment) | B, 7.5 of 10 | D, 2.5 of 10 |
| Kohärenz (coherence) | A, 10 of 10 | B, 7.5 of 10 |
| Wortschatz (vocabulary) | A, 10 of 10 | B, 7.5 of 10 |
| Strukturen (structures) | A, 10 of 10 | C, 5 of 10 |
| Total | 37.5 of 40 | 22.5 of 40 |
Faust marks strictly: on both samples it never gave more points than the examiners. Your score never flatters you.
The texts belong to the Goethe-Institut. They are not printed here.
Read them in the Handbuch Prüfungsziele und Testbeschreibung C1: Teil 1: PDF page 47, marks on PDF page 48, Teil 2: PDF page 49, marks on PDF page 50.
The two older samples
The other two samples come from the 2016 Übungssatz. That format has four criteria scored 4, 5, 5 and 6 points, not the bands A to E. The scale does not map onto the bands, so Faust did not mark them. PDF page 41: 18 of 20, PDF page 42: 20 of 20.
Good to know
- Faust is an independent trainer that marks on the published criteria of the Goethe-Institut, telc and the other boards. The certificate comes from the exam centre.
- Faust marks strictly and never flatters a score. When your Faust score sits above the pass mark, you are well prepared.
- Reading and listening are scored by the answer key the moment you hand in.
What Faust marks
| Board | Schreiben marked | Sprechen marked |
|---|---|---|
| Goethe | A1 to C2 | A1 to C2 |
| telc | A1 to C2 | A1 to C2 |
| DTZ | marked | You practise every Sprechen task in the real exam format, with photo cards and guides. |
| TestDaF | marked | You practise Sprechen with timed recording and playback, as on exam day. |
| DTB | PDF practice | PDF practice |
| ÖSD | PDF practice | PDF practice |
| DSD | PDF practice | PDF practice |
| ÖIF | PDF practice | PDF practice |
Lesen and Hören: scored by the answer key
Frequently asked questions
- Who marks my writing and speaking?
- The Faust marker. A language model reads your answer and reports facts only, for example corrections, task points done and moves made. Fixed rules from the official exam criteria turn those facts into the bands A to E and the points. That is why the score stays stable from run to run.
- How is this different from asking a chatbot to mark a letter?
- Three things differ. First, the same answer gets the same score. Faust reads your text 3 times for writing and 5 times for speaking at temperature 0 and takes the majority. On a strong B1 speaking exam it returned 79 of 84 points on 10 of 10 runs. A chatbot has no fixed rule table and can name a different score each time you ask. Second, Faust uses the official criteria of the exam, with bands A to E for each criterion. A chatbot picks its own scale and its own number. Third, Faust shows the rule and the words that cost you each lost point.
- How close is the Faust score to the real exam?
- Faust marks on the official criteria and point scale of each exam, so your score shows where you stand before exam day. Faust marks strictly and never flatters a score.
- Which exams and parts does Faust mark?
- The table above lists every board and level where Schreiben and Sprechen are marked. Lesen and Hören are scored by the answer key.
- Are the two samples real learners?
- No. The Faust team wrote both texts and labelled them as samples. They show the method on one text each. They are not a quality claim.