German writing, graded to the point.

Writing is the one skill you can't self-assess, and the one you get the least feedback on — a teacher can't mark an essay a day for you, and most apps don't try. So the writing tool on Crosshire Learn does exactly that, and only that. You photograph the page — or type it, or upload a PDF — and it reads the whole text and marks it the way a Goethe, telc or DTB examiner would: a score on each of the four criteria, the reason behind every mark, each mistake fixed inside your own words, and a model answer rebuilt from what you actually wrote. Rather than describe it, here is exactly what it did with one real essay.

You're not filling in blanks. You pick a prompt, set your target level, and hand over a finished text in whatever form it already exists. Three ways in:

Two settings frame the grade. You choose the target level — the run below was set to B2 Beruf (DTB) — and a mild or strict mode, so the same text can be marked gently for encouragement or held to the full exam line. The marking then comes back in whatever language you set for feedback: the learner below chose English, so the notes are in English while every German phrase they quote stays German.

Every text is marked on the four criteria a real B2/C1 exam uses — each out of five, on the half-point — and given one overall CEFR verdict. No private rubric, no extra axes, and the same grader behind every chapter exercise and every free essay.

One learner wrote a workplace statement by hand — should everyone take part in a training they pay for themselves, held during work hours? — photographed the page, set the level to B2 Beruf, and pressed grade. Here is the whole report, start to finish. Every number below is exactly what the engine returned; only the learner's name has been taken out.

Most tools give you a figure and leave you to trust it. This one argues for it — and, unusually, tells you why the mark is not a point lower and not a point higher. Two of the real justifications from this essay, verbatim:

The grammar mark comes with its receipts too. Rather than a feeling, the tool lists every error it found and counts them by class, so you can see exactly where the points went:

And none of this rests on the model behaving itself. Every correction you see is a verbatim quote from your own text — anything the grader can't point to in what you actually wrote is discarded before it reaches you, so it can't invent a mistake you didn't make. The jump from error-count to points is a fixed ladder applied on our side, the same for everyone. The overall CEFR verdict is recomputed from the four scores, not taken on the model's say-so. The model reads and reports; the arithmetic is ours.

It doesn't rewrite your work out from under you. Each mistake is corrected where you made it and tagged by type, so the same slip is easy to recognise next time. The full set from this essay:

Two things worth noticing. It separates errors that cost points from optional polish — stylistic suggestions like Anderseits → Andererseits are filed as hints and never touch the score, so you always know what actually mattered. And vocabulary is checked by counting, not scolding: it flagged Weiterbildung four times and Firma four times, offered the swaps — Fortbildung, Schulung, Qualifizierungsmaßnahme for the first, Unternehmen, Betrieb, Arbeitgeber for the second — while crediting the business lexicon already there (finanzielle Belastung, Geschäftsleitung, Belegschaft, Akzeptanz). The instruction is precise: vary two repeated nouns so no key word appears more than twice, and that is the +1.

Last comes a model version — and this is the part most tools get wrong. It is not a generic sample pulled off a shelf, and it is not pitched two levels above you to admire from a distance. It is your essay, rebuilt at your target level, keeping your points and your argument. The opening the tool returned:

Liebe Kolleginnen und Kollegen, habt ihr schon gehört, dass alle Mitarbeitenden unserer Firma an einer Weiterbildung teilnehmen sollen? Die Kosten sollen wir selbst tragen, die Schulung findet jedoch während der Arbeitszeit statt. Ich möchte mich kurz dazu äußern, denn ich sehe sowohl Vorteile als auch Nachteile.

The informal Hallo an alle becomes a workplace greeting; the tangent is folded into the argument; the repeated nouns are varied. The length is capped to the level you were graded against — an A1 learner is never shown a C1 model — and the framing is deliberately modest: the tool labels it a version built from your text, for comparison, not the only correct answer. It shows you a reachable target in your own voice, not a verdict on it.

Written from the live tool — the scores, corrections, tally and model answer above are one real grading run, reproduced as returned. The learner's name has been removed; nothing else is changed. — Crosshire.