Every score on this site comes from AI.

Every submission you send — for Questions 51 through 54 — is graded by Anthropic’s Claude, not a human. The three rubric scores (content, structure, language), the written feedback, and the corrections you see are all AI output.

Short-answer questions (51, 52) are graded by Claude Haiku 4.5. Long-form writing (53, 54) is graded by Claude Sonnet 5, and its output also feeds question 51 in some flows. See exactly which fixtures and models were tested on the grading verification page.

Your submitted text and the official TOPIK II rubric are sent to Claude for grading. Where relevant, your past recurring mistake patterns are also included so feedback can call out repeated errors. No data beyond what’s needed to grade the current submission is sent.

• Scores are practice estimates, not official TOPIK results, and do not predict one.
• Sseudam is an independent study tool, not affiliated with the National Institute for International Education or the TOPIK examination.
• Grading has measurable variance — re-submitting the same essay can shift a borderline score by a few points. That’s measurement noise, not a change in your writing.

We test our own grader against essays with known outcomes — including one written to instruct the grader to give full marks — and publish the results, pass or fail, on the verification page. If you want the receipts, not just the claim, that’s where they are.

No human reviews or edits your feedback before you see it. If a score looks wrong, treat it as a signal to revise, not a verdict — and see the limits above.