We test the grader on essays that people have already scored with a rubric: 40 college essays scored by their instructors and 28 school essays scored by trained raters. Our estimate missed the real score by 11 % of the maximum on average, about 1.6 points on a 15-point rubric. That is why every score is shown as a wide range.
What it cannot do: know your instructor, read the assignment sheet you did not paste, or tell a 3.0 from a 3.5. Read the number as a rough band. Read the evidence (which criterion, which sentence, what is missing) to decide what to fix before the deadline.
Every score in the report is labelled a prediction and shown as a range.