ABOUT · SCORING
The lead TOEFL instructor at MasterPrep, Korea's #1 study-abroad academy, designed every question and the scoring engine from ETS's published research. Here's exactly how your work gets graded.
Every question and every score starts from ETS's published research and rubrics.
You don't just get a number — error rates, vocabulary, development. You see exactly where points went.
Know you're ready before test day. Catching weaknesses early means fewer paid attempts.
ETS grades Writing and Speaking with humans and machines together. Machines measure language and delivery precisely; humans judge content.
Grammar, usage and mechanics error rates, development, and lexical level come from e-rater; ideas and logic from the human rater. ETS reports human–machine agreement of r=0.86 — on par with human–human (0.85).
Speaking rate, pauses, pronunciation, and lexical/grammar metrics come from SpeechRater; the content of your answer is judged by the human rater. Human–machine agreement r=0.89.
Agreement figures are from ETS's published 2026 iBT Technical Manual.
Each section implements the axes the real rubric demands. What we measure is what you fix.
Algo Instant Scoring is free and unlimited. Algo Precision AI Scoring uses credits, only when you need it.
The two tiers map onto the real test's two raters. Algo Instant Scoring measures what can be counted, as e-rater does — grammar, usage and mechanics error rates, lexical level and diversity, sentence length and syntactic range, length, bullet fulfillment on email, and off-topic signals. Algo Precision AI Scoring reads and judges, as a human rater does — whether you answered the question, whether the argument holds, whether the examples fit.
Neither score is an official ETS score; both are estimates for practice.
The real SpeechRater excels at measuring delivery, but judging the content of your answer remains the human rater's job. Algo Precision AI Scoring delivers both at once — machine-grade delivery analysis plus the evaluation a human rater would give.
The engine is designed against ETS's published research literature, with every metric implemented to its published definition.
Mock results include an estimated band on the 1.0–6.0 scale. The anchor is the band-to-legacy-score concordance ETS itself published. We convert your section result to a legacy-scale (0–30) equivalent, look up the band in that table, and compute the overall score with the same rule as the real test — the average of the four sections, rounded to the nearest 0.5.
In Reading and Listening, attempts routed to the easier second module are capped at 4.5: that path contains no upper-level content, so there is simply no opportunity to demonstrate C1-level (5.0+) ability. Reaching the upper module opens the full range to 6.0.
This is an estimate. The real test applies undisclosed statistical equating per form, and even the official test has a measurement error of roughly ±0.3 bands. That is why the estimated band is always shown with ±0.5 and never replaces an official ETS score.