How Scores Are Calculated and Checked
Every practice score on the platform follows the exam board's own published rules. Writing and speaking are marked by the AI examiner against the board's own criteria, and the scoring is measured against answers the exam boards themselves have scored.
Objective sections
Reading, listening, maths and verbal answers are marked against verified answer keys:
- The exam's own marking rules. Where an exam uses partial credit or negative marking, so does the practice test. PTE, for example, scores Re-order Paragraphs by adjacent pairs, Write from Dictation word by word, and takes a point off for each wrong choice in multiple-answer tasks, never below zero.
- The exam's own score scale. Raw marks are converted to the official scale: IELTS bands, TOEFL 1–6 bands, PTE 10–90, CELPIP levels, OET 0–500 and A–E grades, Duolingo 10–160, GRE 130–170, GMAT 205–805, SAT 400–1600 and dMAT module scores.
- Checked conversions. Our conversions reproduce every published official conversion point.
- Verified maths keys. Every SAT, GRE and GMAT maths answer key, and every dMAT key, is checked by an independent reasoning solver before a paper reaches students.
Writing and speaking
The board's own criteria, word for word
The AI examiner marks against each exam's published band descriptors and criteria, taken from the board's own documents rather than paraphrased. It works in two steps:
- It records the evidence in the answer: what the task asked for and what was covered, the range and accuracy of the language, and the organisation.
- It picks the descriptor that matches that evidence for each criterion.
The score is computed from those picks. Every criterion on the result page therefore comes with the reason for its band.
Comparison with examiner-rated performances
For IELTS Speaking, CELPIP Speaking and English placement tests, the candidate's performance is also compared with performances that the exam boards' examiners have rated. Each comparison is made in both orders, and the candidate is placed among the rated speakers. Deciding which of two speakers is stronger is the judgement AI makes most reliably, so the level comes from where the candidate sits on a ladder of officially rated speakers.
IELTS Speaking is rated once over the whole test, as the IELTS descriptors require, rather than answer by answer.
Exam by exam
| Exam | Objective sections | Writing and speaking |
|---|---|---|
| IELTS Academic & General Training | Raw marks out of 40 on the band tables for each module; overall band rounded as IELTS rounds it | Four public band descriptors, Task 2 counting double; Speaking rated over the whole test and placed against examiner-rated speakers |
| TOEFL iBT (2026) | 1–6 band per section | ETS's published rubric for each task; task points added up as ETS's technical manual describes and linked to the 1–6 band |
| PTE Academic & Core | Pearson's task rules, partial credit and negative marking included | Pearson's published criteria for each task; each skill score integrates the tasks that count towards it |
| CELPIP | Paragon's score charts | Paragon's four performance dimensions; speaking placed against officially rated responses |
| OET | OET's 0–500 scale and A–E grades | OET's official criteria and level descriptors; speaking adds the clinical-communication indicators, judged on evidence quoted from the role play |
| Duolingo English Test | 10–160 in steps of 5, with subscores | Aligned, skill by skill, to Duolingo's officially scored example answers |
| GRE | 130–170, section-adaptive | ETS's 0–6 scoring guide for Analytical Writing |
| GMAT, SAT & dMAT | Adaptive scoring on the official scales; dMAT module scores and their sum | — |
| CEFR placement & custom tests | Each skill on its own A1–C2 band ladder | English writing on the Council of Europe's CEFR descriptor scales; English placement speaking against rated speakers; other languages against exemplars written for that language |
How accuracy is checked
- Against the boards' own scored answers. We use the examiner-marked samples the exam boards publish for IELTS, GRE, PTE, OET, the Duolingo English Test and CELPIP, plus an independently CEFR-rated research corpus for CEFR writing.
- The production engine, as students meet it. The same model, criteria and code path that score a student's answer are the ones measured.
- Never graded on its own examples. A sample used as a comparison anchor is never used to measure itself.
- Before every change ships. Scoring changes are measured on the same samples first.
- Student data is never used. Customer and student answers are never used to validate or tune scoring.
What the measurements show:
- IELTS Writing and Speaking: within half a band of the official examiner band on average.
- Objective sections: every published official conversion point reproduced.
- CEFR writing: within one band of trained human raters about 9 times in 10.
- CELPIP, OET, Duolingo English Test and GRE writing and speaking: close to the officially awarded scores.
What students and staff see
- Results are always delivered. If an AI provider is briefly unavailable, the answer is retried. The result page says the attempt is being scored, and the result appears as soon as the scoring is stored. A score is never invented.
- One rating per attempt. If a test is submitted twice at the same moment (a double click on Finish, or an exit as the timer runs out), the attempt is still rated once and scored once.
- Re-evaluate. Admins and examiners can re-evaluate a completed attempt from the result page or from analytics. It re-scores the attempt with the current scoring and tries again any speaking rating that could not be completed earlier.
- If the attempt is being rated at that moment, Re-evaluate says so and changes nothing; try again a minute later.
- If a speaking rating still cannot be completed, the response tells you, and you can run Re-evaluate again later.
- Estimates on the official scale. Practice results are estimates on each exam's official scale; official scores are issued only by the exam boards.