Scoring & calibration
How Lexis scores your mock exams
An AI score is only worth what it was measured against. Here is what each of our three scoring engines was calibrated against, what we measured, and where each one's limits are.
The short version
Every Lexis mock is scored by AI, and every AI engine we run was tuned against real marks rather than left as an off-the-shelf language model guessing at a band. Our PTE scoring maps to real Pearson scaled scores from students who sat the actual exam. Our IELTS scoring was measured against hundreds of attempts already marked by trained human raters. Our MET scoring applies the published Michigan Language Assessment rating scales, criterion by criterion, word for word.
None of that makes a practice score an official result. It makes it a score you can plan around. The rest of this page shows you the actual numbers.
PTE — calibrated against real Pearson results
A PTE score is a number from 10 to 90, and the only thing that makes any practice number meaningful is whether it tracks what Pearson would have given you. So that is what we measured it against.
We took students who had sat the real PTE and paired each one's mock performance with their actual Pearson scaled score, then fitted the raw-to-scaled curve to those pairs, skill by skill. Those fitted anchors — not a placeholder or a generic curve — are what runs in production when your mock is scored.
What we measured, and where it is strongest
| Skill | How closely it tracked real Pearson |
|---|---|
| Listening | Strongest of the four |
| Reading | Solid |
| Speaking | Directionally good, on the smallest sample |
| Writing | The weakest fit of the four |
What this does not mean
We say calibrated, and we mean measured against real results. We do not mean Pearson-exact. The fit is provisional and built on a small number of students, and predicting a real exam score from any mock carries roughly a band of noise no matter who builds it. We refit the curve as more students report back their real scores.
IELTS — measured against attempts human raters had already marked
An IELTS band is a human judgement, so the only fair test of an AI band is whether it lands where trained human raters landed on the same work.
We built a harness that runs our production scoring logic over attempts that had already been marked by a review centre's trained human raters, then reports how far our AI bands sit from theirs.
- Speaking was measured across 316 human-rated attempts, with every criterion mapped one-to-one to the human rater's criterion — fluency to fluency and coherence, grammar to grammatical range, lexical resource to lexical resource, pronunciation to pronunciation.
- Writing was measured across a larger set of human-rated essays, replayed through the scorer using the exact production prompt. On that set the scorer landed within half a band of the human mark on 84% of essays, and within one full band on 97%, with no systematic tendency to score high or low.
- The logic that decides your score is the same code in the live scorer and in the calibration harness, so what we measured is what you get.
The number we deliberately do not publish
We also built a system that keeps nudging AI bands toward human marks over time, from a running log of the gap between them. It is built, it is switched on, and it has not yet accumulated enough logged samples to be adjusting anything. We will say so on this page when that changes. We would rather tell you what is dormant than let you assume it is running.
Where human marking actually appears in your package
Human rating is a specific, limited thing at Lexis, and we will not blur it:
- The IELTS Mock Package includes one examiner-rated Writing and Speaking review, on one designated mock, on top of the AI scoring you get on every mock.
- Success Packages and all PTE and MET packages are AI-scored. They do not include examiner rating, and no Lexis page should tell you otherwise.
MET — the published rating scales, applied criterion by criterion
The MET's Writing and Speaking rating scales are published by Michigan Language Assessment. We did not paraphrase them, summarise them, or invent a Lexis version. The band descriptors our AI marker works from are the published descriptors, copied word for word.
So your Writing is marked on five criteria and your Speaking on three — the same criteria, on the same 0–4 scale, that the published scales define:
| Skill | Criteria (each 0–4) |
|---|---|
| Writing | Grammatical Accuracy · Vocabulary · Mechanics · Cohesion and Organization · Task Completion |
| Speaking | Task Completion · Language Resources · Intelligibility and Delivery |
Your criterion marks then feed a projected MET-equivalent score and a CEFR estimate. See how MET scores and CEFR levels work.
What this does not mean
Using the published scales word for word is not a relationship with the people who published them.
What a Lexis score is, and what it is not
| A Lexis score is | A Lexis score is not |
|---|---|
| A projection, measured against real marks, of where you would land today | An official result from any test provider |
| Accurate enough to tell you which skill is holding you back | Accurate to the point |
| Consistent — the same logic marks every attempt, every time | Immune to the roughly one band of noise between any mock and any real exam |
| Re-measured as more real results come back to us | Fixed forever |
The most useful thing a mock can tell you is not your score. It is which of the four skills is costing you the most, and how much time you have left to fix it.
Questions
Everything you might want to know.
Is Lexis AI scoring accurate?
Accurate against what we measured it against, and we publish both sides of that. Our PTE scoring is fitted to real Pearson scaled scores from students who sat the actual exam. Our IELTS scoring was measured against hundreds of attempts already marked by trained human raters, landing within one band of the human mark on almost all of them. Our MET scoring applies the published Michigan rating scales word for word. What no honest provider can promise is a practice score that matches your real result to the point: predicting a real exam score from any mock carries about a band of noise for everyone.
Does a real teacher look at my work?
On the IELTS Mock Package, yes. One designated mock gets an examiner-rated Writing and Speaking review with annotations, an instructor note, and a sub-band breakdown, on top of the AI scoring you get on every mock. Every other package, including the Success Packages and all PTE and MET packages, is AI-scored.
Is Lexis connected to IELTS, Pearson, or Michigan Language Assessment?
No. Lexis is an independent preparation provider and is not affiliated with, or endorsed by, any of them. MET is a trademark of Michigan Language Assessment. Practice scores are estimates, not official results.
Why is my mock score different from my real exam score?
Some gap is expected, for every provider and not just us. A mock measures you on a different day, on different material, without the pressure of the real room, and every mock-to-real projection carries roughly a band of uncertainty. Use the score to find your weakest skill and to check whether you are trending up. Do not treat it as a guarantee.
How often is the scoring updated?
The PTE curve is refitted as more students report their real Pearson results back to us. A change to a scoring model is gated on replaying stored work against the human-rated ground truth: it has to land within tolerance of the human marks, with no systematic bias, before it ships.
Which parts of the test are AI-scored at all?
Reading and Listening are objective. They are marked against an answer key, with no AI judgement involved. Writing and Speaking are the parts an AI marker judges, and those are the parts we calibrated.
Where to go next
- IELTS mock examsAcademic and General Training, AI band scoring plus one examiner-rated review.
- PTE Academic and PTE CoreMocks in the real item types, scored on the Pearson-calibrated curve.
- The complete MET guideWhat the test is, how the 0–80 score works, and how to prepare.
- Compare packages and pricingEvery Lexis package for IELTS, PTE and MET.
See the scoring for yourself
Every exam has a free way in. Take one and read the report before you spend anything.
Compare packagesLexis is an independent preparation provider and is not affiliated with, or endorsed by, Pearson.
Lexis is an independent preparation provider and is not affiliated with, or endorsed by, the IELTS partners.
MET® is a trademark of Michigan Language Assessment. Lexis is an independent preparation provider and is not affiliated with, or endorsed by, Michigan Language Assessment.
Projected scores are estimates, not official MET results.