How Erosophia estimates your Linguaskill score

In short. The scores shown on Erosophia are unofficial estimates, for Linguaskill and for Linguaskill Business alike. Only Cambridge University Press & Assessment issues Linguaskill scores. Erosophia is not affiliated with or endorsed by Cambridge. This page explains how each score is calculated, what follows Cambridge's publications and what cannot follow them.

Why no website can reproduce the official score exactly

Reading and Listening are adaptive: the difficulty follows your answers and the test stops once your level is established precisely enough [1] [2]. Cambridge publishes neither its items, nor their difficulty, nor any conversion table.
In the exam, Writing and Speaking are marked by examiners and by automated marking technology ("hybrid marking") [1]. On Erosophia, an AI scores the answers; for Speaking, it only reads an automatic transcript.
Cambridge does not publish the weighting of the Speaking parts, the rounding of scores or the measurement error of the current version.

Scale and levels

As in the exam, each module is scored on the Cambridge English Scale, from 140 to 210 [3]:

CEFR level	B1	B2	C1	C2
Score	140–159	160–179	180–199	200–210

Below 140, Cambridge gives no score and no level: the certificate shows "Not reported" for that module [3]. Erosophia then shows "< 140" and "Not reported".
The official certificate also gives the average of the modules scoring B1 or higher [3]; Erosophia shows one score per module, with no overall average.

Reading and Listening

Marking: each answer is right or wrong. For typed answers, spelling counts; capital letters and typographic apostrophes or dashes do not. In Listening, the three-word limit of the instructions is applied.
Adaptive test: as in the exam, the test starts with B2 tasks [2], then each task is chosen close to your estimated level. After each task, your level is recalculated from all your answers (Bayesian estimate). The test stops when the estimate is stable to within about ± 7 points, after at least 6 and at most 12 tasks, and in any case after 59 minutes, the official maximum [1].
Limit: the level of each task (B1, B2, C1 or C2) was set by its author, not measured on thousands of candidates as Cambridge does. The score is therefore a rough guide and may be one band away from your real result. In a simulation over the 40 tests, and provided the task levels are right, it falls in the right band about 80% of the time.

Writing

Format: as in the exam, two parts in 55 minutes: an email (Part 1, at least 50 words recommended, 25% of the mark), then an essay (Linguaskill) or a report (Linguaskill Business), at least 250 words recommended, 75% of the mark [1].
Criteria: the AI scores each part on Cambridge's three criteria, Communicative Achievement, Organisation and Language [4], against level descriptors from B1 to C2: 5 for C2, 4 for C1, 3 for B2, 2 for B1, 1 below B1. A criterion gets the highest level whose descriptor is fully met.
Rules applied by the site after the AI:
- the score of a part is the average of its three criteria;
- an answer that is not in English, or unrelated to the task, scores 0;
- Cambridge presents 50 and 250 words as recommended minimums, with no published penalty. Erosophia does not cap a text that reaches 90% of that minimum; below it, the score is capped at 3.5 out of 5 (70 to 90%), 2.5 (50 to 70%) or 1.5 (under 50%), and at 0.5 under 10 words;
- a part left blank counts as 0.
Module score: weighted average of the two parts (25% and 75%), converted to the 140–210 scale: 5 out of 5 gives 210, each point is worth 22 scale points, and the 140 threshold (B1) corresponds to 1.8 out of 5. Cambridge publishes no conversion; a single rounding is applied, at the end.

Speaking

Format: as in the exam, five sections in 16 minutes (presentation, summary, three-question discussion, recommendation, argument), after an unmarked warm-up [1].
Transcription: each recording is transcribed automatically, then the AI scores it on Cambridge's three criteria: Pronunciation and Fluency, Language Resource and Discourse Management [5], against their descriptors from B1 to C2, taking into account what the section asks for.
Pronunciation: the AI cannot hear it, so it does not assess it. It only scores the fluency visible in the text (amount of speech, continuity), and that score cannot exceed the better of the other two.
Rules applied by the site after the AI: the score of an answer is the average of the three criteria; an answer that is not in English, or unrelated to the question, scores 0, as in Cambridge's criteria [6]; an answer of fewer than 10 words scores 1 at most; below 55 words per minute of the speaking time allowed, fluency scores 1 at most, since the instructions ask you to speak for all the time you have.
Silence and failures: a silent recording counts as 0. If the microphone, the transcription or the marking fails, the answer is not counted, the site says so and the partial score is not saved to your progress.
Module score: the five sections count equally (Cambridge does not publish its weighting); in the discussion, the score is the average of the three answers. The conversion to the 140–210 scale is the same as for Writing.

How this scoring is checked

Before each change, Erosophia has the site score a set of 60 reference answers. It includes emails, essays, reports and spoken answers from A2 to C1 level, in Linguaskill and Linguaskill Business, plus edge cases: text too short, off-topic, copied instructions, answer in French, silent microphone, transcription failure… These answers were first scored blind by three independent markers; these markers are AI models distinct from the one that scores the site.
At the last check, on 29 October 2026:
- all 14 edge cases were scored correctly;
- in Writing, the site gave the same band as the markers for B2 and C1 scripts;
- in Speaking, it gave "Not reported" to A2-level answers, and C1 to C1-level answers;
- it still scored emails slightly high, and B1-level spoken answers as B2.
These markers are not Cambridge examiners. To check the scoring against real results, you can enter your official score, if you wish, on your "My progress — Linguaskill" page.

Official sources

Linguaskill is a trademark of Cambridge University Press & Assessment. Erosophia is not affiliated with or endorsed by Cambridge University Press & Assessment. Page updated on 29 October 2026.