NO BLACK BOX

Here is exactly how every number on this site is produced.

Four things in the product produce a score. This page names what goes into each one, what computes it, and what comes back — and then it names the places where the number is weaker than it looks.

Describes the system as it works on 10 September 2026. If a line here stops being true, the line changes.

THE FOUR SOURCES

Every score starts in one of four places.

Read each row left to right: what you gave it, what ran on it, what came back.

The placement test
At setup, before a plan is built for you
What goes in
Six quick questions to find roughly where you are, then a longer set near that level: grammar, vocabulary, one reading passage and a single writing task, saved as you go.
What computes it
The ladder narrows the level as you answer, so where the questions settle matters more than any single one of them. The writing task is graded by our AI against the same four criteria the writing marker uses.
What comes out
A CEFR level for each skill it could grade, an overall CEFR, and — on the matching track only — an estimated IELTS band or an estimated TOEFL total out of 120. Skills it could not grade come back in an explicit unmeasured list rather than as a zero. Speaking is not part of this test.
The reading mock test
On demand, and as one half of Exam Day
What goes in
One passage of 380–450 words, ten four-option questions, and a fifteen-minute wall clock.
What computes it
Your correct answers as a share of the questions, read off a fixed table. On a ten-question mock, ten right returns 9.0, nine 8.0, seven 6.5, five 5.5, and one or none 2.5, the lowest it returns. The same ratio always gives the same band — nothing about it varies between attempts.
What comes out
One estimated band, for reading, from a short mock. It is a single section against a single passage — not a simulation of an exam, and there is nothing else inside it.
A graded essay
Any time, and as the second half of Exam Day
What goes in
Your essay, written on a generated topic, optionally under an exam clock. It saves as you type, so a closed laptop does not cost you the draft.
What computes it
Our AI marks it on the four IELTS criteria — task response, coherence and cohesion, lexical resource, grammatical range and accuracy — each in half bands from 4.0 to 9.0.
What comes out
An estimated band, a short verdict, two to three strengths, and up to six corrections, flagged only for real errors, that each quote the exact words you wrote beside the corrected version. The corrections are marked in place, inside your own paragraph. On the TOEFL track the band is the same IELTS-style figure, labelled as not a TOEFL score.
A spoken answer
Any time, from a cue card matched to your track and level
What goes in
A recording. IELTS Part 2 gives you 60 seconds to think and 120 to speak; the TOEFL track’s opinion task, 20 and 60; General English, 30 and 90. Under fifteen seconds or under twenty transcribed words it refuses to score the take rather than guess at one.
What computes it
Our speech assessment transcribes the audio and scores pronunciation, fluency, coherence, grammar and vocabulary. Our AI reads that and writes the coaching — the verdict, what worked, and the lines worth saying differently.
What comes out
On the IELTS track, an estimated band. On the TOEFL track, the same criteria as Liguastic’s own estimate from 0 to 4, where 4 is full marks; it is not an ETS score. Pronunciation appears only when there was audio to hear it in.

Every other number in the product is a count of what you did rather than an estimate of your level — per-skill accuracy, the review queue, the streak, days active. Those are arithmetic on your own answers, and nothing infers a band from them.

SCALES

A 6.5 and a 3.0 can describe the same standard of English.

IELTS counts in bands out of nine. On our TOEFL track a spoken answer is estimated out of four, where four is full marks: that scale is ours, not ETS's, whose score reports have used 1–6 since 21 January 2026. We never plot one against the other, because we have no defensible way to do it.

Illustration — not a real result

/ 9

Speaking, IELTS track

Estimated IELTS-style band, 0–9

/ 4

Speaking, TOEFL track

Liguastic TOEFL speaking estimate, 0–4 — not an ETS score

Two tracks, two ladders. Each dial prints its own scale beneath the label, so a 3.0 is never read as a near-fail when it is close to the top of its own scale.

Where a score out of 120 appears: In exactly one place: the placement result, on the TOEFL track. It is never assembled from later work, and no other screen prints a total out of 120. It is our estimate, not an ETS score: TOEFL iBT itself now reports on a 1–6 scale, with a comparable 0–120 score during a two-year transition.
Why progress draws two charts: Your trajectory plots every graded moment in time order, split into a separate chart for each scale. Putting a 3.0 out of 4 on the same axis as a 6.5 out of 9 would invent a conversion nobody has given us.
General English carries no band at all: If you are not preparing for an exam, there is no band target on your account. The field is empty, and nothing on the page fills it in with a guess.

Rows of empty desks in a school exam hall, set out before a sitting

WHAT AN ESTIMATE IS

Nothing here is marked by the people who mark the real thing.

These are estimates produced by the process above. They are not official scores, and there is nowhere you can submit one.
No examining body has reviewed, approved or endorsed a single number on this site.
An estimate can move half a band between two attempts that felt identical to you. Several attempts in a row carry more information than any one of them.
We never write “your IELTS score”. It is an estimated band, and it is ours.

No affiliation, of any kind

Liguastic is not affiliated with, endorsed by, approved by or connected to IELTS, IDP, the British Council, Cambridge, ETS or TOEFL. Those names appear on this site only to say which exam a learner is preparing for.

WHERE IT IS WEAKER

We would rather you knew where this is weak.

Speaking scored from a transcript has no pronunciation score.

Pronunciation needs a recording. If you type your answer instead of speaking it, the pronunciation score is not guessed, greyed out or set to zero — the row is simply absent, and the result says why.

This is what that result looks like. Nothing was substituted for the measurement we did not take.

Estimated band

Sample result

6.5

Graded from the transcript only — pronunciation needs a recording, so it isn’t scored here.

Fluency

6.5

Vocabulary

6.0

Grammar

6.5

This attempt was graded from a transcript. Pronunciation is missing above because nothing listened to it — we would rather leave the row out than print a number we did not measure.

Verdict

You used the whole two minutes and finished your last idea instead of stopping mid-sentence, which is what Part 2 is really testing. The ideas were ordered clearly, but three linkers carried the entire answer and a few noun phrases lost their articles. Widening the linking is the fastest gain available to you here.

What worked

You spoke for the full time without a long silence.
The answer had a beginning, a middle and an end.
Two topic words — “coastline” and “off-season” — were used accurately.

Say it better

I depend of my sister for the difficult words.
I depend on my sister for the difficult words.
“Depend” takes “on” (or “upon”) before its object. This one has come back three times in review.
It was very good experience for me.
It was a very good experience for me.
“Experience” is countable when it means a single event, so it needs “a”.
And then, and then we go to the beach and it was nice.
After that, we went to the beach, which was lovely.
A repeated filler and a tense slip in one line — “after that” buys the same thinking time.

Next focus: pick two contrast linkers and use each one once in your next take.

The mock band comes from a fixed ladder.
An exam board equates — what equating means its papers, so a harder passage needs fewer correct answers for the same band. Ours does not. Eight out of ten returns the same band whether that passage was gentle or brutal, which means an easy set flatters you and a hard one under-reads you.
The grader is a model, and a model can be wrong.
Our AI marks the writing and writes the speaking coaching. On a borderline essay, half a band either way is entirely plausible, and it marks what you actually wrote rather than what you meant. If a correction looks wrong to you, read the reason it gives before you accept it — the reason is there so you can argue with it.
One essay is not a writing band.
The writing level in your placement comes from a single task on a single prompt. It is the most honest read available from one task, and it is thinner than a band built from several essays on several topics. Treat the trajectory across attempts as the signal and any one result as a reading.

VALIDATION

No validation study has been published.

Not a partial one, not an internal one we are sitting on. Nobody has independently measured how close these estimates land to a real exam result — which is why there is no accuracy percentage anywhere on this site.

If you are comparing products and one of them prints an accuracy figure, ask for the method, the markers and the sample size. We have not printed one because we have not run the study.

What we intend to run, in order

A blind re-mark
The same essays scored by our grader and by qualified human markers, reporting exact agreement and agreement within half a band — including the cases where the two disagree badly.
A comparison against reported results
Learners who sit the real exam and volunteer what they got, published beside the sample size however small that number is.
A check on the reading ladder
Whether the fixed ratio-to-band table in code matches the reading bands those same learners actually report.

Until those exist, the honest version is the one at the top of this page: here is the process, and here is where it is thin. If a number ever appears, it arrives with its method, its sample and its failures, or it does not appear.

No black box.

START

Now you know how the number is made.

Measure where you actually are with a placement that says what it could not grade, then work on what it found.

Start learning See the method

Think a line on this page is wrong? Tell us at hello@liguastic.com — we would rather fix the copy than defend it.

Here is exactly how every number on this site is produced.

Every score starts in one of four places.

The placement test

The reading mock test

A graded essay

A spoken answer