Flagship · BandBoost / v1.0

The examiner, simulated.

BandBoost grades full-length IELTS mock tests across all four skills — against the exact band descriptors examiners use — and tells you the score you'd get today, criterion by criterion.

Listening· Reading· Writing· Speaking — graded to the half-band
// Calibration readouts
4
IELTS skills, graded end to end
41
Band descriptors modelled
0.32
Mean band error vs. examiners
82s
Avg. time to a scored report
The product · BandBoost

Grades like an examiner.
Explains like a tutor.

BandBoost runs full, timed mock tests, then evaluates every response against the official nine-band descriptors — the same criteria a certified examiner applies. You get a band per skill, the breakdown behind it, and the precise moves that lift it.

01
Listening
sectional accuracy · distractor analysis
02
Reading
inference · scanning · paraphrase tracking
03
Writing
task · coherence · lexical · grammar
04
Speaking
fluency · lexical · grammar · pronunciation
See how grading works
Score report · Writing T2
Graded
7.5 OVERALL
Predicted band
Within ±0.5 band of certified examiner marking.
Task achievement7.5
Coherence & cohesion7.0
Lexical resource8.0
Grammatical range7.0
// examiner note  Clear position throughout. Vary complex structures to reach Band 8 grammar.
Technology · Capabilities

Intelligence, built for assessment.

A purpose-built evaluation stack — not a general chatbot in a costume. Every capability is anchored to the rubric.

// Core engine

Criterion-level grading

Scores against the official band descriptors — not a black box. Every band arrives with the evidence and the criterion that produced it.

Examiner-aligned model

Calibrated against thousands of certified examiner judgements.

Speech & fluency analysis

Pronunciation, pacing and lexical range, scored from your spoken answers.

Exam-faithful mocks

Full-length, timed and formatted exactly like test day.

Band trajectory

Track every attempt and watch the curve bend toward your target.

The Lab · R&D

Tested like an instrument.

Scoring you can trust is scoring you can measure. Every model release runs the same loop, against held-out responses a certified examiner has already marked.

01
Experiment
Form a hypothesis about what moves a band.
02
Test
Score against held-out, examiner-marked responses.
03
Iterate
Tune until error falls inside half a band.
04
Deploy
Ship to BandBoost; monitor continuously for drift.
benchmark.log live
$ calibrate --cohort=certified n=1284 ok $ eval writing.task2 QWK=0.91 MAE=0.28 $ eval speaking.pron MAE=0.21 band $ agreement ±0.5 band ............ 97.4% $ drift check ................... nominal $ deploy bandboost@1.0 .......... live
Examiner agreement within ±0.5 band
Listening98.6%
Reading98.1%
Writing96.4%
Speaking96.9%
n = 12,840 examiner-marked responses · updated weekly
About · Asmriti Labs

We build evaluation intelligence — and we started with the hardest rubric in language testing.

BandBoost reads an IELTS response the way a certified examiner does: criterion by criterion, against the published descriptors — never by vibes. It is the first instrument out of the lab, and the method generalises far beyond it.

asmṛti — Sanskrit: that which is remembered, recorded with precision.

01 / Rubric first

Every score traces back to a published band descriptor. No mystery numbers.

02 / Measured, not claimed

We publish agreement, error and drift — and re-run them on every release.

03 / Built with examiners

Certified markers sit in the loop, from calibration through to review.

Early access · Cohort 01

Find your real band.

Request access to BandBoost. We're onboarding test-takers and teaching teams in small cohorts.

// Enter a valid email address to continue.
No spam · Unsubscribe anytime
You're on the list.
We'll reach out when Cohort 01 opens a seat.
// confirmation queued for