Tests & benchmarking

Make speaking assessment consistent, explainable and easier to operate.

Run computer-based speaking tasks with stable scoring, item-level evidence and reports that teachers and administrators can review.

Explore assessment moments

Batch assessmentStable auto-scoringCohort and item reports

Students completing a computer-based speaking assessment in a school lab

Designed for

Schools, districts, publishers and examination service providers

Use case

Cohort scale: parallel delivery and batch assessment
Item level: evidence for review and teaching use
Reviewable: clear exception and validation paths

/three assessment moments

Deliver the task, review the exceptions, then publish a report people can use.

The same response set has to be fair for the student, inspectable for the assessment team, and useful for the school—not only fast to score.

Students completing a computer-based speaking test in a quiet exam lab

Computer-based delivery

In operation: A locked prompt, timing and headset path that feels the same at every site.
Product signal: Controlled capture, completeness checks and a response that can enter the scoring pipeline.

Assessment coordinators reviewing a speaking-test item and waveform together

Item-level review

In operation: Reviewers inspect exceptions instead of guessing why a recording failed.
Product signal: Item evidence, audio-quality flags and a path to reprocess or moderate.

School leader and teacher reviewing speaking-assessment results in a meeting room

Cohort reporting

In operation: Leaders see class and programme patterns, not only a list of grades.
Product signal: Cohort, class and item views with access controls for each role.

See the assessment engine, without the agent layer.

Choose the product SDK experience or the MCP agent walkthrough based on how you plan to integrate.

Choose a live demo

/built for operated assessment

A speaking test is a delivery, scoring and review system.

School assessment is not practice with a lock icon. It needs controlled tasks, validation before scoring, a path for exceptions, and reports that different roles are allowed to see.

Built for these activities

Placement tests
Term exams
Benchmarking
Oral proficiency tasks
Item analysis
Moderation

01
Controlled delivery
Lock the prompt, timing, language and recording conditions required by the assessment plan so sites do not invent their own process.
02
Evidence that can be reviewed
Incomplete audio, duplicates and scoring exceptions stay inspectable. Automated output is compared with the intended rubric before release.
03
Reports that guide programmes
Student, class, item and cohort views should help teaching follow-up—not only produce a grade.

/before a score is released

Fit automated evidence, human review and reporting together.

Decide how automated evidence, human review and reporting fit together before a score is released.

The situation

Manual speaking assessment is expensive to coordinate and difficult to standardise.

What changes

Repeatable scoring at cohort scale, with evidence for review and teaching follow-up.

/who acts on the evidence

Students need a fair task. Reviewers need exceptions. Leaders need a programme view.

Student

Needs: A fair task and reliable recording process.

Consistent delivery with a clear path when audio is incomplete.

Assessment team

Needs: Stable scoring and explainable exceptions.

Evidence for validation, moderation and reprocessing.

School leader

Needs: Results that can guide programmes, not just produce grades.

Cohort and item views connected to teaching follow-up.

/from delivery to report

Deliver, validate, score, then publish the right view.

High-stakes use is an operations problem as much as a scoring problem. Each stage needs an owner.

01
Deliver a controlled task
Lock the prompt, timing, language and recording conditions required by the assessment plan.
02
Validate every response
Check completeness and audio quality before a response enters the scoring and reporting pipeline.
03
Score and moderate
Apply the language engine, review exceptions and compare the output with the intended rubric.
04
Publish useful reports
Return student, class, item and cohort evidence with the appropriate access controls.

/ready for assessment design

Align the rubric, the controls and the reporting roles.

Stable auto-scoring is not enough if retries, missing audio and access control are undefined.

Decision 01

Rubric alignment

Benchmark representative responses against the intended proficiency or examination standard.

Decision 02

Operational controls

Plan retries, missing audio, duplicate submissions, moderation and audit trails.

Decision 03

Reporting roles

Separate what students, teachers, reviewers and administrators are allowed to see.

Responsible boundary

High-stakes use requires validation on representative candidates, tasks, devices and recording environments, with human review for defined exceptions.

Signals worth tracking