AI assurance for education
Show what your education AI does.
Show your working.
1
Define the scope
Test Pack. • Agreed Scenarios
Rubric
Teacher review
Borderline cases
Model update
2
Run the agreed scenarios
Your Product
3
Define the scope
Test Pack. • Agreed Scenarios
01
The scope
02
The test record
03
The findings
04
The limits
The decision moment
A reviewer needs more than a demo.
A school review, pilot, or release decision brings practical questions:
Q1
Does the feedback reflect the agreed rubric?
Q2
Can teachers review and correct outputs before release?
Q3
How does the product handle borderline or ambiguous responses?
Q4
Has a model or prompt update changed previously tested behaviour?
starts with the questions relevant to your product and decision.
We agree the scenarios and expected behaviour before testing.
The evaluation
A defined evaluation. A report reviewers can follow.
Test Pack
The evaluation
A Test Pack is the evaluation: an agreed set of scenarios for a specific product version and use case.
Evidence Pack
The report
The Evidence Pack is the report you receive. It connects each finding to the tests and observations behind it.
The boundary
The report supports a decision. It does not certify the product, guarantee safety, or establish learning effectiveness.
Example finding
Follow the finding—not just the conclusion.
1
Product claim
2
Agreed expectation
3
Hypothetical observation
4
What would need attention
5
Next check
Teachers approve generated marks before learners can see them.
How it works
Start with the decision. Agree the tests. Show the results.
01
Define the scope
Tell us which product version you want assessed, who needs the evidence, and which decision it must support. We agree the questions, access, exclusions, timing, and fee before testing.
02
Run the agreed scenarios
We test in an authorised environment against expectations defined in advance. Inputs, outputs, configuration details, and repeat runs are recorded as agreed in the test plan.
03
Review and report
We review the observations and explain the findings, next steps, and limits. The report separates observed behaviour from interpretation and unresolved questions.
Change and retest
The product changed. Does the evidence still apply?
A change to the model, prompt, rubric, data source, or workflow can affect an earlier finding.
A Release Retest reviews the relevant change and tests affected scenarios against the earlier record. New results remain linked to the original findings, with any remaining limits stated.
Retests are separately scoped. A Test Pack does not include continuous monitoring or automatic coverage of future versions.
A change to the
model
prompt
rubric
data source
workflow
Release Retest
↑ linked to original findings
Affected scenarios tested against the earlier record
FAQs
Questions reviewers ask us.
Who is this for?
+
Edtech product teams that need evidence for a defined school review, pilot, or release decision. The report is designed for the people investigating the product and those reviewing the decision.
Will a school accept the report?
+
Does this replace internal QA or other assessments?
+
Can this prove the product improves learning?
+
Do you need real student data?
+
Can we question a finding?
+
Who makes the final decision?
+
What does your next review need to establish?
Tell us about the product, the decision, and the deadline. Start with the question—not a file upload.
This starts a scoping conversation. It does not book or begin testing. Scope, access, timing, and fee are agreed before any evaluation.