Dmitry Reznik

Written by: Chief Technology Officer

Dmitry Reznik

Posted: 20.11.2026

21 min read

EdTech serves five kinds of users at once, from a seven-year-old on a classroom tablet to a university registrar reconciling 40,000 enrolments. A feature that works for one is often broken for another: the grade that is right in the instructor's view and wrong in the student's, the roster that synced for the admin and missed the late enrolee.

This guide tests the platform through each pair of eyes, then covers what sits underneath all of them – interoperability standards, student-data privacy, accessibility, the first day of term, and AI.

We have tested learning platforms, LearnOps tools and childcare systems since 2010; our educational software testing services page describes the service, and this is the method.

Ready to take your software quality to the next level?

Why edtech software testing is different

Educational software testing is different because the users span every age and ability, the load follows the academic calendar, the integrations are mandated by standard, and the data belongs to minors.

1.

Users span ages, abilities and devices. The same platform serves a child who cannot yet read error messages and a professor who will not tolerate one. The device matrix includes a 2015 Chromebook on a school network with half the ports blocked.

2.

The calendar is the load profile. Term start, assignment deadlines at 23:59, exam week – demand is predictable and extreme, and a quiet August load test proves nothing about the first Monday of September.

3.

Interoperability is mandated. SCORM, xAPI, LTI, OneRoster and QTI are not optional features; institutions test vendors against them before buying, and a failed LTI launch in the customer's LMS is a lost contract.

4.

The data is about minors and is regulated. FERPA for US student records, COPPA for children under 13, GDPR and its member-state rules for children in Europe, and a growing set of US state laws. Internet Safety Labs' benchmark of 1,357 apps recommended or required by US schools found 96% shared student data with third parties, 78% of it with advertising and analytics companies – and regulators have noticed.

5.

Outcomes are the product. A wrong grade, a lost submission or a completion certificate issued early is not a bug ticket. It is a grievance, an appeal, or an audit finding.

Our earlier eLearning QA checklist still covers the basics well; this article is what has changed since – standards, privacy enforcement, live classrooms and AI.

The learner: "my progress disappeared"

E-learning app testing starts with the learner, because the learner is on the worst device, the worst network and the tightest deadline of anyone who uses the platform.

A day in the platform. She logs in on her phone on the bus and finishes the second half of a video lesson. At lunch she takes a timed quiz on a shared school laptop. At 23:58 she uploads a 180 MB assignment over mobile data. In the morning she checks her grade and finds the lesson she finished marked incomplete.

Where it breaks for her

1.

Progress is not saved when the app is backgrounded, the connection drops or the browser tab is closed mid-lesson.

2.

The video plays on home Wi-Fi and fails on the school network – blocked ports, a DRM licence server the proxy does not allow, or a CDN domain on a filter list.

3.

The quiz timer runs on the device clock; a submission at 23:59:50 by her clock is 00:00:10 by the server's and is marked late.

4.

The large upload fails at 90% with no resume, and the retry starts from zero.

5.

The lesson is complete on the device but the completion statement never reached the learning record store (LRS), so the gradebook shows nothing.

6.

The video has no captions; the quiz cannot be completed with a keyboard.

Tests

  • Offline, background and resume scenarios on real devices, with progress asserted server-side after each interruption.

  • A network-restricted lab profile that mimics school and corporate networks: filtered domains, blocked ports, throttled bandwidth, proxy authentication.

  • Timer tests against server time at deadline boundaries, including clock skew on the client.

  • Large-file upload tests with resume, over throttled mobile profiles.

  • SCORM and xAPI round trips: launch → statements → LRS → gradebook, with every completion and score status asserted end to end.

  • Captions, transcripts and keyboard navigation checked per content type, not per page.

  • A device matrix that includes low-end Chromebooks, older iPads and whatever the customer's districts actually issue — the method for building and maintaining one is in our guide on how to test mobile applications.

Field note – Learningbank. Learningbank is a digital learning platform for corporate training, built around the learner, available on all devices and customised for each client. When we joined there was no structured QA flow: developers and product owners tested in a single environment, there were no test cases and no regression, and post-release bugs were common. Over two years our two manual QA engineers built the flow from nothing – about 500 test cases, about 100 regression tests, more than 1,000 bugs reported, over 300 of them critical, blocking or major, across more than 100 released features. The learner-facing defects clustered in content rendering across customer brandings and devices, so a per-client configuration matrix became part of regression. Read the Learningbank case study.

The instructor: "the roster is wrong and the gradebook won't add up"

LMS testing from the instructor's side is mostly about two things: whether the people in the course are the people who should be there, and whether the arithmetic of grading is right in every view.

A day in the platform. He opens the course and a student who enrolled yesterday is not on the roster, while one who dropped two weeks ago still is. He grades 60 submissions against a rubric, posts the scores, and three students email to say their displayed grade differs from his. He runs a live session for 30 students; two cannot see the screen share, and the breakout rooms do not appear in the recording.

Where it breaks for him

1.

The roster is out of sync with the student information system (SIS): drops not processed, late enrolments missing, a section merge creating duplicates.

2.

Gradebook arithmetic – weighted categories, dropped lowest score, extra credit, late penalties, excused assignments – produces one number for the instructor and another for the student.

3.

A rubric edited after grading began silently re-scores already-graded work.

4.

Bulk operations (grade import, mass message, due-date shift) fail partway and leave half the class updated.

5.

The virtual classroom degrades unevenly: screen share fails for a few participants, breakout recordings are lost, attendance is recorded for students who never joined.

Tests

  • OneRoster or SIS sync contract tests with recorded payloads, plus a daily reconciliation of roster versus SIS, per section.

  • Gradebook golden tests: a fixed set of submissions and policies → expected grades for every policy combination, asserted in both the instructor and the student view. This suite catches the most grievances per hour of effort of anything in edtech.

  • Rubric and assignment versioning tests: changes after grading starts must prompt, not silently re-score.

  • Bulk-operation partial-failure tests with rollback or clear reporting of what did and did not apply.

  • Live-classroom tests under network degradation across the device matrix, and concurrency tests at class and lecture-hall scale – the tooling choices are compared in our guide to load testing tools.

Roster sync, gradebook logic, content standards and live-session reliability are the core of our learning management system testing work.

The administrator and L&D manager: "the report says 83% completed, HR says 61%"

For the administrator the platform is a rules engine and a reporting system, and the defects are mismatches between what the rules did and what the report says.

A day in the platform. She assigns mandatory compliance training to 4,000 employees by role, location and hire date, configures a learning path with prerequisites, and pulls the completion report an auditor asked for. HR's headcount does not match her assignment count; the completion percentage does not match the certificates issued.

Where it breaks for her

1.

Assignment rules misfire: contractors receive employee-only training; new hires in one region are missed because the hire-date rule used the wrong time zone.

2.

The completion report disagrees with source records – completions timestamped in UTC fall into the wrong month; re-takes count twice.

3.

The HRIS sync creates duplicate users on a name change or a transfer.

4.

Certificates are issued before every required module is complete, because the rule checks module count, not module identity.

5.

Learning-path prerequisites can be bypassed by a deep link to the final module.

Tests

  • An assignment-rule matrix run against a synthetic workforce covering every role, location, employment type and hire-date boundary.

  • Report reconciliation to raw completion events, with time-zone and re-take fixtures.

  • HRIS integration contract tests and de-duplication fixtures for renames, transfers and rehires.

  • Prerequisite enforcement tests through the UI, the API and deep links.

  • Certificate-issuance rule tests against every path configuration.

Field note – Cognota. Cognota is a LearnOps platform that helps learning and development teams run training intake, capacity planning and content design. Our two automation engineers inherited a Cypress suite that produced around 120 failures per regression run – static data, interconnected tests, deprecated interfaces – which meant nobody acted on it. In a year we refactored about 130 tests, cut flakiness by 90% and execution time by 20%, delivered 70% of the automation roadmap, and brought a regression run to zero or one failure. In admin-heavy platforms the suite's credibility is the product: a 120-failure run is ignored; a 0–1 run gets fixed the same day. Read the Cognota case study.

The parent: "who else can see my child's record?"

Childtech testing has one suite that matters more than all the others: the matrix of which guardian can see which child's data, because the product's core promise is showing parents their child – and only their child.

A day in the platform. He checks what his daughter ate and when she napped, pays this month's invoice, signs a field-trip permission slip, messages the teacher about a rash, and adds his mother as an authorised pick-up for Thursday.

Where it breaks for him

1.

A photo or record of another child appears in his feed because a classroom-level permission was applied where a child-level one was required.

2.

The authorised-pick-up change is saved in the app but not visible at the check-out tablet in the classroom.

3.

The invoice payment is charged twice after a timeout and retry; a late fee is applied on day two of a five-day grace period.

4.

The daily report includes a photo of a child whose parents declined photo consent.

5.

Notifications continue to a parent whose custody rights were removed from the record last week.

6.

A child-facing surface collects personal information without verifiable parental consent, which COPPA requires for children under 13.

Tests

  • A guardian × child × data-type access matrix – every pair, every type (schedule, meals, medical, photos, billing, messages) – run on every build.

  • Pick-up authorisation lifecycle tests across the parent app and the classroom device, including offline sync.

  • Payment idempotency and fee-rule tests with grace-period fixtures.

  • Consent-flag propagation into every generated artefact: reports, feeds, photos, newsletters.

  • Guardian-removal tests: access, notifications and messaging must stop on the effective date.

  • A data-collection audit of every child-facing screen against the consent record. The FTC's 2023 action against Edmodo – a $6 million penalty and the first order prohibiting an edtech provider from collecting more data than necessary – made data minimisation a test requirement, not a policy statement.

Parent apps, classroom tools and the access and consent logic between them are covered under our childtech software testing service.

Field note – a childcare management system. The platform serves daycare centres and preschools – enrolment, scheduling, reporting – and gives parents an app with schedules, meals, medical records, payments and activity sign-ups. Since 2021 our six full-stack QA engineers have built 3,000+ test cases, 1,600+ web and 150+ mobile automated tests running on 14 parallel threads, 20+ end-to-end load tests, and made regression six times faster, reporting 300+ bugs along the way. The guardian × child access matrix runs on every build; nothing else protects a product whose core feature is showing parents their child's data. Read the case study.

The institution: "does it pass our integration and compliance review?"

Platform functionality testing for academic platforms is, from the institution's side, a procurement gate: the vendor passes the integration, accessibility and load review or does not get the contract.

A day in the platform. The university's IT team evaluates the vendor: an LTI 1.3 launch from the campus LMS, single sign-on through the identity provider, roster sync from the SIS, grade passback to the campus gradebook, an accessibility conformance report they can believe, and proof it will survive 40,000 students enrolling at 08:00 on the first day.

Where it breaks for them

1.

The LTI launch works in the vendor's sandbox and fails against the institution's LMS version or configuration.

2.

SSO attribute mapping drops or mis-maps a role, and a cohort of students arrives as instructors.

3.

Grade passback rounds differently from the campus gradebook; every grade is off by a hundredth.

4.

The accessibility conformance report (VPAT) claims support the product does not deliver.

5.

Day-one load takes the platform down, or takes the SIS down through unthrottled sync calls.

Tests

  • LTI 1.3 Advantage conformance through the 1EdTech certification suite, plus launch, deep-linking, roster (Names and Roles) and grade (Assignment and Grade Services) tests against each supported LMS – Canvas, Moodle, Blackboard, Brightspace – at the versions customers actually run.

  • An SSO attribute and role mapping matrix for every identity provider configuration.

  • Grade-passback precision and rounding tests against each target gradebook.

  • An accessibility audit that reproduces the VPAT line by line.

  • A day-one load model with rate-limited, back-off-aware sync, and a browser and device matrix built the way institutions ask for it – our guide on what compatibility testing is covers how to derive and maintain one.

Underneath every persona: standards, privacy, accessibility, peaks and AI

Five concerns cut across all five personas, and each needs its own test discipline.

Interoperability standards you will be tested against

Interoperability testing means passing the published conformance suite and then testing against real partners, because partners deviate from the standard in ways the suite does not model.

  • SCORM 1.2 and 2004. Launch, suspend/resume, completion and score reporting across the platform's player and third-party content; SCORM Cloud for reference behaviour.

  • xAPI and cmi5. Statement generation, LRS storage and retrieval, and the mapping from statements to gradebook and reports.

  • LTI 1.3 Advantage. Secure launch, Deep Linking, Names and Role Provisioning Services, Assignment and Grade Services – conformance plus per-LMS contract tests.

  • OneRoster. Roster, class and enrolment sync from the SIS, with reconciliation.

  • QTI. Assessment item import and export without losing scoring rules or media.

  • Caliper. Learning analytics event emission and schema validity.

Student data privacy: FERPA, COPPA, GDPR and state laws

Privacy testing in edtech means proving data minimisation, consent and access control in code, across every system the data reaches.

  • A data-collection audit of every child-facing surface against what the educational purpose requires.

  • Consent flows: verifiable parental consent under COPPA for children under 13; school-authorised consent only where its limits are respected; GDPR consent and age thresholds per member state.

  • Role-based access tested as a matrix, not a checklist.

  • Data subject access and deletion requests fulfilled across the LMS, SIS, video provider, LRS and analytics warehouse.

  • Retention schedules enforced and tested; sub-processor flows documented and verified.

  • US state student-privacy laws (California's SOPIPA, New York's Education Law 2-d and their counterparts) mapped to requirements and tests.

The validation method – consent, DSARs, retention, lawful basis – is laid out in our GDPR testing guide; it is framed for finance, and every item transfers.

Accessibility is a procurement gate

Accessibility testing in education is not an enhancement; public institutions are legally required to buy accessible software, and the baseline is poor. WebAIM's 2024 analysis of the top million home pages found 95.9% with WCAG failures and an average of 56.8 errors per page – with education and government sites doing better than average, precisely because they are regulated.

  • WCAG 2.2 AA, Section 508 and EN 301 549 as the targets, with the VPAT as a tested document.

  • Captions and transcripts on every video; keyboard operability on every interaction; screen-reader order on every page; colour contrast everywhere; math, code and diagram content made accessible, not just text.

  • Testing with real assistive technology – NVDA, JAWS, VoiceOver, switch access – not only automated scanners. The framework we use is in our accessibility testing guide.

The first day of term and exam week

Performance testing for edtech is modelled on the academic calendar, not on average concurrency.

  • Enrolment spikes at term start; assignment-deadline spikes at 23:59; exam-start concurrency when a whole cohort opens the same assessment in the same minute; proctoring video load.

  • Graceful degradation when the video provider, the SIS or the identity provider is slow – a lesson must still open, a submission must still save.

  • Load tests run before each term on a frozen build, with time to fix what they find.

AI tutors, adaptive learning and auto-grading

AI features in education are classified as high-risk under the EU AI Act, and they need evaluation, not just testing.

  • Golden-answer evaluation sets for tutors and auto-graders, re-run on every model or prompt change.

  • Grounding checks: answers trace to course content, not to the open internet.

  • Bias testing across learner groups for adaptive paths and auto-grading.

  • Explainability of why a learner was routed to a given path or given a given score.

  • Age-appropriate content filters tested with adversarial prompts.

How to build evaluation sets and release gates for AI features is covered in our guide to QA strategies for AI solutions.

Automation for academic and learning platforms: what to automate first

Academic software test automation pays back fastest on the suites that protect grades, rosters and access, and there is no single "QA tool for education" that delivers them.

Automate first

1.

SIS and HRIS sync contract tests and daily reconciliation.

2.

Gradebook golden tests, asserted in every view.

3.

Access matrices: guardian × child; role × data; party × document.

4.

SCORM, xAPI and LTI round trips against reference content and real LMS versions.

5.

The assignment-rule matrix on a synthetic population.

6.

The day-one load test, in CI, before each term.

Keep manual. Content review for pedagogy and age-appropriateness; live-classroom facilitation UX; accessibility sessions with real assistive-technology users.

The honest answer to "QA tool for educational institutions." There is no single tool. A working stack has layers: Playwright or Cypress for web, Appium for mobile, Postman or REST Assured for APIs and standards contracts, k6 or JMeter for calendar-driven load, the 1EdTech conformance suites for LTI and OneRoster, SCORM Cloud for content behaviour, and axe plus real screen readers for accessibility. Institutions asking for "a QA tool" usually need a QA process that uses several.

Choosing among elearning testing companies

Among elearning testing companies, the useful filter is evidence that the vendor has already done the things this article describes.

1.

Education case studies with numbers – gradebook, roster, access and standards work, not generic web testing.

2.

Named experience with the standards – SCORM, xAPI, LTI 1.3 Advantage, OneRoster – and the conformance suites that go with them.

3.

A real-device lab that includes school-grade devices, not only current flagships.

4.

Privacy competence in FERPA and COPPA, not only GDPR, with data minimisation treated as a test requirement.

5.

A proof of concept on one flow before any contract.

We work this way: learning platforms, LearnOps and childcare systems since 2010, and the first flow is free.

EdTech testing checklist

Learner

  • Progress is asserted server-side after background, offline and tab-close interruptions.

  • Video and quizzes work on a school-network profile with filtered domains and blocked ports.

  • Deadline timers use server time; large uploads resume.

  • SCORM/xAPI completions round-trip to the gradebook.

  • Captions, transcripts and keyboard operability are tested per content type.

Instructor

  • Roster reconciles to the SIS daily per section.

  • Gradebook golden tests cover every policy combination in instructor and student views.

  • Rubric and assignment changes after grading prompt rather than re-score.

  • Live sessions are tested under network degradation at class scale.

Administrator

  • Assignment rules run against a synthetic workforce with hire-date and time-zone boundaries.

  • Completion reports reconcile to raw events; re-takes count once.

  • HRIS sync is de-duplicated on renames and transfers; prerequisites hold via deep links.

Parent

  • The guardian × child × data-type matrix runs on every build.

  • Consent flags propagate into every report, feed and photo.

  • Guardian removal stops access, notifications and messaging on the effective date.

  • Child-facing surfaces collect only what the educational purpose requires.

Institution

  • LTI 1.3 Advantage conformance plus per-LMS version tests; SSO role mapping is a matrix.

  • Grade passback precision matches the target gradebook.

  • The VPAT is reproduced by audit; day-one load is tested before each term.

AI

  • Golden-answer sets, grounding checks and bias tests gate every model or prompt change.

Conclusion

Three defects cause the most grievances in education software: a lost submission or progress record, a gradebook that disagrees with itself between views, and a parent or student seeing another's data.

None is visible from one persona's screen alone, which is why testing through each pair of eyes finds what feature-by-feature testing misses – and why the gradebook golden suite and the access matrix are the first two things we build on any learning platform.

If you would like to see the method applied to your product, talk to us about a free proof of concept on one persona's day: a gradebook golden suite across your grading policies, or a guardian × child access matrix on your parent app.

Book a strategic QA consultation

FAQ

What does educational software testing include?

Testing an LMS, learning app, virtual classroom or childcare platform from the perspective of each user — learner, instructor, administrator, parent and institution — covering progress and content delivery, roster and gradebook logic, assignment and reporting rules, guardian access and consent, and institutional integration; plus the cross-cutting layers of interoperability standards (SCORM, xAPI, LTI, OneRoster), student-data privacy (FERPA, COPPA, GDPR), accessibility (WCAG 2.2, Section 508), calendar-driven performance and AI evaluation.

How do you test an LMS?

Reconcile the roster to the SIS daily; run gradebook golden tests for every grading policy in both instructor and student views; round-trip SCORM and xAPI content to the gradebook; test LTI 1.3 launches and grade passback against each supported LMS version; run the assignment and prerequisite rules against a synthetic population; test live sessions under network degradation; and load-test on the academic calendar before each term.

How do you provide end-to-end QA testing for higher education software?

By following the student and the course through every system they touch: SIS enrolment → identity provider → LMS → LTI tools → gradebook → transcript, with contract tests at each boundary, reconciliation at each data store, accessibility and device coverage institutions require, and a day-one load model — delivered with a traceable test suite the institution's IT team can review during procurement.

Who offers compatibility testing consultancy for higher education software?

Look for a partner that derives the browser and device matrix from the institution's actual analytics, includes school- and campus-issued devices and network profiles, tests against the LMS versions in use rather than the vendor sandbox, and reports accessibility against the VPAT. DeviQA provides this as part of educational software testing engagements, starting with a free proof of concept on one integration flow.

How do you test student data privacy (FERPA, COPPA)?

Audit every child- and student-facing surface for data collected against what the educational purpose requires; test consent flows, including verifiable parental consent for children under 13; run the role × data access matrix on every build; fulfil and verify access and deletion requests across every system the data reaches; and test retention schedules as rules with evidence.

Is there a QA tool for educational institutions?

Not a single one. Effective edtech QA uses layers — web and mobile automation, API and standards contract tests, calendar-driven load testing, the 1EdTech conformance suites, SCORM Cloud for content behaviour and real assistive technology for accessibility — organised into a process. Institutions asking for one tool usually need that process, in-house or from a partner.

Dmitry Reznik

About the author

Dmitry Reznik

Chief Technology Officer

Dmitry Reznik is the Chief Technology Officer and co-founder at DeviQA, bringing deep technical expertise across software architecture, implementation, and long-term system operation.