Question and Test Interoperability (QTI) 3 implementation evidence review Reviewed: November 9, 2026 Publisher: Longsight Repository: https://github.com/LongsightGroup/qti3 Commit: aca58bd4cdc2a8c048e4739934f5db3b95f24bab Package version: 0.13.4 Environment used - macOS (Darwin) - Node.js 22.14.0 - pnpm 10.30.1 (qti3 package-manager version) - Vitest 4.1.12 - Playwright 1.64.1 - xmllint for separate XML Schema Definition (XSD) validation Reproduce the review Use a clean checkout of the public repository at the commit above. Install locked dependencies, then run the public release gate: pnpm install --frozen-lockfile pnpm release:check This review ran release:check in the existing clean local checkout, using its installed dependencies. No private certification inputs were configured. The schema checks verify pinned hashes of the official schema sources. If browser binaries are missing, install the browsers before rerunning: pnpm exec playwright install chromium firefox webkit Observed results - Formatting, TypeScript checks, and lint: passed. - Full Vitest suite: 233 files passed, 1 skipped; 3,487 tests passed, 3 skipped, 0 failed. The skipped external-content suite requires official certification content and validator evidence not configured for this review. - Dependency policy, package build, source-map checks, and exports: passed. - Transcoder evidence: 176 executable profile/interaction cases passed. - Information-model gate: 20 reviewed requirements passed executable checks. The gate reports that 1,886 of 1,921 headings have no reviewed requirement; headings with evidence are not fully audited. This is not full conformance. - Semantic mutation gate: all 19 curated faults were detected. - Legacy transcoder XSD gate: 83 vendored schema hashes verified; all 154 profile/interaction instances, 107 variants, and 7 assessment-test serializers passed schema validation. - QTI 3 XSD gate: 278 documents passed official ASI schema validation with pinned source hashes; 433 tests in 29 selected suites passed. These overlap the full suite and are not additional unique tests. - Browser suite: 664 passed, 2 failed, with no retries. The two failures were Firefox browser-launch failures: "Could not find profile folder." Neither reached its behavioral assertions. Chromium and the two WebKit slider checks passed. Firefox slider behavior remains unverified in this environment. - Overall release:check: failed at the browser stage because of those launch failures. Earlier successful checks do not make the complete release gate pass. The final CLI checks occur after the browser stage in the release command. They were therefore run separately against the same built revision: node packages/cli/dist/index.js assert-support node packages/cli/dist/index.js run-fixtures Observed CLI results: 346 support checks passed, 0 failed; all 43 reference fixtures passed, 0 failed. Source references at the reviewed revision Base URL for every path below: https://github.com/LongsightGroup/qti3/blob/aca58bd4cdc2a8c048e4739934f5db3b95f24bab/ - CHANGELOG.md: changes in 0.13.2 through 0.13.4 since the previous review. - packages/migrator/src/lms-migration.test.ts: synthetic Moodle XML and GIFT conversion, expected scores, partial credit, manual grading, source metadata, embedded asset bytes, and refusals of unsupported scoring or content. - packages/migrator/src/legacy-lms-fidelity.test.ts: selected Canvas/Sakai identities, percentage-to-point grades, HTML, manual grading, source groups, and explicit refusal of score-changing programs. - packages/writer/src/human-grading.test.ts: extended-text and upload SCORE remains pending; competing automatic processing returns a conflict diagnostic. - packages/core/src/qti-package-encoded-assets.test.ts: encoded filenames and URL suffixes; malformed encoding, absolute paths, and traversal rejected. - packages/core/src/test-ordering.test.ts: fixed slots, section boundaries, exact saved-order restoration, invalid permutations, and unsupported policies. - packages/core/src/test-ordering-instruction-sections.test.ts: instruction-only sections retained without artificial question references. - packages/core/src/test-rubrics.test.ts: authored rubric scopes and audiences, dynamic-content refusals, and unchanged execution-profile restrictions. - packages/player/test/test-rubric-delivery.test.ts: candidate-safe projection of static instructions, sanitized content, and asset handling. - tests/browser/player-feedback-release.spec.ts: host-controlled feedback withholding through processing and restoration, followed by authorized release. - packages/fixtures/CHALLENGES.md: synthetic item collection and fixed answers. - docs/test-execution-profile-review.md: finite execution profile and refusals. - scripts/check-test-xsd.mjs: selected positive fixture schema checks. - scripts/check-semantic-mutations.mjs: curated faults and behavioral witnesses. Scope and limitations The matrix's Tested status identifies public automated coverage. The results above identify checks actually run in this review. All evidence applies to the fixed source revision, selected fixtures, and stated local environment. Tests use synthetic content. They do not report an institution's exam corpus or independent vendor import, export, rendering, scoring, or round-trip results. Moodle/GIFT conversion is limited to documented supported forms; source metadata preservation does not imply that all source delivery settings execute in QTI 3. Fixed-test ordering and static rubric APIs are separate from the finite branching executor. That executor still rejects unsupported rubrics, timing, nesting, and selection policies. Legacy QTI assessment-test structure is not preserved by the migrator; it returns items in a flat review part with an explicit diagnostic. Feedback withholding controls the rendered DOM. Hosts must separately protect trusted XML, outcomes, assets, preview endpoints, and result access. Hosts also own attempt storage, candidate authorization, timing, accommodations, navigation, scoring authority, review, and LMS integration. This review does not establish WCAG conformance or report screen-reader testing, production load, network or lab-outage recovery, cost comparisons, or security qualification for QFlowLearn or any delivery platform. The release gate is separate from the official certification process. Longsight's previously published application status remains pending; this review does not independently verify that status or establish certification approval. Matrix: https://www.longsight.com/research/qti-3-implementation-matrix/ Delivery guide: https://www.longsight.com/guides/secure-assessment-delivery/ Previous review: https://www.longsight.com/downloads/qti3-evidence-review-2026-10-05.txt