Release notes

Open Quizzes guest links now require an expiry. Links created before this release keep working until 2026-09-30; after that every public attempt link must carry an expiry date. Re-issue any link stored outside LEAP.

Paper Quiz grading is grounded in the extracted answer sheet. Upload the marking scheme with the paper and each graded response shows the criterion it was matched against, with a one-click override.

Voice Feedback batches replace one-off prompts. Send a prompt to a section, track who has responded, and read transcripts beside the mastery record for each gate.

GeoGebra and PhET simulations in the materials studio. Interactive simulations are generated per gate alongside mind maps, infographics and podcasts, and embed directly in the lesson workspace.

Integrations catalogue. One page listing every system LEAP connects to, what each connection moves, and where it stands — available, requires setup, in development or planned.

ERP integration architecture published. The ownership model and the five information flows between your ERP and LEAP are documented. Nothing syncs yet — the contract is being finalised with the first ERP vendor.

Cascade edits on Review Extracted show a diff first. Renaming a strand or reordering gates updates dependent learning outcomes, lesson plans and quiz groups in one pass, with a preview before you commit.

The mastery heatmap groups by gate instead of by quiz. Columns follow the knowledge graph sequence, so a gap reads as a position in the course rather than a bad assessment.

Gmail connections must match your registered institutional address. LEAP validates the Google account you connect against the address on your LEAP account, so notifications can only be sent from your own institutional mailbox.

Attendance was lost when a lesson was rescheduled twice. Rescheduling now moves the attendance record with the lesson instead of orphaning it against the original slot.

Proctoring captures are encrypted per tenant. Webcam stills and screen events from Open Quizzes attempts are written with a per-institution key and purged on the retention schedule.

The generation jobs queue drains about three times faster. Materials, quizzes and scripts for a gate are fanned out in parallel, cutting a full course generation from roughly 40 minutes to 14.

Per-institution subdomains are generally available. Admins claim a subdomain during institution setup; branding, sign-in and email links all follow it.

Rubric-based (HUC) assessments dropped the last criterion on save. The final criterion in a rubric was discarded when saving from the quiz builder. Existing rubrics are repaired automatically on first open.

Quiz groups with no questions blocked publishing. An empty quiz group no longer prevents a course from going live; it is skipped with a warning in the pre-publish check.

Session cookies are rotated on role switch. Switching between roles on a multi-role account now issues a fresh session identifier rather than reusing the existing one.

Rubrics are extracted from uploaded marking schemes. Criteria, levels and descriptors are pulled out and mapped to learning outcomes; you edit them inline before approving.

Remediation follows the knowledge graph. When a student misses a gate, LEAP walks back to its prerequisites and assigns the material that covers the actual gap.

The quiz builder was rebuilt around question banks. Questions are drafted once and reused across quiz groups, with per-use difficulty and Bloom level overrides.

CSV question import reports row-level errors. A failed import now returns a downloadable file with the offending rows annotated instead of rejecting the whole upload.

Exam mode could be exited with a browser gesture. Full-screen exit and tab switching are now recorded as attempt events and the timer keeps running.

Report builder queries are cached per institution. Benchmarks across more than 40 sections render in under two seconds on second view.

Conflict resolution missed clashes across sections. Two sections sharing a teacher in the same slot are detected again during timetable generation.

Bulk upload rejected valid enrollment files. Files with a trailing blank row are accepted; the row is skipped rather than failing validation.

The study materials studio generates five formats per gate. Mind map, infographic, podcast, flashcards and a Gamma slide deck, each traceable to the gate and outcome it covers.

AI Tutor is available inside every course. The tutor answers Socratically and refuses to run ahead of the gates the student has reached.

The lesson workspace gains Transcript and Recording tabs. Recorded sessions are transcribed and linked to the lesson plan, so notes and attendance sit beside the delivery.

Grounding is enforced at generation time. Every generated artefact carries the source spans it was drawn from, and generation fails loudly rather than inventing content.

Concept map and misconceptions views in class analytics. Misconceptions are clustered from wrong-answer patterns and mapped onto the concept graph.

Generation jobs are durable across deploys. Jobs resume from their last completed gate instead of restarting when a worker is replaced.

Open Quizzes support public links and guest attempts. Share a link with candidates outside the institution; results land in the same quiz group as enrolled attempts, tagged as guest.

Exam mode locks an attempt down. Server-side timing, full-screen enforcement and per-question navigation rules set by the teacher.

Timetable generation with conflict resolution. Generate a timetable from sections and teacher availability, then resolve flagged clashes one at a time.

Bulk upload handles teachers, students and enrollments in one file set. Validation runs before anything is written and reports per-row outcomes.

Quiz groups lost their gate mapping on duplicate. Duplicating a quiz group now copies its gate and outcome mapping with it.

Tenant isolation checks moved into the data layer. Every query is scoped by institution at the repository boundary rather than in route handlers.

Review Extracted supports inline editing. Course details, strands, learning outcomes, assessments and the schedule can all be corrected before approval.

The faculty dashboard shows gate progress per section. Coverage against the knowledge graph, upcoming lessons and pending generation jobs in one view.

Syllabus extraction handles scanned PDFs. Image-only documents are run through OCR before extraction, with a confidence flag on low-quality pages.

Gate ordering was non-deterministic after re-extraction. Re-running extraction preserves the approved sequence instead of re-sorting alphabetically.

Syllabus upload and gate extraction. A syllabus becomes strands, learning outcomes and sequenced gates mapped to Bloom and DIKW levels.

Lesson plans and Socratic teaching scripts. Each gate generates a plan and a script the teacher can edit before the lesson.

Students learn and are assessed against gates. Progress is recorded per gate, producing the first mastery signals.

Multi-tenant foundation. Institutions, roles and white-label branding read from configuration from day one.