AI writing critique for German Abitur preparation: students submit a text, the system evaluates it against 16 formal criteria of the Textform plus teacher-defined task criteria, and returns criteria-based feedback — diagnoses and questions, never solutions, never a grade. These clickable prototypes cover the student and teacher surfaces for the October rollout decision.
Each prototype completes its spec's demo script end to end — submit, evaluate, override, publish.
Every page has a "Demo" pill to jump between states: weak / medium / strong submissions, K.-o. case, second attempts, error states.
All surfaces run on the same mock class (Maria Stuart Vergleich, Deutsch LK 12) — what a teacher overrides is what the student sees.
Nicknames + XplifyID throughout. Staff appear by name. No real student data anywhere.
Task view → upload → staged evaluation → results: per-criterion verdicts with evidence from the student's own text, three-stage feedback (content → logic → language), one clear next step, strengths. No overall score — the tier strip is the only aggregate. Includes K.-o. case, attempt-2 deltas and teacher-reviewed states.
→ !The stakeholder-requested points visualizations: points total with per-layer rings, progress radar, skillset overview, weekly track record. The default results page now carries a compliant version of the teaser; this page remains the fuller exploration for the points debate.
→Textform selection → task starters → task definition with materials and licence labels → AI-extracted qualitative criteria (review, edit, confirm) → sandbox try-out with the exact student result → sign-off and versioned publish. Formal criteria are read-only everywhere.
→ T2Roster with status, K.-o. flags and criterion distribution ("most missed first" — click to filter). Per student: the exact student view side by side with the teacher-only layer (extended reasoning, material quotes, verifier log), audited score overrides and criterion disables, comments, release gate, full history.
→Working rule: every iteration ships with an entry here — why it happened and what changed — so reviews always have a reference point.
Why: initial design contract (specs 00–04, PRD).
Student critique flow, teacher authoring wizard, teacher review & override — one shared mock dataset (Maria Stuart Vergleich), all demo scripts complete, hard rules enforced (no scores, no pasteable feedback, evidence split, pseudonyms).
Why: dedicated teacher-surfaces instruction set.
Staff appear by real name (C. Weber); starter-library filters made functional; verified spec checklists embedded in both teacher files; teacher design notes added.
Why: stakeholder mockups pushed toward warmer, more motivating results.
Compliant adoptions on the default results page (personal framing, progress strips, teacher session card). The points/radar visualizations were built as a separate gamified page so the team can debate them against the no-score rule — declined elements and reasons logged in the design notes. Admin console also built in this period; excluded from this hub pending rework.
Why: first structured client review prioritized results-view clarity and lab-readiness of the task template.
"What this expects" definitions on every verdict; criterion IDs removed from student surfaces; formal vs. task-specific split explained; attempt-2 deltas split into improved / slipped / not changed yet; attempt limits removed platform-wide; AFB markings removed; handwritten-photo submission mode added.
Why: board review (Mathias, Dejan) pushed a condensed, visual-first results screen and a leaner authoring flow. Implemented as “Option A”; deeper restructures (authoring step-merge, full profile model, view convergence) are scoped as the next iteration.
Student results reordered: a visual teaser (radar + points + AI summary) opens the screen; criteria become the drill-down; a fold separates the evaluation from the guidance; “Profile criteria” (task-independent, with avatar) replaces the old competency block; Next Step is merged with the Revise CTA. Task page: criteria now sit behind an AI summary with a show-option; an audio-recording submission mode was added. Authoring: operator field removed, submission settings became three independently-toggleable input modes (document / photo / audio), topic filter moved up to Textform selection. Review renamed to Teacher Analytics. Per client direction, the teaser shows real points. Next iteration: merge “pick a starter” into “define the task,” the full Profile-criteria model, and converging the default and gamified results views.
Why: Figma board (Mathias) — "starters are just pre-defined tasks"; eliminate the separate picker step.
The "Pick a task starter" step is merged into "Define the task" — the wizard is now 5 steps. The merged step opens with an explicit start from an existing task vs. start from scratch choice; the starter library filters by grade level and literature only (topic is already fixed by the Textform); picking a starter loads its task text inline with a "Clear & start blank" escape; and the selected Textform's formal criteria sit right at the top of the step (read-only), so criteria stay visually tied to the template. The operator field, attempts control, and global topic filter were already removed in iteration 5. (Dejan's MVP-planning note is roadmap/presentation work, out of prototype scope.)