Skip to content

Whitepaper visual refresh — 21 September 2026

Editorial image decisions

  • Lead with learning-house.png: mobile guided learning and prerequisite-based progress as a solid house, requested by the Product Owner. Its shared text-free illustration serves both languages; the caption explains the current Claude Mathematics/Physics upper-secondary scope. The phone is an illustrative coach metaphor, not a fabricated product screenshot or a claim of automatic mastery. Its exact prompt, provider, hash and review are in learning-house.provenance.json.
  • Keep the original introductory and Champion comics. Place the original intro in section 1 beside the shared-curriculum/different-starting-points explanation.
  • Add learning-moment.png in both languages: a text-free, friendly learning story showing a starting sketch, an initial attempt, and a completed explanation. The flat notebook faces the student; only the raised notebook faces the audience. It is an AI-generated illustration, not observed learner evidence or a product UI. The final prompts, provider, byte hash and visual review are recorded in learning-moment.provenance.json. Rejected drafts are not publication assets.
  • Keep the Cockpit as implementation evidence. Put the recall-status and route figures in the technical-detail appendix, linked from their explanations, rather than interrupting the main visual story. Remove obsolete ChatGPT coach/handoff screenshots and inaccurate older architecture/coach diagrams from the narrative, not from the historical assets.
  • Leave the long configuration-form capture out of the opening; show the actual two-subject planning component instead of the generic planning schematic in 3.4.
  • Retain the content-selection/material captures as technical documentation. Section 5.2 now focuses on the strategic direction and its narrative illustration, rather than on short-lived details of the PoC interface.
  • All displayed figures can be opened at original size in the WebGUI, including on small screens. Existing artwork is retained; no original cartoon is overwritten.
  • A single italic AI-generation notice appears beside Version/Project in the document header: “Ein Teil der Illustrationen ist KI-generiert.” / “Some illustrations are AI-generated.” It replaces repeated per-image labels in both languages and the PDF. Keep provenance out of narrative captions; documentary UI screenshots are not labelled as AI-generated. Asset-level provenance and generation metadata remain available in their own records.

Six-station visual story

The main illustrations now follow the same sequence in both full Whitepapers and the short storyboard.de.md / storyboard.en.md picture-reading editions:

  1. Build on solid knowledge: learning-house.png opens with mobile learning, personal effort and prerequisite-based progress.
  2. Shared curriculum, personal path: retain the original introductory comic; typeset lead and caption carry the message even when its embedded copy is small.
  3. Think, get help, explain: learning-moment.png keeps the learner's own work at the emotional center.
  4. Progress stays connected: new learning-core.png separates the phone coach from the stable SkillPilot curriculum/progress/plan board. The connecting tokens stand for structured state exchange, not storage of complete dialogues.
  5. One goal, different materials: new learning-materials.png visualizes book, explanation and exercise choices around one goal. The in-development vision label precedes the image. The text describes teacher/learner selection, providers connecting materials to goals and plans, and the starting point with Physik Libre. Fast-changing pilot details stay in the technical documentation.
  6. Improve it together: the existing Champion comic closes the story, followed by goal-linked feedback and the free SkillPilot / paid Claude Pro beta invitation.

learning-core.provenance.json and learning-materials.provenance.json record exact generation/edit prompts, tool, selected output IDs, hashes and review. Both are built-in image-generation PNGs, copied unchanged with their metadata. The material draft was corrected before use: flat notebook/book drawings now face the learner, unlike the audience-facing conceptual material cards.

The localized reading-view selector keeps full text as the default and offers Die Geschichte in Bildern / The story in pictures at ?view=story. Both editions retain one italic header disclosure and original-size image links. The short edition contains the six narrative figures only; the full version adds real UI captures and technical detail. No existing cartoon was overwritten.

Read-only independent AI review checked the six actual images and DE/EN copy for story continuity, learner agency, paper orientation and current/vision distinctions. This is editorial review, not a human comprehension study or evidence of educational effectiveness.

Local publication checks

  • Actual DE/EN pages in both reading views at 375px and 1280px: six story figures / ten full-text figures load, no horizontal overflow or page errors, one italic header disclosure, and links to the original images.
  • UI regression tests: reading mode, language changes, late fetch responses, retained query/hash state, keyboard access and existing media playback.
  • TypeScript, scoped ESLint, Markdown links, documentation indexes and the AI-transparency inventory pass. Four illustration provenance hashes match source and runtime PNGs byte-for-byte; original comics remain unchanged.
  • German full PDF regenerated and rendered for inspection. Overlong German expressions were rephrased without changing their meaning; a text-box check confirms that no words reach the outer 40pt page margins.
  • Publication copies are synchronized locally. These checks are not a production deployment, native-mobile acceptance, or a test with actual readers.

UI capture scope

The bilingual captures below show current application components rendered locally with explicit synthetic example data. They are illustrative UI captures, not screenshots of a real learner, backend integration tests, production evidence or Claude/ChatGPT host acceptance.

Reproduce

Start the application Vite server on loopback, then run:

npm --prefix app run dev -- --host 127.0.0.1 --port 4183 --strictPort
WHITEPAPER_CAPTURE_BASE_URL=http://127.0.0.1:4183 app/node_modules/.bin/tsx app/scripts/captureWhitepaperUi.ts

The capture harness is app/scripts/captureWhitepaperUi.ts; its page and React fixture are app/scripts/fixtures/whitepaperUiCapture.{html,tsx}. Each image is captured at an 800 CSS-pixel viewport width and device scale factor 2, with the application's real fonts, styles and German or English language context.

Learning plan

learning-plan-ui.de.png and learning-plan-ui.en.png render the production LearnerPlanTodayOverview and its child components. No application component or native control is restyled or replaced for the capture. Action callbacks are deliberately inert: the fixture cannot continue or switch an actual learner's session. Plan details can still expand using the component's native controls.

The synthetic state matches the Whitepaper's illustrative table:

Subject Scheduled through today Achieved in total Today Separate balance
Mathematics 13 9 1 of 3 2 learning goals behind
Physics 10 11 2 of 2 1 learning goal ahead

As in the current backend formatter, a fulfilled daily target is labelled “Tagesziel erreicht” / “Daily target reached”, not a newly invented numeric label. The current goal remains “Lineare Gleichungen lösen” / “Solve linear equations”. Both continuing mathematics and switching to physics remain enabled. The capture asserts the exact localized status, direction, active goal and enabled actions; it does not independently recompute the backend's plan balance. No learner API request is permitted for these planning captures.

Optional content

The capture described below predates the Product Owner's access-model clarification later on 21 September 2026. Its credential-field and default-off observations are historical capture evidence, not the current UI contract. Material selection now uses the ordinary Cockpit profile access without a separate credential; content support defaults to on while package selection remains explicit. See the current PoC runbook.

content-materials-ui.de.png and content-materials-ui.en.png render the production MaterialSelectionPanel and GoalAdditionalMaterials. The fixture reads the real public package metadata from content/physik-libre/1.0.0/package.json. Its current four material links map to four Physics goals; only the one material bound to goal ae67bcf1-f3ee-50d6-9a12-25a159dff659 is shown below the package selection. The German title of that external resource remains German in the English UI, accurately reflecting the available material language.

Only the synthetic profile whitepaper-demo exists in the fixture. The two permitted read requests (content selection and goal materials) are intercepted and fulfilled locally. All other learner API requests, every API write, and external network requests are blocked. The configuration credential field stays empty; no real profile ID, grant, session, learning progress or chat is accessed. The external link is displayed but never opened by the capture.

The example shows an opted-in package as an illustration of the implemented PoC; it does not imply that optional content is enabled by default or in production, that all Physics goals have material, or that a provider partnership exists. See the content PoC runbook for real activation, authorization and outstanding host acceptance requirements.

Checks performed by the capture

  • Exact German/English planning status and active-goal assertions.
  • The real four-link package inventory and exact one-goal material binding.
  • Only the expected synthetic reads, no private API or external requests.
  • Empty configuration credential field, no page errors and no horizontal overflow.
  • PNG output paths, SHA-256 digests and CSS dimensions reported by the script.

These checks validate the rendered fixture and image provenance. They are not a replacement for ordinary application, backend, privacy or provider-host tests.