Innta

Turn a source into concept notes once, so the source never has to be read again.
Concept notes are infrastructure for flashcards, for Claude's context and as targets for the learner model. The user does not read them, so they are short, not textbook prose.

All user-facing text and notes are Norwegian bokmål, with the English term in parentheses. Never use an em dash.
Paths and course table: _system/config.md.

1. Find sources and course

• From okt startup: /usr/bin/python3 _system/scripts/pdfverktoy.py nye.
• From the læreplan: the lecture's sources (Kilder column) in ~/Nextcloud/skole/ntnu/<mappe>/. For a book, only the chapter pages the lecture covers (find them in the table of contents with pdfverktoy.py tekst <pdf> 1 20).
• Derive the course from the file name and the first page. Ask only if it is ambiguous. Skip non-course PDFs (receipts, shipping labels) silently.
• A PDF that belongs to a personal topic (a book for a code in "Personlige temaer"): file it as in "Personal topics" below and add it to that topic's sources, but do not ingest it now. It must never change the topic of a school session.
• For a module in a personal topic, follow "Personal topics" at the end of this file.
• TDT4136: skip sections listed as not in the syllabus in 10 fag/intro ai/læreplan.md.

2. Read the pages as images

All PDF work goes through /usr/bin/python3 _system/scripts/pdfverktoy.py (run it without arguments for usage). It is the only tool this skill needs permission for.

/usr/bin/python3 _system/scripts/pdfverktoy.py sider "<pdf>" [FRA TIL]    # prints the folder with s-NN.png at 150 dpi
/usr/bin/python3 _system/scripts/pdfverktoy.py bilder "<pdf>" [FRA TIL]   # embedded raster images

Read the page images, not just extracted text: formulas, diagrams and tables live there. Use pdfverktoy.py tekst only as a helper for long text.

3. Split into concepts

One concept per note: something the user must be able to define, apply or recognize on the exam.
Before creating a note, check 20 notater/ (file names and aliases:) for an existing note on the same concept:

• It exists: add only what is missing (figure, formula, source line, emner: entry). Do not rewrite the user's text.
• It does not exist: create it.

4. Figures

For each figure worth keeping: decide the crop box (pixels) on the 150 dpi page image, then:

/usr/bin/python3 _system/scripts/pdfverktoy.py beskjaer <side.png> X0 Y0 X1 Y1 <EMNE> <navn>

This writes 90 vedlegg/figurer/<EMNE>/<navn>.webp.
Never a whole page unless the whole page is the figure. File names: lowercase Norwegian, hyphenated, descriptive (samlebånd-fem-steg.webp). Look at the result image to check the crop.

5. Write concept notes (aim for 10-25 lines)

File: 20 notater/<norsk tittel>.md (lowercase, Norwegian, linkable).

---
type: konsept
fagområde: [<eksisterende fagområde-slug>]
emner: [<EMNEKODE>]
aliases:
  - <english term>
---
<Definisjon i 1-3 setninger, med emnets notasjon og engelsk fagterm i parentes.>

<Nøkkelformel eller nøkkelresultat, i $...$ eller $$...$$.>

![[<figur>.webp]]

**Fallgruver:** <typiske misforståelser, bare hvis kilden eller gamle eksamener viser dem>

**Kilde:** <filnavn i Nextcloud>, s. <side>

## Med mine ord

## Relevante lenker
[[<beslektet konsept>]]

## Flashcards
#flashcards-vent/<domene>

• Content is only: definition, key formula or result, figure(s), pitfalls, source. No textbook-style explanations.
• fagområde uses existing slugs (see 30 kart/ and _system/config.md). Ask before introducing a new one.
• The flashcard tag domain is the note's first fagområde with spaces as hyphens, and always #flashcards-vent/ for new notes.
• Do not write cards. kort does that when the concept is learned.

6. File the source in Nextcloud

For PDFs from ~/Downloads:

1. Name it by the user's scheme: <EMNE>-F<NN>-<pre|inclass>-<norsk-tema>.pdf for lectures (TDT4258-F05-pre-hurtigbuffer-og-virtuelt-minne.pdf), Ø<NN> for exercise lectures, G<NN> for guest lectures. Exams: <EMNE>-<år>-<H|V>[-kont][-l].pdf, where -l is the solution.
2. Copy it with pdfverktoy.py kopier "<pdf>" <mappe>/<slides|bøker|øvinger|eksamener|notater> <nytt-navn>.pdf. It refuses to overwrite.
3. At the end, show one summary line per file (gammelt navn → ny sti) and ask: Slette originalene fra Downloads? (skriv ja). Delete (rm) only after the user confirms. This asks for permission, on purpose.

7. Update state

• 10 fag/<fag>/læreplan.md: add the new concepts to the lecture's row, set status innlest, and add innlest: <fil> s. x-y to the Kilder column.
• _system/elevmodell.md: one new row per new concept: level 0, next action forklar i økt (pretest først).
• If you processed PDFs from Downloads: pdfverktoy.py marker.
• Delete the page images: pdfverktoy.py rydd.

8. Never reread

A source recorded as innlest in læreplan.md is never read again in later sessions. Use the notes.

Personal topics

Called from egen-okt with one module of 15 egne studier/<kode>/læreplan.md. Steps 3, 5 and 8 apply unchanged. The rest changes as follows.

Sources, in this order:

1. The user's books in ~/Nextcloud/personlig/egne-studier/<kode>/. Read only the chapter the module covers, as page images (step 2).
2. The anchor source in the læreplan, at the pinned version. Fetch the relevant pages with WebFetch.
3. Other reputable sources when the anchor is thin: official docs, university course pages, textbooks. Find them with WebSearch. Avoid content farms and SEO blogs.
4. Your own knowledge, to fill gaps and connect the pieces.

If a fetched source and your memory disagree, the source at the pinned version wins. Note the difference as a pitfall if the user is likely to meet the old behaviour.

Notes: the format in step 5, with emner: [<kode>]. Use the English term as the title when practitioners use it untranslated (pod (Kubernetes), ingress (Kubernetes)), and write the note in Norwegian.
One **Kilde:** line per source, so it is clear what comes from where:

• **Kilde:** <tittel>, <url> (versjon <x.y>, hentet <dato>)
• **Kilde:** <boktittel>, s. <side>
• **Kilde:** Claude (egen kunnskap) for content not backed by a fetched source.

Figures: from books as in step 4. From the web, only images with a free licence (Wikimedia Commons), saved to 90 vedlegg/figurer/<kode>/ with the licence in the Kilde line. For structures and flows (architecture, request paths, rock cycle), a Mermaid diagram in the note is often better.

Filing books from ~/Downloads: pdfverktoy.py kopier-egen "<pdf>" <kode> <kode>-<forfatter>-<kort-tittel>.pdf, then ask about deleting the original as in step 6. Add the book under ## Kilder in the topic's læreplan.

State: the module's row in the topic's læreplan gets the concepts, status innlest, and innlest: <url | bok s. x-y> in the Kilder column. New rows in elevmodell.md get the topic code in Emner. Rows written by the diagnostic test in nytt-tema already exist: keep them.