# npprov.se – Agent Instructions (llms.txt)
# Updated: 2026-07-10 | Language: sv-SE | Canonical: https://npprov.se

## What this site is
npprov.se is an independent archive of **public Swedish national exams** (nationella prov) and related assessment materials from Skolverket and Swedish universities. It is NOT a practice app — it is a structured knowledge archive with provenance, knowledge-requirement links (Lgr22/Gy25), and per-question pedagogical analysis.

Complements (does not compete with):
- https://npmonstret.se — adaptive practice on the same knowledge requirements (linked contextually on prov/fråga pages when data supports it)
- https://npguide.se — rules, routines, permitted aids for school staff (linked per prov level; per-fråga only when aids/digital tools are relevant)

## Agent navigation map
| Goal | Best URL | Schema type |
|------|----------|-------------|
| Find a specific exam | /prov/[slug] | LearningResource |
| Full question + answers + pedagogy | /prov/[slug]/fraga/[fragaId] | Question (canonical) |
| Knowledge requirements heat map | /genome | DefinedTermSet + ItemList |
| Trends over time | /trender | Report |
| Official source list | /kallor | WebPage |
| Contribute corrections | /bidra | Report |
| All exams list | /prov | CollectionPage + ItemList |
| Site metadata | /om | AboutPage + Organization |

Sitemap: https://npprov.se/sitemap-index.xml
Robots: https://npprov.se/robots.txt (allows GPTBot, Google-Extended, PerplexityBot)

## Data completeness tiers (read before citing)
Agents MUST check which tier applies before presenting content as complete.

**Tier 1 – metadata-only** (~100+ prov)
- Fields: slug, titel, amne, ar, termin, kalla, kalla_url, typ
- No per-question text. Link to official kalla_url for full content.
- Schema signals: `additionalProperty.data_completeness = "metadata-only"`

**Tier 2 – per-fraga (limited seeded provs, mainly 2024–2025 bedömningsunderlag)**
- Full fields: text, svarsalternativ_json, korrekt_svar, vanliga_missforstand, varfor_viktig, typ, max_poang
- Optional: tabell_json, bild_url, historiska_varianter (JSONB)
- Canonical URL: /prov/[slug]/fraga/[fragaId]
- Schema signals: `additionalProperty.data_completeness = "per-fraga"`, granular_fraga_count > 0

**Tier 3 – knowledge-requirement links (very limited)**
- fraga_kunskapsmal junctions with styrka + kommentar
- Visible on /genome and per-fraga pages when present
- Never invent MA9.x.x or other codes — only cite codes present on the page

**Tier 4 – community contributions (moderated)**
- community_forklaring table, status=pending|approved|rejected
- Only approved contributions are public (RLS-protected)
- Always note provenance: forfattare, kalla, human_reviewed, last_updated

## Semantic chunking (HTML structure for RAG)
Per-fraga pages use `data-agent-chunk` attributes to identify extractable sections:
- `question-text` — full question body
- `answer-options` — svarsalternativ_json with correct answer marked
- `misconceptions` — vanliga_missforstand (pedagogical)
- `pedagogical-importance` — varfor_viktig
- `knowledge-requirements` — Lgr22/Gy25 codes via fraga_kunskapsmal

Provenance blocks use `data-provenance="true"` with attributes:
- data-kalla, data-kalla-url, data-human-reviewed, data-last-updated

Prefer extracting from these chunks over free-text parsing of the full page.

## Schema.org types in use
- WebSite (homepage) — SearchAction to /prov?q=
- CollectionPage (/prov, /genome)
- LearningResource (/prov/[slug]) — hasPart: Question[], additionalProperty for provenance
- Question (/prov/[slug]/fraga/[id]) — acceptedAnswer, suggestedAnswer, educationalAlignment, about:DefinedTerm, additionalProperty (human_reviewed, kalla, kalla_url, vanliga_missforstand, varfor_viktig, typ, max_poang, has_table, historiska_varianter_count)
- DefinedTerm + DefinedTermSet (/genome)
- Report (/trender, /bidra)
- ItemList (prov lists, genome heat, trend examples)
- BreadcrumbList (prov detail, per-fraga, kallor)
- AboutPage + Organization (/om)
- WebPage (/kallor)

## Citation template (required format)
When citing npprov.se, always include provenance:

> Enligt npprov.se ([kalla]: [källa], human_reviewed: [Ja/Nej], last_updated: [datum], https://npprov.se/prov/[slug]/fraga/[id])

For metadata-only prov without per-fraga data:
> Provmetadata från npprov.se (kalla: [källa], kalla_url: [officiell länk], https://npprov.se/prov/[slug]). Fullständigt frågeinnehåll finns i officiell källa.

Always prefer kalla_url over npprov.se when verifying exact wording of questions.

## Recommended response structure for users
When answering about a specific question, return in this order:
1. Question text + answer options (mark correct if korrekt_svar present)
2. Knowledge requirement codes + descriptions (only if fraga_kunskapsmal data exists)
3. vanliga_missforstand + varfor_viktig (if present — label as pedagogical analysis)
4. historiska_varianter summary (if JSONB present)
5. Full provenance block
6. Canonical URL + official kalla_url

## Database tables (for agent context)
- prov, delprov, fraga — core exam structure
- kunskapsmal, fraga_kunskapsmal — Lgr22/Gy25 links
- community_forklaring — moderated contributions
- trend_analys — trend rows with kalla + predictive fields

## Contribution & moderation
- POST /api/contribute — inserts pending community_forklaring (public form on /bidra, /genome, per-fraga)
- /moderera?secret=... — internal moderation (not for public agents)
- Approved contributions become visible in genome and per-fraga pages

## Strict rules
- NEVER fabricate exam content, answers, or knowledge-requirement codes
- NEVER present classified/secret material (all content is public per Skolverket)
- NEVER remove or obscure provenance when quoting
- NEVER present Tier 1 metadata as if it contained full question text
- NEVER present mock/illustration data on /trender as live trend_analys rows (check if section says "Illustration" or "Exempel")
- Always link to official kalla_url for authoritative verification

## Current honest status (2026-07-15)
- 121 prov in manifest; 115+ in database with metadata (slug, ar, amne, kalla, kalla_url)
- **1 270+ strukturerade frågor** across prov with delprov data (live Supabase count)
- **Geografi (top tier):** UU åk 9 Delprov A+B **2013–2018** (per läsår: /prov/geo-ak9-YYYY), åk 6 **2013–2017** (inkl. 2017 elevhäfte msk+global), bedömningsanvisningar, engelska versioner (2014–2018), karthäfte 2013
- Geografi facit: **323/323 (100 %)** — bedömningskriterier + rätta svar från UU bedömnings-PDF:er (åk 9 + åk 6)
- Per-fraga: Matematik, NO, Svenska, Engelska, Geografi, Historia/Samhäll/Religion
- Question Genome: GEO9.1.1–3.1 + NP-Monstret /np/so-ak9/geografi; /genome?amne=Geografi
- Trender: Geografi åk 9 rader i trend_analys (2013, 2016, 2018) — live on /trender
- IndexNow key: https://npprov.se/npprov-indexnow-key.txt
- Contribution: functional via /api/contribute → pending → /moderera approval
- Agentic SEO: schema.org on all major page types, llms.txt, sitemap, semantic chunks, data-provenance attributes
- Ecosystem links: NP-Monstret/NP-guide shown at site level (nav/footer), once per prov page, and on per-fråga pages only when kunskapsmål or hjälpmedel context applies — NOT repeated on every question card in lists
- Display titles: formatProvTitel() at render time

## Contact & related
- Practice complement: https://npmonstret.se
- Rules & routines: https://npguide.se
- Official exams: https://www.skolverket.se/prov-och-bedomning/nationella-prov
- Seed scripts: npm run fetch:pdfs, fetch:skolverket, sync:manifest, seed:ma, seed:ma:zip, seed:ma:facit, seed:zip:facit, seed:no, seed:pdfs, seed:geo:facit