Glömskan

Det som ingen längre nämner, försvinner.

A Swedish vocabulary card game where forgetting is the enemy. Every card is a word. You play a card by producing the word from a Swedish clue, and a spaced-repetition scheduler (FSRS) decides which words are in your deck — so the game loop and the learning schedule are the same thing. Built for Swedish schools (högstadiet and gymnasiet), entirely in Swedish, with no accounts and no server: progress stays in the student's browser.

▶ Play: https://kunskapshunger.github.io/glomskan/

What's playable now

A first slice, to find out whether the loop is fun:

• A run of four encounters and a boss. Each Tystnad is drawn from the letters of its own name, and every hit erases some of them.
• 50 hand-written B1–B2 words (påverka, trovärdig, ifrågasätta, förmodligen …).
• Word class is card type: verbs attack, nouns defend, adjectives strengthen the next card, adverbs draw cards.
• The task grows with your memory of the word: choose among four → type with a first-letter hint → type freely → type the right form in a sentence. Stronger memory, stronger card.
• Words you got right come back later; words you missed come back tomorrow.

Not in the slice yet: placement test, branching map, relics, grammar combos, the teacher page, save codes and sound. See the design spec for the full game.

Keys: 1–5 pick a card · E ends the turn · Enter answers and continues · Esc closes the answer sheet.

Answer judge (optional)

When a typed meaning answer is wrong by the card but might be a synonym the card doesn't list, the game asks Jev (TypeSafe) through a Supabase Edge Function for a second opinion. Only the card (word, definition, sentence) and the typed answer are sent; nothing about the student. The judge is on when content/config.json has a judgeUrl, and off without it. Deploy steps: supabase/README.md.

Run it locally

npm install
npm run dev     # → http://localhost:5174
npm test        # unit tests, including a check of every word card
npm run sim     # 60-day simulation of the learning engine

No build step: the browser loads native ES modules, and ts-fsrs is served from vendor/ through an import map.

Docs

• Design spec: docs/superpowers/specs/2026-09-25-glomskan-design.md
• Learning-engine plan: docs/superpowers/plans/2026-09-25-plan-1-learning-engine.md

Credits

• Scheduling: ts-fsrs (MIT), vendored in vendor/.
• Word selection is guided by the Swedish Kelly list (Språkbanken, CC BY 4.0).
• Fonts: IM Fell English and Spectral, via Google Fonts.