Graphle

Match words to their Google Books Ngram graphs.

The game: four graphs across the top, a slot under each, and the word bank below. Three guesses. Each submission says only how many are right; the y-axis scale is revealed after one wrong guess, and after two a graph the player already had right is confirmed. Past guesses are colored by whether the current arrangement is still consistent with them.

Two side pages are linked discreetly in the footer:

• Rules & sharing — words per round (3–5), guesses, feedback style, hint timing, graph variety (R² cap), word frequency floor, opening view, and hand-picked words. Every setting is encoded in a share link.
• Saved & deleted — words starred with ☆, and words removed with ✕, both kept in the browser and exportable as text.

Rules live in the URL query (n, g, fb, lg, mp, r2, pf, h, and p for hand-picked words). Hand-picked words only apply to the first round.

Run locally

fetch doesn't work over file://, so serve the folder:

python -m http.server 8000

Then open http://localhost:8000.

Word list

Words and their definitions live in themed files under scripts/words/, one entry per line:

hornswoggle | verb | To swindle, cheat, or hoax someone.

Only single lowercase words work (no hyphens, spaces, or digits). After editing, rebuild the data:

node scripts/fetch-ngrams.mjs

This writes data/ngrams.json and data/definitions.json.

node scripts/fetch-books.mjs writes data/books.json: a book published around each word's peak whose title contains the word, used by the "Peak-era book" hint. Raw Ngram responses are cached in scripts/.ngram-cache.json, so only new words are downloaded. Words peaking below 50 per billion words (MIN_PEAK in the script) are dropped and listed in the output.

The Ngram endpoint doesn't allow cross-origin browser requests, so the data is fetched ahead of time and served as a static file.

Default rules and the hint catalog are in rules.js. When you change the game files, bump the ?v= numbers in index.html and game.js so browsers fetch the new versions.

Deploy to GitHub Pages

Push to a GitHub repo, then Settings → Pages → Deploy from branch → main / root.

Ads

Three reserved slots, all at standard sizes and all below or beside the game, never between the graphs and the words:

• bottom: 728×90 leaderboard under the game (320×100 on phones), on every page.
• rail: 160×600 skyscraper in the right-hand margin, only on windows wide enough that it can't reach the board (1600px for 4 words).
• bottom-2: a second banner under the first, standing in for the rail on narrower screens.

Until AdSense is set up they show house banners for Word Salad and fillmein. To go live, put the publisher ID and each slot's ad-unit ID into ads.js. Any slot left blank keeps its house banner.

Pick words

A third side page for building a round by eye: scroll the whole word list as graphs, search words and definitions, sort by how common a word is (or A–Z, or shuffled), and click graphs to fill the tray at the top. Once it holds a full round, play those words or copy a link to them. Cards load a page at a time as you scroll, so the several thousand graphs don't all render at once.