Furikake

Furikake adds furigana (ruby annotations) to kanji on any Japanese web page — Wikipedia, news sites, blogs, forums, storefronts, social feeds — entirely offline. No page text, no reading data, nothing ever leaves your machine. WHY ANOTHER FURIGANA EXTENSION? Existing tools' own user reviews converge on the same complaints, and this one was built against each of them specifically: • Wrong compound readings (一人 read いちにん instead of ひとり) — Furikake's aligner is anchored on JmdictFurigana, a dataset that maps kana spans to individual kanji rather than guessing from frequency alone. • Can't fully undo it — toggling Furikake off restores the page's DOM byte-for-byte, verified by an automated test that cycles it on/off 20 times in a row. • Crashes or lags infinite-scroll feeds like X/Twitter — annotation only touches text currently in the viewport, work is chunked to an 8ms budget per slice, and a guarded observer handles content that loads in later. • Settings that don't stick, no size/colour control — per-site rules and every setting persist across sessions. • Breaks page layout — line-height is compensated so reserving space for ruby text causes zero layout shift, in every reveal mode. DESIGNED TO BE OUTGROWN Furigana pasted over every kanji forever makes you worse at reading it, because there's no reason for your brain to stop leaning on the crutch. Furikake defaults to progressive disclosure instead: • Three reveal modes — always visible, blurred until you hover, or hidden until you tap. • JLPT-level filtering — show readings only for kanji above the level you've already learned, so known characters stop being annotated as you improve. • Confidence signalling — guessed readings (proper nouns, ambiguous compounds) render with a visibly different (dotted-underline) style, so you're never silently taught a wrong reading as if it were certain. MEASURED, NOT ASSERTED Reading accuracy is checked in CI against Aozora Bunko text that already carries human-verified ruby — tens of thousands of real (kanji, reading) pairs, not a hand-picked demo set. The current gate requires 100% on that suite before any change ships. WHAT IT DOESN'T DO • Google Docs — it renders to a canvas, so there's no text in the page for any extension to find. This is a platform limitation, not a bug to report. • Manga, images, or video subtitles — that needs OCR, which is a different tool (see Mokuro for manga). • Machine translation, or competing with dictionary-lookup extensions like Yomitan/10ten on definition depth — Furikake is built to run alongside them without conflicting. PRIVACY Furikake runs its Japanese analyzer as WebAssembly, inside the browser, using dictionaries bundled in the extension itself. Annotating a page makes zero network requests. The only thing stored is your own settings (on/off, reveal mode, JLPT level, per-site rules), in chrome.storage.local, never transmitted anywhere. Open source, MIT-licensed code; dictionary data is credited in full inside the extension's own About screen, per each source's licence.