Reading

The reader presents an open document as e-reader-style pages of clickable text in the active language — Japanese or an installed language pack — with a dictionary panel on the right. It opens from the library or the home page's continue-reading card; the Reader icon in the rail is enabled only while a document is open.

Pages and navigation

Text is paginated to fit the window — there is no in-page scrolling. To turn pages:

• scroll the mouse wheel,
• press PgUp / PgDn, or
• click the ◀ / ▶ buttons in the bottom bar.

The bottom bar shows page N / M and a thin progress strip representing how far into the book the end of the current page is. Resizing the window repaginates the book and keeps you on the paragraph you were reading.

ArrowRight / ArrowLeft (rebindable in Settings → Shortcuts) move the selection to the next or previous phrase group, following it across page boundaries.

Clicking words and phrase groups

Every token in the book's language is clickable. Tokens that belong together are selected as one phrase group, highlighted in a single color:

• Conjugated verbs select with their endings. Clicking 読んでいる selects the whole phrase, and the panel's Form box explains the conjugation component by component (te-iru, polite past, passive, causative, …).
• Nominal compounds (noun + suffix, prefix + noun) are tried against the dictionary as a single word. When that succeeds, the compound takes the headline and the clicked token is shown below as its component.

Selected words are always highlighted. An optional checkbox in Settings → Reading additionally tints every unknown word in the text.

The dictionary panel

The right side panel shows the headword with furigana positioned over each kanji run, the surface form from the text when it differs from the lemma, part of speech, current knowledge status, corpus frequency rank, usage register tags (colloquial, formal, archaic, …), and the numbered dictionary senses — JMdict for Japanese, the pack's dictionary otherwise — with cross-references and antonyms.

Pack languages add a few panel elements of their own:

• Decoded parse — in pre-annotated texts (the Koine Greek pack), an italic "this form: …" line decodes the clicked occurrence's stored parse to prose.
• IPA — with Settings → Reading → Pronunciation → "Show IPA with dictionary entries" enabled (off by default), entries from packs built from Wiktionary show their IPA under the headword. Japanese readings are unaffected.
• Contractions — a fused function word (French au = à + le) shows its components under the headword, each a clickable button that looks the part up in the Dictionary-and-Kanji view. Leaving the reader this way credits the page like a pause, same as the kanji chips below.
• Compound splitting — when a word in a Germanic pack has no entry of its own, the panel shows "compound of: …" with each part clickable (Arbeitsmaschine → arbeit + maschine): no entry for the whole word, but its parts are in the dictionary.
• Ambiguous forms — when a form has more than one possible analysis, an "Ambiguous form" box lists every candidate (lemma plus decoded parse); one click re-points that single occurrence — the manual override above the frequency vote.

The panel's buttons assign a knowledge status:

Buttons only appear when they would change something (e.g. no Ignore button on an already-ignored word). Shortcuts are rebindable in Settings → Shortcuts.

Kanji chips: under the headword, one small button per unique kanji in the lemma. Clicking a chip opens that character's card — readings, meanings, stroke order — in the Dictionary-and-Kanji view. Leaving the reader this way credits the partially read page like a pause.

If an LLM backend is configured (Settings → AI), an Explain this sentence button (shortcut E) asks it about the selected sentence. Without a dictionary installed, the panel says so explicitly — distinct from "no entry found for this word".

Furigana and glosses

Settings → Reading → Furigana offers four modes:

X is set next to the mode (1–50, default 3). The first-X mode is instance-anchored per book: occurrence indices are computed in reading order when the book opens, so the same X occurrences carry furigana no matter how you flip around or resize the window — later occurrences never do. X applies independently to each book.

The unknown-based modes track status live: assigning any status (learning/known/ignored) removes a word's furigana everywhere. For Japanese, furigana is drawn only over tokens containing kanji, and a conjugated stem keeps just the reading of its kanji run (走っ from 走る shows はし, not はしる).

In pack languages with pre-annotated texts (the Koine Greek pack), the furigana slot doubles as an interlinear gloss layer: each token's hand-verified gloss is drawn over it, governed by the same four modes and the same live status tracking — no kanji requirement applies.

Reading position

Your position — the first sentence of the current page — is saved on every page flip and again when the app closes. Reopening the book returns you to the saved page.

Reaching the last page saves a one-past-the-end position, which the library reads as 100% (finished): it completes the progress display and unlocks the finish sweep in the book's info panel. Opening a finished book again lands on its last page.

Pause and the reading clock

Reading time and velocity (characters per minute) are tracked per sitting; see Statistics. The ⏸ button in the bottom bar, or its shortcut (default P, rebindable in Settings → Shortcuts), pauses the clock and shows a "Reading paused" overlay over the dimmed page. Any click or key press resumes — and that resuming input is swallowed, so it cannot flip a page or trigger a shortcut.

Auto-away: any interaction anywhere in the reader — a click, a scroll, a key press — resets an idle timer. When idle time on a page reaches 2× the page's expected reading time (its character count ÷ your measured velocity; never less than 20 seconds), the paused overlay appears with a 5-second grace: re-engaging within 5 seconds means it was a hard page, not an absence — the overlay is dismissed and the clock keeps running with full credit. Otherwise the absence is treated as real and the clock stops. Before a velocity statistic exists, a flat 5-minute idle threshold is used instead.

What gets credited when a page visit ends:

Finishing a book

When a book reaches 100%, its info panel in the library offers Mark remaining words as known. The sweep sets every word still at the default unknown status to known; untouched proper nouns become ignored (names are not vocabulary). Words you explicitly marked while reading — learning, known, or ignored — are never overwritten.

Before applying, the confirmation dialog flags suspicious words you plausibly never actually learned. A word is suspicious only when both signals fire: its corpus frequency rank is far beyond where your known vocabulary lives (or it is absent from the frequency list entirely), and it appears at most twice in the book. Suspicious words are excluded from the sweep by default — a false "known" hurts more than a false "unknown" — with per-word checkboxes to include them. The rarity signal needs at least 50 known words before it activates. See Reviews-and-SRS for how newly known words affect scheduling.