Afterword — RSVP with peripheral context
A mobile-first experiment in Rapid Serial Visual Presentation (RSVP), with an approximate Spritz-style optimal recognition point and two competing fields of context. Previous language sits above the anchor, and upcoming language below it. The context is not merely decoration: it invites prediction and regression without losing the central reading point.
Touch
Tap the left third of the reading surface to go back one word, the right third to step on, the centre to start or pause. Stepping on is also how you set the pace (see The beat). Drag horizontally to scan as you'd drag a page (leftward brings on what's next), reversible from the original touch-down position. The timeline seeks anywhere, and ±10 buttons make coarse steps. The whole player folds away (⌄) to a slim bar with play, progress and the way back (⌃), leaving the page to the text. Keyboard arrows and Space work on desktop (→ steps on, and sets the pace, as a tap does).
The beat
The reader runs on a steady beat, shown top right in beats per minute with a dot that lands on each word. Every word takes one beat (a long one, eleven letters or more, a beat and a half), and punctuation adds rests in whole half-beats: a comma half a beat, a clause or sentence end one, a paragraph end two, all scaled by the rests setting. So the pulse never wavers: the words, the clicks and the bass all fall on the same grid, and it doesn't drift (each onset is counted from the last, not from a chain of timers). The tempo is set by stepping: tap forward through the words (the right third, › or →) and, from the third step within two and a half seconds of the last, the rhythm of your steps becomes the BPM, each interval counted against the beats the word it left holds (so stepping off a sentence's end, with its rest, counts as the longer beat it is). Playing, your step lands at once and the beat runs on from it at the new tempo, so you can tap along to speed it up or hold it back; paused, Play goes on at the pace you stepped. It holds until you step again or move the slider; stepping back, seeking and pausing leave it as it is. Each start (Play, after a pause or a seek) eases in: the first few words (the warm-up, 4 by default, 0 to 8) held longer, the first by 1.6×, coming up to the tempo by the last.
Reading
The anchor letter is determined approximately by word length: first for 1-character words, second for 2–5, third for 6–9, fourth for 10–13, fifth for 14+. Typographic widths are accommodated by laying out characters before and after that anchor on opposite sides of a fixed centre. A word too long for the page at your size is drawn smaller, that word only. Modes: Off, Nearby words, Extended passage. Context strength and span are adjustable.
The context reads as prose flowing through the word: the previous words sit just above it, set right, their last line ending where the word begins; the next words sit just below, their first line starting where the word ends and running on from the margin (when the word leaves under 30% of the width beside it, the line is a whole one of its own). The passage starts at a sentence, never mid-sentence at the top edge, and a blank line in the text is a paragraph break there, as on the page.
Writing
At the end of the text the reader becomes a pen. Step past the last word, or let playback reach it, and the current word is what you type: it stands where every word stands, its anchor letter marked, a caret after it, and the text so far runs into it from above as the previous words always do. A space sets the word down and the next begins; Return sets it down and starts a paragraph (a ¶ shows it's coming); deleting past a word's start takes the last word back up to edit. The tap zones still work: the middle brings up the keyboard, the right sets the word down, the left (or ←, or Escape) steps back into reading. What you write goes into the text, and so into the link.
Comprehension check
Check (in the player) runs a small test. Passages are made from templates whose facts are known, five sentences each: who gave whom how many of what, what colour thing is on what, who arrived after whom and before whom, who works as what and where, who did not take one thing but took another, who met whom on what day and where. Each sentence gets one question with four answers, the wrong ones of the same kind (names from the same passage among them), so none can be ruled out by type and chance is 1 in 4. Pick the speeds (200 to 800 bpm) and how many passages at each; the passages come in a shuffled order so practice favours no speed. Each plays once at its speed with your context and sound settings, untouchable (no stepping or pausing, so the speed is the speed), then its five questions, with no word on right or wrong until the end. Then a score at each speed, the passages with your answers, and the last few runs on this device (kept in this browser only). Your own text, place and tempo come back afterwards.
Light and dark
Auto (as the device is), Light or Dark, from the player; kept in the link as theme. Dark is warm ink on slate, the anchor's accent lifted to read on it.
Audio
Click marks every word onset. Bass is a soft synthesized falling sine pulse on the words that start a clause, or a sentence, or (as a comparison) every 2, 3, 4, 6 or 8 words. Both combines them. The sounds follow actual word onsets, including punctuation delays; they are not independent beats that drift away from the text. User-gesture play is required to enable Web Audio. No external assets.
Shareable state
Everything is held in the URL fragment: \#text=…&i=…&bpm=…&mode=…&span=…&opacity=…&pause=…&font=…&sound=…&volume=…&group=…&warm=…&ui=…&theme=…\ (an older link's \wpm\ is read as the bpm). The address is rewritten when you pause, seek or change a setting, not on every word as it plays. The position and settings travel with the source text, and no localStorage/server data is needed. The URL itself contains the source text: do not publicly share private text. Long passages make long URLs. Playback does not auto-start on reload due to mobile audio restrictions.
Research questions
Does subdued preview help prediction without creating attention competition? Does previous-word context make regression useful? When does visibility harm comprehension? Can an audible pulse scaffold reading rhythm, or impose artificial grouping (bass on clause starts against every N words)? Does a steady beat with rests read better than durations that vary word by word? The Off mode provides a minimal control condition. This is an interaction study rather than evidence for higher comprehension.