DeepTutor: Lifelong Personalized Tutoring. https://deeptutor.info/.
Python ai-agentsai-tutorclawbotcli-tooldeepresearchinteractive-learninglearningmulti-agent-systemsrag
/**
* Where to scroll a transcript list so the active caption row stays readable.
*
* Following captions is not the same as centring them. A row already
* comfortably in view must not be nudged on every cue, or the list creeps
* continuously while the reader is trying to read it. So the row is given a
* band — the middle half of the viewport — and the list moves only when the
* row would leave that band, and then only as far as the nearest edge of it.
*
* Kept as a pure function because the arithmetic is the whole behaviour: the
* first attempt at this subtracted the band from the centring offset
* unconditionally, which pushed the active row three quarters of a viewport
* below where it belonged (off screen, or pinned to the top for early cues).
*/
export function transcriptFollowScrollTop(metrics: {
/** The row's offset from the top of the list's scrollable content. */
rowOffset: number;
rowHeight: number;
/** The list's visible height (``clientHeight``). */
viewportHeight: number;
/** The full scrollable height (``scrollHeight``). */
contentHeight: number;
currentScrollTop: number;
}): number {
const { rowOffset, rowHeight, viewportHeight, contentHeight, currentScrollTop } = metrics;
// The scrollTop that puts the row's centre at the viewport's centre.
const centred = rowOffset - viewportHeight / 2 + rowHeight / 2;
// A larger scrollTop moves the row *up*, so these two positions are where
// the row sits at 75% and 25% of the viewport respectively.
const band = viewportHeight * 0.25;
const nearest = Math.min(Math.max(currentScrollTop, centred - band), centred + band);
const maxScrollTop = Math.max(0, contentHeight - viewportHeight);
return Math.min(Math.max(0, nearest), maxScrollTop);
}