# Page Loading

When a page counts as loaded, how the scroll loop captures dynamically loaded content, and when it stops.

## Settling

Before anything is extracted, the page settles in two steps:

1. `WaitDOMStable` with a window of `min(400ms, IdleWait)` and a 0.01 change threshold, capped at `IdleWait` overall
2. `SettleJS`, capped at `IdleWait`; the default `listener.js` resolves after one `requestIdleCallback` (250ms timeout) or after 1 second, whichever comes first

Settling runs again after each consent pass and after each scroll. Timeouts here are not errors: extraction continues with whatever the DOM holds.

## Scroll loop

After consent handling, the first snapshot is the full page HTML. Then, up to `ScrollCount` times:

| Step | Detail |
|---|---|
| Check | Stop when the document is no longer taller than the viewport plus 4px |
| Wait | 150ms plus 0–300ms of random jitter; stops if the context ends |
| Scroll | Smooth scroll to the bottom over 300ms with `requestAnimationFrame` |
| Settle | Same two-step settle as above |
| Snapshot | Stop when the new HTML is byte-identical to the previous snapshot |

`ScrollCount` defaults to 3; a negative value disables scrolling entirely, leaving only the first snapshot.

## Why multiple snapshots

Infinite-scroll feeds and lazy-loaded articles replace or append content as you scroll. Each snapshot captures one state, and the output stage combines them: `TypeHTML` merges their `<body>` children, while `TypeMarkdown` and `TypeJSON` run readability on each snapshot and join the article contents. Repeated paragraphs introduced by overlapping snapshots are removed afterwards. See [Output Formats](/configuration-output-formats).

## Viewport

The page uses `Option.Viewport`, 1280 × 960 by default, with `DeviceScaleFactor` 0 treated as 1. A taller viewport loads more content per snapshot.