max / quasi

0.31.0: a tab strip fetches its panel, it does not reveal it `dfbc88ce`, ruled by Max 2026-08-18 as option (a). Found writing mnw-server's library strip, which is the first description anyone has written with a labelled `Showing::One` region whose children carry addresses. What it did before: the strip was a client-side reveal, buttons carrying `data-shows`, and every child with a `fed_by` emitted `hx-trigger="load"` regardless of its frame being hidden. So a five-tab strip fetched five panels on page load and pressing a tab issued nothing. MNW's library page does one read and four more on demand, so the description as ruled would have cost it four database reads per view. node.rs said so itself -- "route-bound movement available and unbuilt. Nothing asks for it today" -- and this is what asks. The strip carries the address now. Where a child of a labelled selective region carries `fed_by`, that action goes on the tab button and the child emits no load trigger of its own; both marks sit on the button, so htmx makes the request and the binder moves the frame. Presence picks between the two idioms and nothing reads the region's kind, so a carousel of frames already downloaded is untouched. Uniform across every child, the shown one included. `9b958e7b` is why: a screen renders once at final geometry with no placeholder before first content, and a shown panel arriving on `load` is a placeholder on the one panel being looked at. So the shown panel comes with the screen and carries no address, which is what the Askama page it replaces already does with its include. A tab is a `button` even though its address is a read, which inverts `control_tag`'s rule in the one place it should be inverted: a panel route answers with a fragment, so an `href` to one is a link that middle-click and copy-link would follow to a shell-less scrap. `Fires::ClickInStrip` is that, and it is the second member to suppress the href after `Fires::Load`. The retained-screen hosts get it through the router rather than through their own walk. `Screen::feeds` no longer returns a tab's panels, so a terminal or an egui host stops asking for all five on the way up, and `Slot::asked_for` is the address a strip presses. quasi-tui presses it: Left/Right on a strip answers `Step::Call` for a panel that has not arrived, and `Step::Idle` for one that has. quasi-immediate draws no strip yet, so it takes the `feeds` half and nothing else. The walk still descends past a panel, because a panel that arrived can hold a region that genuinely is waiting, and that one is nobody's tab. Not done, and it is written down on `asked_for` rather than smoothed over: a webview re-reads a panel on every press because the markup keeps the address after `Screen::replace` has dropped it from the tree, while a retained host reads it once. Both agree on the thing that mattered.