max / quasi

Read every branch, loop and arm off the marks, and delete the comparisons The containers rustc named now carry marks, so the mechanism reaches the whole corpus: Table (columns, then rows, then the pager), Field's options, Node's five list-holding shapes, and Slot and Row from the first pass. The renderers record where they wrote each member, and where assembly belongs to makeover-webview -- a row's cells, a chart's bars -- it is asked rather than searched, through cells_html_placed and chart_html_placed. The whole workspace is green on this. quasi-bench's residual tests are the real check: they compile MNW-shaped screens and compare the filled residual against the renderer's own output, and they pass with nothing diffing anything. Two things follow that are worth stating plainly. A guard no longer changes what the twin draws. It places its members and marks them, so one render answers every guard, and the test that asserted the opposite now asserts this. An arm may hold loops and guards of its own. Each arm is compiled from its own render, where its own marks were recorded, so "the arms are flat" is gone rather than worked around. PARKED, precisely: a guarded SETTING. A mark covers a run of members and a setting varies markup inside the member its own statement produces. Two shapes hide under that -- placing settings (figure, bar, more, back, forward, jumping), which are members in all but name and want counting in their container's index space, and settling settings (chosen, here), which want two arms over the member they settle. Four bench shapes use one and are unstaged until it exists, behind the guarded-settings-stage feature so the shape and its residual tests come back together. symbolic's PLACED list is deleted: the question it answered cannot be asked of a mark.