{# NO HEADING OF ITS OWN, and no close: this panel is the BODY of the card core builds for a loaded tool, and that card's header already carries the name, the key that opens it, an eye and an X. See views/toolLoader.js. NO `data-tool-extras` either, and that is the change. The CSV pair used to be lifted into that header as two unlabelled arrows whose only account of themselves was a tooltip -- the same mistake the base image's channel-rename button made, and fixed the same way. They are labelled actions on a line of their own now, at the top of the panel, exactly as Opacity and "Upload channel names" sit at the top of the image card's body (see index.html). Loading and saving gates act on the whole tool rather than on any one marker, so the line comes before the marker picker and everything below it. #}

{# ONE LINE: `MARKER [ CD3 v ]`. The caption stood on a line of its own above the picker, which in a 300px sidebar spent a whole row on six characters -- and the panel below it is all one-line controls, so the stacked pair read as a different kind of thing. `.control-row` is core's label-beside-control shape (viewer.css), the same one the Point size and Opacity rows use. #}

Marker

{# THE WHOLE GATE ON ONE LINE: `0.42 ---o=====o--- 9.10 *`, which is the image channel's contrast window exactly (viewerSidebar's createChannelSlot). The two thresholds had a row of their own above the track, and Auto Threshold a full-width button below it -- three lines for one control, in a 300px sidebar, when the numbers are three to six characters each. The numbers are the slider's own `.plx-number` boxes, inline at the two ends of the track they name: drawn as plain text until they are clicked (`.plx-slider.is-plain-numbers` in main.css), still typeable, which is what a gate copied off a paper or from another sample has always wanted. The button at the end is Auto, and afterwards the way back from it. #}

{# WHERE THE THREE INPUTS DISAGREE, if they do. A table, a mask and an image come from three steps of a pipeline and nothing in any of them says they belong together, so pairing a table with the wrong image, or a mask exported at another resolution, produces a panel that works and is answering about a different sample. Directly under the distribution, because the distribution is the thing being doubted: this is the marker's own numbers, and this is what may be wrong with the frame they are in. Quiet by construction -- muted text on the soft amber DESIGN.md keeps for a non-blocking warning -- because every one of these states is one a real project is allowed to be in. See core's models/consistency.py, which decides them; this panel only shows them. #}