246 pixels back
One correction first: the LDX page is the panel — the admin menu runs down the whole thing. What's absent there is the step rail, and only because a what's-new table isn't a wizard; there are no steps to draw. So the question is really about the wizard case, and it turns out to answer something I'd already flagged as open.
On the full-scale page I noted that the right-hand aside — the most persuasive thing on the sheet, because it argues against the pre-selected option using the reader's own data — needs about 260px it doesn't reliably have. Killing the vertical rail frees 246px. The two problems cancel, which is a good sign the horizontal rail is right rather than merely different.
Same steps, same content, different geometry
All three keep the tile mark, so the identity survives the move. Note in each that the content area now runs the full width and the aside fits without argument.
SearchSync
The wire is the same 1.5px device as the schematic's, which is why a new rail doesn't read as a new component. Nodes are 30px tiles rather than numbered circles, so the mark keeps doing the progress work. The skipped step stays on the rail, dashed and tagged, with its explanation moved to a single line beneath — see §3.
SearchSync
Costs about 46px of height against A's 88px, and reads as progress immediately. Two real losses: the skipped segment is a 26px stub that cannot carry a label at all, so its explanation has to go into the body copy where it competes with the step's own message; and a segment is not clickable, so nothing is revisitable.
SearchSync
Uses core's real .nav-tab-wrapper, so it is the most native
option available and the greyed tab handles the skipped step for free. It also
stops being a wizard: tabs say "go anywhere in any order", which suits
a flow where nothing is gated and undermines the one thing a first run wants, which is
a sense of finishing. Note the setup is already reachable from the submenu, so
some of this addressability is redundant.
Where the 246px goes
| Viewport | Content area | Vertical rail leaves | Horizontal rail leaves | Effect |
|---|---|---|---|---|
| 1440 | 1240px | 934px | 1180px | Aside fits either way. Cards go from ~215px to ~300px each. |
| 1280 | 1080px | 774px | 1020px | The decisive row. Vertical: three cards at 165px plus a 262px aside does not fit — the aside drops below. Horizontal: it fits. |
| 1100 | 900px | 594px | 840px | Vertical is unusable — the rail is 41% of the content area. Horizontal still holds three cards. |
| < 960 | menu collapses to icons | ~740px | ~860px | Both work; the horizontal rail wraps to two lines, which is graceful. A vertical rail at this width is most of the screen. |
Honest counterweight: the horizontal rail costs 46–88px of vertical space at the top of the panel, which is the scarcest space on an admin page because the notice zone is already eating it. On a site with four plugin notices, variant A's 88px rail may be below the fold entirely — at which point it is decoration with a scroll cost. Variant B's 46px is the safer number for that reason.
The skipped step loses its home
The vertical rail had a sub-label per step, and one of them was load-bearing:
not asked · on file, under Email & diagnostics. That
line is the whole resolved-manifest idea — it shows we chose not to re-ask rather than
hiding that a decision was already made. A horizontal rail has no room for it, so it has
to move. Three options, and they are not equal.
One sentence, its own strip, referring to the skipped step by name. Keeps the explanation adjacent to the rail and out of the step's own copy. Best of the three, at the cost of another 34px of height.
Cheapest, and it competes. The reader is here to answer "where should search run", and the first thing they read is about diagnostics. It also appears on every step or on none, and neither is right.
Free, and the weakest: a greyed tab reads as not yet available rather than deliberately not asked. That is precisely the wrong inference, since the honesty is the entire point of showing it.
A, and drop the vertical rail entirely
Keep B in the back pocket for a two- or three-step flow, where nodes and labels are overkill and 46px is the honest price. And C is worth revisiting for the resumable case — someone who abandoned setup and came back a week later genuinely wants to jump around, and core tabs are the right shape for that. Possibly the answer is A on first run and C on return, which costs one conditional.
Fullscreen §5 Q1 — "does the aside survive a narrow admin?" — answered yes, once the rail is horizontal. And the 246px rail width stops being a number the extension contract has to specify, since a plugin's injected steps now only need a short label rather than a label plus a sub-label.
Still open and unaffected: the six-tile cap. A horizontal rail actually makes it more visible, since all the marks are now on one line where a reader can compare them — five steps of six tiles reads fine, seven steps would read as broken.