One exchange, one band

Rendered evidence for PR #7482 — furniture and segment headers stop splitting the conversation highlight.

Captured 2026-09-09 from the branch’s own dev server (highlight-band-spans-furniture, commit 8a4683e) · fixture data only, no PMS read and no customer row

A merged person read anchored on one voice call painted as three tinted boxes. Every splitter was a row the thread injects between two rows of the same call — not a message from another one. It now paints as one.

The report

Gera, 2026-09-09, on /people/pers_4a21…/conversation?thread=conv_voice_2944…:

“This one has three highlights, though, but they’re all essentially the same, but they’re being blocked by like some components like listen to transfer or went to voicemail or the calendar like pill. … this could have been one big card background as compared to three.”

Before / after

origin/main — 3 boxes

Three separate tinted bands broken by the transfer divider and the email segment header

this branch — 1 box

One continuous tinted band spanning the whole exchange

Same fixture, same anchor, same viewport — only the two source files differ. The legend reads 6 highlighted on both sides: set membership did not move, only where the band is drawn.

What was splitting it

The transfer divider — “WENT TO VOICEMAIL · Listen to transfer”. It is chrome (isThreadChromeRow), so it is never in the highlighted set however the span is painted — and it cut the band in half.
The __SEGMENT__ header — the “Sep 9 · Email” pill. Clara’s “I emailed the office about this call” follow-up is the same conversation id, 82 seconds later, but a channel flip opens a new section — and a band is drawn inside a section body, so no body-level fold could reach across it.
The call player — furniture of the same kind, handled by the same rule.

Two rungs, because the band is drawn at two: bandBridge buffers furniture inside a section body rather than closing the run, and coalesceBandedSections folds a run of wholly-highlighted sections above it. A real message from another call still breaks the run — it must, or the band would claim rows it is not about.

The border the first fix erased

The bot review caught the second half. .chat-section-header[data-sticky]::before paints a 12px rectangle directly above the header unconditionally, not only while stuck. Rebinding --chat-surface-bg to the band’s tint made it visible: at the band’s own 6px inset it overhung the box as a square-cornered tab, with the top border erased across the header’s width.

Top: the band's top border missing across its width. Bottom: the border continuous after the fix.

Band top edge. Top: colour rebind alone — only the corner arc survives. Bottom: after opening the band by the extension’s own height — border continuous, header with proper air above it.

Measured rather than eyeballed. Sampling the first non-ground pixel at the band’s top edge, at four x positions:

	first non-ground pixel	reading
colour rebind alone	`(236,236,250)`	no border at all
after the geometry fix	`(201,198,246)`	the 22%-primary border

The stuck state still works

The extension exists to cover the scroller’s own top padding when the header pins — without it a bubble shows for those few pixels and is then chopped by a hard edge. Scrolled so the header sticks:

The segment header pinned at the top of the scroller, band tint continuous behind it

Header pinned at the scroller top, band tint continuous behind it, no chopped text and no pale strip — the rebound variable doing its job in the band’s own colour.

What deliberately did not change

Set membership. A bridge row is never stamped data-highlighted; the legend prints the same count it did.
The negative control. A real message from another thread still breaks the run.
The work-order lens. A partially-highlighted section is not “banded”, so its slice keeps the existing body-level path byte for byte.
Segmentation. Nothing about which rows form a section moved.
The tail case. A body-level band meeting a wholly-highlighted section stays two boxes — merging them lifts part of a body out of its own section wrapper and the header above it stops sticking.