Finding: a record:highlights chip clips multi-element cell renderers with NO ellipsis (the residue of #5066)

Migrated from objectstack-ai/objectstack#6950 under the file-at-destination ruling (objectstack-ai/objectstack#7167, maintainer 2026-08-10). Originally filed 2026-08-09T06:13:43Z. The full prior thread — including triage rulings and hold/restart conditions — remains on the source issue and MUST be read before acting on this card.

What this is

Observation-class finding split out while fixing objectstack-ai/objectstack#5066 (objectui PR for the percent renderer). Filed unassigned, no pm:queue — nothing here is known to be hit by a user today, and the one instance that WAS hit is fixed by objectstack-ai/objectstack#5066.

The mechanism objectstack-ai/objectstack#5066 fixed only one instance of

HeaderHighlight (objectui packages/plugin-detail/src/HeaderHighlight.tsx) wraps every cell renderer's output in a single clipping span:

block min-w-0 truncate text-sm font-semibold

truncate is overflow:hidden + text-overflow:ellipsis + whitespace-nowrap. The ellipsis half only draws for text in the block's own inline formatting context. When the renderer returns a block-level flex row instead of text, an overflowing flex item is simply cut — no ellipsis, and nothing in the accessible name. objectstack-ai/objectstack#5066 measured exactly that (79px clip box, 32px text node, 25px overflow, 33% reading as 3), so the "clips silently" half of the mechanism is empirical, not inferred.

objectstack-ai/objectstack#5066 fixes the percent renderer from the renderer side: the number gets shrink priority and the decorative bar shrinks away. It deliberately did not take that issue's third suggested route (make the chip itself tolerant of renderers that are not single-line text, or measure before clipping), so the chip-side mechanism is untouched and still applies to every other multi-element renderer.

Candidates, and how far the analysis actually goes

Class-level reading of packages/fields/src/index.tsx only — no browser measurement, because the container this ran in has no Chromium binary. So treat the list as leads, not verdicts:

GeolocationCellRenderer — an inline-flex row: a 14px MapPin plus text like 31.2304, 121.4737 (~110px). In a chip squeezed toward min-w-[7rem] (~72px of content) the tail goes. Because the text sits inside the atomic inline-flex box rather than the chip's own inline context, no partial ellipsis is expected.
ColorSwatchCellRenderer — same shape, 14px swatch plus a 7-char hex. Marginal (~64px), probably fits; listed for completeness.
The percent renderer's own floor after Console: percent display renderer's fixed-width progress bar clips the value in a record-highlights chip (33% renders as 3) objectstack#5066: below roughly 40px of content the bar has collapsed to nothing and the number starts clipping like any other single-line cell. That floor is intrinsic to a self-contained renderer fix — the chip's width is a function of how many chips the record has, so the strip's @container cannot see it.

Renderers whose text carries its own truncate / min-w-0 (lookup, user, file, select-dot — the objectui#3466 family) are NOT affected: the ellipsis draws inside the label.

If someone picks this up

Two shapes, both objectui-only:

Chip-side: stop putting a bare truncate around arbitrary renderer output — e.g. keep the clip but let the renderer opt into its own overflow handling, so a multi-element row is responsible for the ellipsis it can actually draw.
Renderer-side, per renderer: give the value text min-w-0 truncate so it ellipsizes in place (the feat(storage): surface sys_file id on upload-complete — ADR-0104 D3 wave 2 (PR-1) objectstack#3466 convention), which is the cheap local version.

Worth deciding once at the chip, rather than re-litigating per renderer as each one turns up in a highlights strip.

Verification note for whoever takes it: a DOM-only probe reports this surface healthy — the DOM always carries the full value. It needs getBoundingClientRect() plus a screenshot, the way objectstack-ai/objectstack#5066 was measured.

Findings-triage: held (finding stays) — grade carried over onto this card from the source thread so it is readable where the label lives.

The disposition recorded on objectstack#6950 (objectui shard PM, 2026-08-09 09:51Z) stands unchanged: the worst instance (percent renderer) is fixed via #3927; the remaining candidates (Geolocation / ColorSwatch chips, the ~40px floor) are class-level reading with no browser measurement. Restart condition: a measured, user-visible clipped instance (needs getBoundingClientRect() + screenshot, the way objectstack#5066 was measured — a DOM-only probe reports this surface healthy).