A fact row leads with the fact

[WaylandYang]

Closes #500.

A fact row was one line carrying the direction arrow, the predicate, the confidence and stale and contested marks, the corrected ⟲, the interval, the object name, the sources link and the interval pencil. Grouped under From X / To X the line is 249 px wide. The trailing group was shrink-0, the hover-revealed actions hold their width whether or not anyone is hovering (u-reveal only changes opacity), and the predicate was the only item that could give — with no floor. So it gave everything.

On one entity in the benchmark base, 144 rows: 35 predicates rendered at zero width. → 2023-03-02 ~ now Project Aurora says a fact exists and refuses to say which one.

The row becomes two lines

• Line 1 is the fact: the direction arrow, the predicate, and the object on the right. Read across, it is the sentence.
• Line 2 is what we know about it: the interval, the confidence, stale and contested marks, the corrected ⟲, and on the right the sources link and the interval pencil.

The predicate takes min-w-0 flex-1 on line 1, so it gets the leftover rather than absorbing the deficit. The object keeps its max-w-40 cap and truncates first, which is the right order — the object is a node you can click through to, the predicate is the fact.

The two hover actions move down with the interval for the same reason: invisible is not the same as absent, and they were holding width on every row.

Every predicate now carries its full text in a title, so the one that still truncates is recoverable. An inferred predicate's title carries the name and the hint together instead of replacing the name with the hint.

Measured, same entity, same 144 rows

	before	after
predicate readable	62	143
predicate truncated	47	1
predicate invisible	35	0

continued separate lawsuits is the one that still truncates, by 13 px, with its full text on hover.

Also checked

The ontology page's panel was rebuilt in the same PR (#480) and is a different, milder case: its rows are property → range classes, there is no interval competing for the line, both sides truncate at a deliberate cap and nothing reaches zero width. Left alone.

tsc and the style guard are clean. Verified in the browser against a copy of the benchmark base.

🤖 Generated with Claude Code