A judged winner looks like one (ksdrd)
The three moments the product is actually about — the judge's pick, a DiffJudge apply-and-typecheck
pass, a consensus verdict — used to render in the same tone as any successful thing on screen: a
cost figure, a flash marker, a completed step. The run detail's verdict block now paints the
winning agent id, and the winning output's label, with the palette's verified token.
Restraint is deliberate and enforced. That token is the brightest in the dark palette by a wide
margin, so on more than a couple of elements per screen it would dominate everything including the
accent. npm run selftest:tui-verdict-token asserts it has exactly ONE consuming file and at most
two uses inside it, and that no surface paints a winner with the generic success or done tone.
A verdict with NO winner is no longer painted like a success either — "the judge picked nobody" was rendering in the same green as "the judge picked someone", which was the same conflation in the other direction.
The light variant does not reuse sulfur: it measures 1.38:1 on white, so that variant carries its own darker verdict green, and the smoke asserts the substitution is still in place.
Code fences use four tones, not eight (kptd7)
A rendered fence carried eight distinct colours. One was the raw ANSI name magenta, which bypassed the palette entirely and so rendered the same in every variant. One was the brand accent doing...
The fuzzy picker highlights what it matched (kenqx)
Typing in any fuzzy picker now paints the characters that matched, the way fzf, helix and telescope do. The scorer already knew them — it tracks the position of every hit to compute its...