fix(web): keep the contrast slider from flattening text hierarchy

[rynfar]

The last open finding from the #68 review, now measured in a browser rather than reasoned about.

The problem

#7906's contrast boost mixes every foreground role toward one shared target. Per CSS Color 5 an omitted color-mix() percentage is 100% - other, so at a full 100% boost the inner mix gets 0% and each token resolves to the target itself.

Measured against the real Pylon light-theme tokens, reading back getComputedStyle().color:

contrast	--contrast-foreground L	--contrast-muted-foreground L	separation
100 (default)	0.274	0.552	0.278
160	0.110	0.221	0.111
180	0.055	0.110	0.055
200 (max)	0	0	0

At 200, --contrast-foreground, --contrast-muted-foreground, --contrast-placeholder and --contrast-icon-muted all come back oklab(0 0 0). Body text, muted timestamps, icon labels and placeholder text render identically — so an empty composer is indistinguishable from a filled one, and the whole secondary hierarchy disappears at the setting's own maximum. It degrades well before the maximum, too.

The fix

Scale the foreground mix by 0.6, so the maximum lands at a 60% boost instead of 100%:

	fg L	muted L	separation
default	0.274	0.552	0.278
200 as shipped	0	0	0
200 proposed	0.110	0.221	0.111

The slider keeps its full 50–200 range and its direction; the top is still clearly darker than the default; the roles stay apart. The ratio between roles was already preserved at every level — it's the absolute separation that collapses — so capping the mix is enough.

Borders keep the full curve. They carry a single role, so converging on the target costs no hierarchy, and their existing quarter-weight already bounds them at 25%.

0.6 is a judgment call. 0.5 gives separation 0.139 if you want more headroom; the constant is named and documented at the one place it's applied.

Notes

• Arrived with #7906 in chore(upstream): adopt the 2026-08-23 T3 Code batch #68. Not a Pylon-authored defect, and upstream has the same behavior.
• Verification was a standalone CSS harness driving the real token definitions, because the point in question is how the browser resolves nested color-mix() — not app wiring.
• vp test run apps/web/src/appearanceContrast.test.ts — 4 passed; the max-boost test now also asserts the boost is not 100%.
• @t3tools/web typecheck clean; vp check clean on both files.

---

Claude Opus 5 in Pylon.

---

^{Need help on this PR? Tag @codesmith with what you need. Autofix is disabled.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.4 KiB	13.4 KiB	−13 B (−0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.5 KiB	−15 B (−0.2%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.0 KiB	55.0 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	13.4 KiB	13.4 KiB	−4 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−5 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.6 KiB	+1 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	55.8 KiB	55.8 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: ea55ccb · PR result: f9d1de8 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.0 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.