Tier 1 palette collision: graphite-700 and graphite-800 are both #232634, so dark-mode borders measure 1.00:1

[mpstaton]

Context:

• Two-Palette-Steps-Share-One-Hex-So-Dark-Mode-Borders-Vanish.md

Four federal control-boundary findings from a corpora-curator probe; three confirmed, one rejected as stated (it measured --color-border-strong against --color-surface when secondary draws on --color-surface-raised — same error made earlier in the session).

1. CONFIRMED — --color__graphite-700 and --color__graphite-800 are the identical hex. Dark-mode --color-border on --color-surface = 1.00:1. Light 1.41, vibrant 1.45 — both inside gate A22's intended band; dark is supposed to be and isn't.
2. REJECTED as stated / real narrower bug — secondary passes in all three modes (3.53 / 3.06 / 3.38). But outline is background: transparent, so its border sits on the parent: 3.44 on background, 3.23 on surface-2, 2.73 on --color-surface ❌. Re-measure after fixing Engineering Review of All Specifications #1; graphite-700 moves.
3. CONFIRMED — destructive sets a fill and no border-color, inheriting transparent. Fill-vs-surround measures 1.02–1.32 in both modes against a 3:1 floor. Text inside is 7.19 / 4.89, which is why it survived nine migrations — every gate we run is a text contrast gate.
4. CONFIRMED — ghost's hover is --color-selected-tint; on a host already painted with it, hover-vs-rest composites to 1.122:1.

Deliberately queued behind wave 4. Four migration agents are in flight with baseline gate numbers captured; editing theme.css now would invalidate their contrast measurements invisibly.

Underneath all four: F7 states a 3:1 control-boundary floor in prose and nothing measures it. The contrast gate passes 30/30 text pairs straight through every one of these.