fix(#838): buttons in the house themes never lifted or pressed

[TortoiseWolfe]

The bug

Buttons on all three ScriptHammer themes never lifted on hover and never pressed on active. Both state rules were out-specified by the resting rule and had never applied — all three states resolved to --sh-plate.

Specificity, not a typo

resting  chain                                    (0,7,0)
hover    .btn:hover                               (0,3,0)
active   chain-minus-two + :active                (0,6,0)

:not() contributes the specificity of its argument, so the five-:not() chain on the resting rule scores 7 and both neighbours lose to it. The chain was added by #379 so this block would stop beating the sh-* utilities — and it silently disabled the two rules underneath it.

Each state now repeats the whole chain, giving a strict ladder 7 < 8 < 9. Verbose on purpose: doing it by counting rather than by source order means re-ordering the file cannot swap the states back.

The literal black that survived nine tickets

The hover rule read rgb(0 0 0 / 0.3) — invisible on a dark surface, grey sludge on a saturated one, and precisely what the ink tokens exist to replace. It survived the entire #376/#426 epic because it was never rendered: no screenshot can show a state that doesn't apply, and no reviewer can see one.

New --sh-plate-lift is built from the same --sh-ink-* tokens as --sh-plate, just raised further.

The gate couldn't see it either

depth-tokens.spec.ts read resting only — every test in it. Added T5: for each house theme, hover and press a real .btn, assert the three states are three different things, and that none uses literal black. It asserts difference rather than exact values, so a legitimate design tweak doesn't break it while the ladder collapsing does.

Mutation-verified: sinking the hover selector back to .btn:hover fails with scripthammer-light: hover must differ from resting.

Two things my first draft got wrong

It read the shadow once, straight after .hover(). DaisyUI transitions box-shadow, so that catches the animation mid-flight — measured 7.05px 14.11px -5.26px, partway between --sh-plate (6/12/-5) and --sh-plate-lift (10/20/-6). A one-shot read is timing-dependent and could land equal to resting and fail for nothing. It polls now — three consecutive runs: 7 passed, 7 passed, 7 passed.

That same truncated read briefly made it look like the fix hadn't worked, because the intermediate value shares its leading offsets with the resting shadow. I confirmed the rule actually wins by enumerating the matching CSSOM rules — exactly one .btn:hover rule carries a box-shadow, and it's ours, resolving to var(--sh-plate-lift).

Verification

Measured after the fix on all three house themes: resting != hover != active, on every one. Before, all three were byte-identical.

lint clean, 4730 vitest tests, 428 script tests, tsc --noEmit clean.

This is a visible change to every button on three themes. The gate proves the states now differ; it can't tell you the lift looks right. Worth an eyeball on the preview before you take my word for the aesthetics.

Closes #838 — and this is the named exit condition of epic #426.

🤖 Generated with Claude Code