[joeblau]

Closes #37

What

When a target is tapped, requestFeedback now returns before it creates or triggers a WebHaptics engine if all of these hold:

• typeof navigator.vibrate !== 'function', checked at tap time rather than through the module-load WebHaptics.isSupported.
• webHapticsOptions.debug isn't set.
• webHapticsOptions.showSwitch isn't set.

On desktop browsers without vibration, the engine's hidden <label> + <input switch> and its rAF pattern loop produced no feedback. The vibration path and the iOS native path are unchanged. Debug audio and the visible switch work exactly as before.

Tests:

• The shared beforeEach now stubs navigator.vibrate. The existing engine tests ran without a stub, so they now exercise the vibration path they assert on.
• New tests in an unsupported desktop suite:
  • With no options, or with debug: false / showSwitch: false, no engine is created, no trigger fires and no body nodes are added, and the host action still fires once.
  • With debug: true and with showSwitch: true, the engine is still constructed and triggered.
  • Vibration support is re-checked on every tap.

Implemented with the Codex CLI on the shared box and verified by hand.

Benchmark (npm run bench, same box)

The desktop-without-vibration path wasn't touched by #42, #43 or #44, so the quieter 78ec2f2 runs from earlier today are a valid baseline. "Interleaved" means back-to-back runs on main at 1ea44a1 while another agent's build was saturating the box.

Desktop without vibration	Quiet baseline (3 runs)	Interleaved main (2 runs)	This PR (2 runs)
Engines after 100 distinct taps	100	100	0
Body nodes / engine labels / rAF per tap after 100 taps	300 / 100 / 2	300 / 100 / 2	0 / 0 / 0
First-tap cost (instrumented ms)	1.56 / 1.75 / 1.85	5.81 / 10.19	0.48 / 0.38 (≥ −69%)
Remaining 99 distinct taps (ms/tap)	0.38 / 0.38 / 0.38	0.49 / 0.79	0.050 / 0.041
Warmed desktop click (median ms/tap)	0.075 / 0.075 / 0.069	0.124 / 0.159	0.074 / 0.033

The vibration and iOS rows show no change beyond noise; vibration engines stay at 100 and iOS engines at 0. Full ESM goes from 4,583 to 4,607 gzip bytes (+24 B).

Verification

• Locally: npm run lint, typecheck (ESM+CJS+DTS build), coverage (146 tests, 100% statements/branches/functions/lines on React 19 and React 18.3.1), size, and pack:check all pass.
• SSR: only the click handler changed. navigator is guaranteed there, and nothing reads it at import or during server render.
• Risk from Skip WebHaptics engine, hidden DOM and rAF loop on desktop browsers without vibration #37: an iOS-like WebKit that isIOSWebKitWithoutVibration() misses would lose the hidden-switch fallback. Detection already covers iPhone/iPad/iPod and iPadOS desktop mode (MacIntel + touch points). In-app webviews still need physical confirmation in Certify iOS and Android behavior on physical devices #9.

Device-testing notes

These are simulated-browser numbers from Node 24 with production React and jsdom, using a stubbed Linux navigator with no vibrate. None were measured on physical devices, and no physical device was used. Physical iPhone/Android/iPadOS/VoiceOver certification is still tracked in #9.