Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Finite browser lowering for portable behavior effects

Document status: verified. Canonical source.

Summary

A browser adapter can lower the first NUIF behavior profile without accepting authored JavaScript. The narrow mapping is enabled native-button activation, the HTML hidden property for visibility, and an ARIA status live region for advisory announcements. The state machine remains validated data interpreted by one fixed generated runtime. An exact CSP hash grants that runtime—and no other inline script—authority inside the self-contained output.

Native HTML invoker commands are not a replacement for this profile. Their built-in vocabulary covers popover and dialog actions. A custom command still requires script and does not carry NUIF state, ordered guards, typed variables or abstract effects. The smallest faithful result is therefore a finite interpreter, not a general code generator or a misuse of popover state.

Evidence

  • The HTML Living Standard defines hidden for all HTML elements; the hidden state is not rendered. This is a direct Boolean host operation rather than a CSS-cascade approximation. Locator: HTML §6.1, lines 92–96 of the retrieved multipage interaction document, https://html.spec.whatwg.org/multipage/interaction.html#the-hidden-attribute, retrieved 2026-08-31.
  • The HTML button commandfor/command surface has built-ins for showing, hiding and toggling popovers and closing or showing dialogs. The standard’s custom-command example installs a JavaScript command event listener. This supports using native commands when their component semantics are exact, but not claiming that they encode a general guarded state machine. Locator: HTML button element, commandfor and examples, retrieved lines 98–175, https://html.spec.whatwg.org/dev/form-elements.html#the-button-element, retrieved 2026-08-31.
  • WAI-ARIA 1.2 defines status as advisory live-region information, says it should not receive focus as a result of status change, and gives it implicit aria-live=polite and aria-atomic=true. This is the correct first target for non-urgent NUIF announcement effects. Locator: WAI-ARIA 1.2 status role, lines 4471–4480, https://www.w3.org/TR/wai-aria-1.2/#status, retrieved 2026-08-31.
  • CSP Level 3 permits policy delivery through an early meta element, while preferring the response header for served resources. Hash sources use SHA-256/384/512 plus base64, and a matching hash can allow an inline script without unsafe-inline. The inline body is hashed after UTF-8 encoding. Locators: CSP3 §§2.3.1, 3.1, 3.3, 4.2 and 8.4, https://www.w3.org/TR/CSP3/, retrieved 2026-08-31.
  • Playwright documents locators as its auto-waiting/retry unit and recommends user-facing roles or explicit test contracts over DOM-structure chains. It exposes ARIA snapshots computed from browser accessibility trees. Stable data-nuif-id is the adapter’s explicit test contract; snapshots provide a foreign host observation rather than a source-string assertion. Locators: https://playwright.dev/docs/locators#introduction and https://playwright.dev/docs/aria-snapshots#aria-snapshots, retrieved 2026-08-31.
  • The HTML activation model says non-click manual activation, including keyboard or voice input, fires a click event at an element with activation behavior. Playwright’s locator.press() focuses the element and produces a key sequence, so separate Enter/Space runs test the native route without adding key handlers to the adapter. Locators: https://html.spec.whatwg.org/multipage/interaction.html#activation-behavior-of-elements and https://playwright.dev/docs/input#keys-and-shortcuts, retrieved 2026-08-31.

Mechanism

nuif-web-behavior-0 composes the behavior and accessibility profiles. It rejects checkbox/radio transition sources because a native click also mutates checked state that the source behavior vocabulary cannot currently express. It rejects disabled transition sources because native disabled controls do not produce the promised activation. All enabled button/switch entities receive a listener so a valid activation with no transition remains an observable no-op. It also rejects a visibility effect that can hide one of those controls through the control itself or an ancestor; otherwise later source events could exist in the abstract trace but not in the native host.

The program JSON is embedded only after <, >, &, U+2028 and U+2029 are escaped. The static runtime contains no evaluation, dynamic import, request, timer or handler-attribute surface. Its meta CSP denies all resource classes and allows the one exact script body. A serving host should deliver and merge its own response-header policy; the self-contained policy is not assumed to survive embedding in another document.

One transition may emit at most one non-empty announcement, and may not repeat the same effect/target pair. A browser can otherwise coalesce multiple run-to-completion effects into one terminal DOM observation, producing a false equivalence claim.

NUIF relevance

The Rust stage validates unsafe embedding strings and every web-specific refusal, computes the exact script/CSP digests and produces the reference run. The Playwright stage performs separate five-event pointer and alternating Enter/Space keyboard sequences in each pinned engine and checks state, selected transition, retained hidden values, live-region text, target attribution, runtime errors and ARIA snapshots after every event. The first macOS/arm64 run passes Chromium 151.0.7922.34, Firefox 153.0 and WebKit 26.5.

This proves one browser-host mapping. It does not prove screen-reader speech, focus order, native UI behavior, checkbox/radio semantics, authored-script round trips, broader events/effects or a wire-format decision.

Open questions

  • Add a portable checked/pressed-state effect before admitting checkbox/radio activation, then test keyboard and pointer traces separately.
  • Run assistive-technology announcement timing experiments before claiming more than DOM and browser accessibility-tree exposure.
  • Define focus effects and focus-restoration laws before adding dialog or popover lowering.
  • Compare a second host family against the same abstract effects before moving the behavior sidecar into a canonical schema proposal.