[FedyaLight]

The shared controller and reader still perform work when the visible output has not changed: repeated screen-label writes during state delivery, repeated geometry signatures, Gamepad polling in hidden native WebViews, and a continuous homepage demo loop even under reduced motion or offscreen.

This change keeps frame-rate scrolling and the existing synchronization cadence while removing that redundant work:

• Update secondary-screen metadata only when its content changes; restore the latest state when returning to a visible preview.
• Cache layout signatures and preserve measured geometry for identical/appearance-only settings; continue remeasuring actual layout/font/text/viewport changes.
• Keep profile and display identity in hello/presence and dedicated updates, and omit those redundant fields from high-rate reader state packets.
• Suspend Gamepad polling while unfocused/hidden, including native visibility, preserving neutral arming and held-input release on return.
• Stop the demo animation for reduced motion, an offscreen demo or a hidden page, and resume when appropriate. Avoid background connection-help UI work and updates to hidden HUD elements.

docs/performance-audit.md records the full startup-to-shutdown audit across the browser, macOS, Windows/Linux and Android, with preserved background tasks and measurement limits. Native server/shell optimizations are independent in #4; the homepage star button is independent in #3.

Measured with one controller and three Chromium readers, eight-second samples:

• Unchanged screen-label DOM mutations: 1476 to 0 during scrolling, and 24 to 0 while paused/voice-settled.
• Controller main-thread CPU: 9.23% baseline versus 7.86–8.50% in three changed samples. Timing is workload-dependent, not a guaranteed saving on every device.
• Aggregate bytes sent by the four clients: 499523 baseline versus 449037 in the final measured changed sample, without reducing state cadence.
• Hidden native controller with Gamepad enabled and reduced-motion demo: 36 callbacks to 0 per 600 ms; the offscreen demo also draws no continuous frames.

Validation:

• Extended tests/performance.mjs passes; baseline code fails the intended label-churn and hidden-Gamepad regressions. The reduced-motion baseline likewise continues drawing.
• Remote, browser, three-display and real PeerJS/WebRTC hosting suites pass. Existing preview error remains below one CSS pixel; local mode makes no external requests.
• 2184 responsive checks pass across six languages, two themes, seven viewport sizes and all five platform/hosting adapters.
• The multi-display test explicitly selects its named source after reconnect, avoiding an assumption about arrival order.
• JavaScript syntax and whitespace checks pass.

These measurements use Chromium and production relays. Physical battery, native rendering engines and actual remote hardware are not certified by the browser suite. Keep-awake protection, voice recognition/transcript quality, scripts/backups, display calibration, smooth scrolling, heartbeats and reconnect remain intact.