System Design: Input Capture, Ordering, and "What Did the User Input?"

Context

Design a client-side input SDK that runs inside a desktop application process (Windows/macOS/Linux). It must capture keyboard and mouse input from multiple devices, reconstruct what the user typed (including IME/composition), and expose APIs for state queries, replay, and persistence.

Assume the app has access to OS event streams (e.g., raw input/Quartz/evdev) and the text input pipeline (key events, text/IME composition events). The system should add minimal latency and be thread-safe.

Requirements

1. Input coverage
   • Keyboard: key press/release, modifiers, auto-repeat, dead keys.
   • Text input: IME/composition (start/update/end), committed text, deletions.
   • Mouse: button press/release, movement, scroll (wheel/touch), cursor position.
   • Multi-device: multiple keyboards/mice connected simultaneously.
2. Ordering and interpretation
   • Preserve total ordering across devices (global timeline) with stable sequencing.
   • Debounce noisy hardware for keys/buttons.
   • Detect double-clicks with configurable time/position thresholds.
   • Reconstruct the typed text reliably across locales and IMEs.
3. APIs
   • Query current state: which keys/modifiers are down, pointer state/position, active composition, last-click metadata.
   • Query derived text: "what did the user type" over a time/window or session.
   • Replay inputs deterministically.
   • Persist sessions and reload.
4. Engineering
   • Specify data models (event schemas), classes/interfaces.
   • Event queues, buffering strategy, backpressure.
   • Concurrency and thread-safety model.
   • Latency targets and non-functional goals.
   • Failure handling and recovery.

Provide a detailed design covering the above, with clear assumptions and trade-offs.