Skip to main content

​
March 2nd, 2026

Resilient ‘Temporal Brain’ Scheduling & Interactive Interaction Today was a major architectural milestone. We successfully shifted our task execution system from a brittle, linked-list “Relay Race” pattern to a centralized, idempotent “Temporal Brain” architecture. This overhaul, combined with the new “Triple-Write” protocol, ensures that our task scheduling remains rock-solid across cloud, local cache, and hardware alarms.

​
1. ‘Temporal Brain’ Scheduling Architecture

We refactored the core task execution system to be significantly more resilient and self-healing:
  • Centralized Scheduler: Implemented syncTaskSchedule(taskId) as the single source of truth for all verification alarms.
  • Self-Healing Mutations: Updated Task Create/Update mutations to trigger automatic schedule resets, ensuring all edits (time shifts or recurrence changes) are immediately reflected in pending alarms.
  • Decoupled Verification: Removed hardcoded next_instance_id chaining. The system now dynamically calculates the next point in time after every check-in, allowing for stateless and derived scheduling directly from database records.
  • Watchdog Sync: Added a background Cron job that “heals” any orphaned tasks missing active alarms every hour.

​
2. Interactive Calendar & ‘Drag-to-Edit’

We transformed the Schedules screen into a highly interactive, performance-oriented interface:
  • High-Performance Dragging: Enabled dragging in @howljs/calendar-kit with 15-minute increments to maintain grid alignment.
  • Validation Lifecycle: Introduced a dedicated confirmation layer (dragConfirm state) that captures gestures and presents proposed temporal snapshots before committing changes.
  • Persistent Selection Mode: Implemented friction-based ‘Long Press’ to activate event selection handles, which persist until a background tap or commit occurs.

​
3. ‘Triple-Write’ Synchronization Protocol

To ensure hardware-level alignment, we established an atomic synchronization lifecycle:
  • Atomic Operations: Every event update now follows a strict path: Authoritative Cloud mutation (Convex) -> Local instance synchronization (SQLite) -> Hardware trigger refresh (Native Alarms).
  • Zero-Stale Data: Implemented the “Search-Clobber-Spawn” pattern in our local-db repositories to ensure local storage always mirrors the cloud state.

​
4. Backend Modularization & Localization

  • Rule Decoupling: Extracted standalone verification rules for ‘Stay Throughout’ and ‘Just Show Up’ tasks, separating business logic from persistence.
  • Centralized Formatters: Unified all date/time formatting in convex/lib/formatters.ts to ensure consistent IST (Asia/Kolkata) localization across the entire system.
  • Conflict Feedback: Enhanced the UI confirmation modal to provide descriptive conflict feedback, including overlapping event titles and time slots.

​
5. Dev Environment & Tooling

  • ADB Version Sync: Resolved “Uncaught TypeError” and hangs by prioritizing modern Android SDK platform-tools in .bashrc.
  • Connection Diagnostics: Overhauled connect-phone.ps1 with smarter device sharing (usbipd) and diagnostic coloring for reliable physical device debugging.