task_intelligence_baseline.md

Task intelligence baseline for #9352

This file records the pre-implementation boundaries that the Ticket 01 fixtures and source manifest protect.

Current authority boundary: all authenticated users share the Candidate, task, goal, workstream, recommendation, and Chat-first logic. Persisted workflow/UI fields remain account-generation and proactive-choice fences; memory storage origin and deployment environment variables are never product gates.

• Backend conversation extraction can write directly to action_items; desktop screen extraction stages and ranks locally.
• Desktop task discovery is coupled to proactive-task notification enablement.
• The backend action-item contract drops desktop-only provenance, confidence, goal, recurrence, and agent fields.
• Mobile goal links and desktop task-agent continuity are not canonical backend relationships.
• staged_tasks and local staged SQLite can both influence promotion and deduplication.
• Feedback cannot yet join an intervention to a Candidate, canonical task/workstream, and later outcome.

Ticket ownership:

• Ticket 02 removes backend contract and Candidate-lifecycle gaps.
• Ticket 03 proves Swift/Dart round-trip parity.
• Tickets 04–06 establish workstream product and execution boundaries.
• Tickets 07–12 replace capture, UX, evaluation, and attention behavior.
• Universal convergence retains staged-task reads only as a compatibility adapter. Physical legacy deletion requires separate evidence and approval.

Remaining compatibility burn-down (not an entitlement boundary):

• Delete TaskPromotionService timer and local staged_tasks GRDB authority once released-client parity is proven.
• Grep ratchet proving legacy staged writers are gone.
• Collapse Swift task_chat_messages into kernel SQLite (INV-CHAT-1 dual-store debt).
• Remove tmux / legacy task_chat surface paths marked for Ticket 14.