# Data Model Research & Simplification Decision (2025-12-23) ## Problem Tinker system had 5 data models with overlapping purposes, unclear usage patterns, and unused tables. ## Research Findings (via Rails console query) **Actual Usage:** - Artifacts: 0 records (unused) - CodeDiffs: 0 records (unused) - Comments: 6 records (actively used) - AgentMemory: 1 record (underutilized) - AgentLogs: 36 records (working well) **Key Insight:** Agents prefer Comments for progress tracking, not Artifacts. Code changes happen via git/PR, not CodeDiff. ## Decision Made **REMOVE:** - Artifact model - redundant with Comments - CodeDiff model - GitHub PR handles code review **KEEP:** - Comment (4 types: note, decision, question, code_review) - AgentMemory (enhance with new types) - AgentLog (audit trail) ## Implementation Ticket #58 created: "Simplify data model and encourage knowledge accumulation" Changes: 1. Delete Artifact/CodeDiff models, controllers, routes, specs 2. Remove `checkpoint_task` and `create_code_diff` MCP tools 3. Add `summary` type to AgentMemory 4. Update MCP tool descriptions to encourage memory storage ## Behavioral Pattern to Encourage Agents should store memories when: - Solving bugs → error memory with solution - Making decisions → decision memory - Completing research → summary memory (new type) - Learning facts → fact memory This builds organizational knowledge over time via semantic search (RAG).