Finding: Research data lifecycle is single-use despite cumulative design
The April 3 research run produced 350+ lines of analysis written to /tmp/ (ephemeral). Nothing was archived to /oracle-vault/. The knowledge graph (Graphiti) is unreachable. The calibration system has zero entries. Result: every research starts from absolute zero — no cross-research intelligence, no institutional memory, no accuracy tracking. The system operates as a stateless one-shot researcher despite being architected for cumulative intelligence.