/goal Destination: Observational Memory has a merged, hardened, fully iterated OM Cluster 0.6.x implementation covering PR #33 and issues #34-#41, plus the additional hardening and roadmap in om_cluster_unified_roadmap_implementation_plan.md. First read that document, PR #33, and issues #34-#41. Work milestone by milestone: merge/prepare PR #33 and harden 0.6.0 preview; implement strict ID/path validation, safe head updates, observation-only reflection catch-up, and honest key-rotation docs; implement #34 interactive approval; preserve #39 and implement #40 Windows cluster hardening; implement #38 feature-flag caching; implement #41 reflection metadata, stale-state pruning, host-memory coexistence, and cluster-aware semantic merge; add namespace/source policy, override semantics, and record indexing; implement #37 as key epochs plus historical rewrap/purge semantics; implement #35 relay and #36 optional P2P only after trust/key foundations are safe. Preserve disabled-mode behavior, local-first operation, generated Markdown as materialized views, untrusted transports, no secret/plaintext leakage, and base install without heavyweight optional transport dependencies. After each milestone update docs/om-cluster-roadmap-status.md, add focused tests, run pytest and ruff, and continue unless a change would delete user data, weaken security, require a large required dependency, or incompatibly change public CLI semantics. Success means full tests/lint pass, docs match behavior, and multi-node temp-install validation proves convergence, approval, redaction, key epoch behavior, Windows coverage, and transport secrecy.
