{"slug":"implementation/adrs","title":"28 · ADRs","section":"implementation","sectionLabel":"Implementation","status":"unclassified","type":"document","path":"implementation/adrs.mdx","bytes":2835,"description":"Canonical architecture decisions — AI Workbench supersedes Twin Workbench, three continuities, Spine Agents, Discovery/Activation boundary, MCP routing, and what each decision supersedes.","updated":null,"source":"integratewise/Integratewise-docs-wiki","content":"\n## Decision Record\n\nEach ADR states **what was decided, why, what it supersedes, and implementation impact**.\n\n| ADR | Decision | Supersedes |\n|---|---|---|\n| **ADR-001** | **AI Workbench supersedes Twin Workbench.** Twin is the human-facing AI coworker *inside* the AI Workbench; the Workbench is the operating environment. | Twin Workbench as a separate workbench |\n| **ADR-002** | **Three continuity scopes.** Organizational Memory (THE memory, in the Spine), Personal Contextual Memory, AI Continuity Memory. Latter two serve the first. | Flat \"Spine Fabric memory\" view |\n| **ADR-003** | **Personal Contextual Memory boundary.** Human working continuity, private unless deliberately shared/promoted. Not silent organizational truth. | Coworker / shared-everything continuity |\n| **ADR-004** | **AI Continuity persistence — semantic defined, substrate decision open.** See G-10 in [Codebase Alignment](/implementation/codebase-alignment). | Unscoped AI memory store |\n| **ADR-005** | **Spine Agents as governed organizational actors.** Identity, role, authority, continuity, capability access, evidence. | Agent-as-tool-wrapper |\n| **ADR-006** | **Discovery vs Activation separation.** Discovery = what exists; Activation = what becomes active for tenant/runtime/context. | Single discovery-as-activation |\n| **ADR-007** | **MCP Pool routing boundary.** MCP is governed and routed via Pool to the Capability Fabric → Governance → Hermes path; not an open public MCP. | Unrestricted MCP |\n| **ADR-008** | **No monolithic AI Orchestrator.** Cognition, model routing, agent runtime, capability resolution, authority, and execution are separated. | Central orchestrator decides authority |\n| **ADR-009** | **Verified closure.** Provider success is not completion; close on verify → reconcile → governed Spine update → rehydration. | Close-on-200 |\n\n\nADRs never redefine the [Founder Canon](/canonical/00-SPINEWORKSPACE-INTEGRATEWISE-CANON). They record how the Canon is interpreted into contracts and code.\n\n\n## Where to Go Next\n\n<CardGroup cols={2}>\n  \n    How each ADR traces to drift G-01 → G-10.\n  \n  \n    Authority-ranked register with explicit supersession.\n  \n  \n    What changed, why, and implementation impact.\n  \n</CardGroup>\n"}