{
  "phase_id": "phase-02",
  "category_id": "architecture_engineering_core",
  "doc_id": "cas",
  "source_path": "/home/yeff/public_html/devon/panel/data/cas.md",
  "documentation_branch_status": "CUSTOM_BRANCH",
  "bucket_statuses_from_docs_index": {
    "Prerequisites": "ACTIVE",
    "Installation": "ACTIVE",
    "Configuration": "ACTIVE",
    "Validation": "ACTIVE",
    "Observable Evidence": "ACTIVE",
    "Failure Modes & Recovery": "ACTIVE",
    "Completion & Promotion": "ACTIVE"
  },
  "exported_at": "2026-04-28T05:56:48.644291+00:00",
  "export_rule": "docs/json export generated from real panel/data source plus docs/index.php branch evidence; no authored content invented by export patch",
  "source_payload": {
    "source_format": "markdown",
    "raw_markdown": "# CAS - COGNITIVE ARCHITECTURE SPEC\nversion: v1.1\nstatus: ACTIVE\nparent: master_architecture_index.md\nphase: 02\nphase_name: Architecture & Engineering Canon\ndocument_role: cognitive_architecture_authority\n\n## 1. OBJECTIVE\nDefine the canonical cognitive architecture of Devon.\n\nThis file owns:\n- architecture placement of layers\n- module responsibility at architecture level\n- separation between probabilistic and deterministic components\n\nThis file does not own:\n- detailed flow behavior\n- latency budget\n- security policy\n- operational flow sequence\n\nThose belong to:\n- cfc.md\n- lpc.md\n- sec.md\n- ofc.md\n\n## 2. CORE PRINCIPLE\nDevon is not an LLM-centric system.\nDevon is a governed cognitive system where the LLM is one probabilistic component under deterministic control.\n\n## 3. CANONICAL LAYERS\n\n### 3.1 Interpreter Layer\n- LLM\n- extraction\n- classification\n- semantic parsing\n\n### 3.2 Deterministic Control Layer\n- FSM\n- policy engine\n- transition enforcement\n\n### 3.3 Validation Layer\n- validator\n- contract checker\n- policy enforcement\n\n### 3.4 Execution Layer\n- executor\n- dispatch\n- bounded side effects\n\n### 3.5 Memory Layer\n- working memory\n- canonical memory\n- audit-linked memory routing\n\n### 3.6 Retrieval Layer\n- embeddings\n- retrieval\n- recall support\n\nRule:\nretrieval is a support layer.\nretrieval is not sovereign decision authority.\n\n## 4. SOVEREIGNTY BOUNDARY\nArchitecture ownership in this file means:\n- where a capability belongs\n- what responsibility it has\n- what it must not absorb from another layer\n\nThis file may not redefine:\n- exact state transitions\n- hot path latency limits\n- security controls\n- end-to-end operational sequencing\n\n## 5. LLM BOUNDARY\nThe LLM may:\n- interpret\n- extract\n- structure\n- propose\n\nThe LLM may not:\n- own final policy\n- override FSM\n- bypass validator\n- execute protected action directly\n\n## 6. RETRIEVAL BOUNDARY\nRetrieval belongs to architecture as a support capability.\n\nRetrieval may:\n- provide relevant context\n- support reasoning inputs\n- enrich decision inputs\n\nRetrieval may not:\n- define sovereign policy\n- replace state control\n- mutate canonical truth by itself\n\n## 7. ARCHITECTURE OUTPUT CONTRACT\nAny architecture derived from Devon must preserve:\n- deterministic control above probabilistic interpretation\n- validator before protected execution\n- retrieval as support, never sovereignty\n- explicit auditability\n\n"
  }
}