{"slug":"triage/06-CLASSIFICATION","title":"06 — Classification","section":"triage","sectionLabel":"Triage","status":"canon","type":"document","path":"triage/06-CLASSIFICATION.mdx","bytes":3019,"description":"How material is classified by type, domain, sensitivity, and confidence before review.","updated":"2026-08-24","source":"integratewise/Integratewise-docs-wiki","content":"\n\n## How Triage Classifies\n\nTriage evaluates every incoming piece of evidence across multiple dimensions:\n\n### Dimension 1: Source and Provenance\n\n- Where did this come from? (MCP, conversation, agent, manual)\n- Who produced it? (user, AI, system)\n- Is the source trusted? (known connector vs unknown)\n\n### Dimension 2: Semantic Category\n\n- What type of knowledge is this?\n- Categories: insight, fact, decision, action_item, preference, relationship\n- Domain: executive, product, engineering, design, sales, marketing, etc.\n\n### Dimension 3: Entity Linking\n\n- Which known entities does this relate to? (accounts, contacts, deals, etc.)\n- Is this a new entity or an update to an existing one?\n- Confidence of the link\n\n### Dimension 4: Duplication and Conflict\n\n- Does this duplicate existing canonical memory?\n- Does it conflict with established truth?\n- Is it stale (old information presented as current)?\n\n### Dimension 5: Sensitivity and Materiality\n\n- How sensitive is this information?\n- Is it material (business-significant) or trivial?\n- Does it require governance review?\n\n### Dimension 6: Memory Scope\n\n- Should this be personal, conversational, agent, or organizational scope?\n- Does it meet the threshold for organizational memory?\n\n## Classification Output\n\nEvery triage evaluation produces:\n\n```json\n{\n  \"classification\": {\n    \"doc_type\": \"lead|document|policy|code|... \",\n    \"workflow_type\": \"sales|marketing|engineering|... \",\n    \"intent\": \"description of intent\"\n  },\n  \"linking\": {\n    \"primary_entity\": \"entity_id\",\n    \"related_entities\": [\"entity_id_1\", \"entity_id_2\"]\n  },\n  \"canonical_review\": {\n    \"is_duplicate\": false,\n    \"has_conflict\": false,\n    \"conflict_reason\": null\n  },\n  \"memory\": {\n    \"scope\": \"organizational\",\n    \"candidate\": true\n  },\n  \"governance\": {\n    \"decision\": \"stage\",\n    \"reason\": \"High-value decision with clear entity links\"\n  }\n}\n```\n\n## Routing Decisions\n\n| Decision | Meaning | Action |\n|---|---|---|\n| `stage` | Candidate for review | Write to triage_results as pending |\n| `reject` | Below threshold or spam | Terminal, no memory |\n| `defer` | Needs more context | Return to review queue |\n| `repair` | Incomplete, fixable | Route for repair then re-triage |\n\n## Confidence Dimensions\n\nConfidence is six things, and one number must never represent all six:\n\n1. **Pattern confidence** — How well does this match known patterns?\n2. **Prediction confidence** — How likely is this to be accurate?\n3. **Input completeness** — How complete is the source data?\n4. **Data freshness** — How current is the information?\n5. **Historical precision** — How accurate has this source been?\n6. **Policy eligibility** — Does this meet policy requirements?\n\nSmall samples use a named conservative lower bound, not raw accuracy. 3/3 is not better evidence than 470/500.\n"}