skills/competitive-research/references/semantic-intelligence-layer.md
Semantic Intelligence Layer
Date: 2026-06-28 Status: Implemented as the first deterministic v1 slice.
Purpose
The CI engine must not publish raw page-change events as intelligence. Scout and direct fetch produce acquisition snapshots; ci_core.py now converts those snapshots into semantic facts, compares facts against the previous snapshot, and publishes only material semantic deltas.
First Intelligence Lanes
-
Customer Proof Intelligence - Sources: customer pages, case studies, customer story pages. - Initial competitors: Constructor, Bloomreach, Coveo. - Publishable deltas: new named customer proof, new outcome, new metric, or new AI/search/product-discovery proof. - Default owner: Sales Enablement.
-
Content/Narrative Intelligence - Sources: blogs, press/news pages, RSS/content pages, AI/search pages. - Publishable deltas: new AI search, agentic search, product discovery, proof-backed narrative, or campaign-worthy claim. - Default owner: Product Marketing.
Ledger Tables
semantic_facts stores extracted source-specific facts with source URL, evidence text, evidence URL, confidence, and typed JSON.
semantic_deltas stores before/after comparison results with materiality score, Algolia implication, action owner, recommended action, evidence URLs, and quality_status.
The legacy signals table remains for compatibility, but direct source collection now inserts signals only from semantic_deltas where quality_status = publish.
Publish Gates
The engine suppresses:
- first-run baseline captures
- collector-method changes
- hash-only movement
- cookie, nav, footer, and boilerplate changes
- changes that do not map to a supported semantic schema
- deltas below materiality threshold
Suppressed deltas stay in semantic_deltas for diagnostics and dashboard trust, but they do not become executive findings.
Report Contract
Daily reports should answer:
- What changed
- Why it matters to Algolia
- Recommended action
- Evidence
- Validation needed
Weekly reports should answer:
- Customer proof movement
- Content/narrative movement
- Campaign opportunities
- Recommended actions by owner
- Battlecard updates
- Suppressed weak signals
- Coverage gaps
Dashboard Contract
Dashboard v2 exposes four intelligence products:
- Customer Proof Radar
- Narrative And Content Radar
- Decision Queue
- Suppressed Signals
Dashboard export still reads archived Markdown briefs, so semantic lane quality depends on daily and weekly reports being generated from semantic deltas.