Skip to content
Help
HelpTrust & provenance

Agent graph & decision trees

How a visitor is routed — classify the network, score confidence, pick a personalization tier, then the agent graph. The router is deterministic; the LLM only drafts; the Gate disposes. · 4 min read

Agent graph shows the live routing made visible: a visitor’s identity key → classification → confidence → personalization tier → the agent graph that drafts and verifies the page.

Two identity keys

Email (first-party) is say — they told you, so you may recite the company. IP only (involuntary) is allude — shape the message, never recite. Both feed the same graph.

Classify → confidence → tier

  • Network type comes from real ip-api flags (corporate · consumer-ISP · mobile · VPN/proxy · hosting), with a corporate-VPN exception.
  • Per-field confidence (location / company / industry) — each inference hop loses confidence.
  • Tier 0 neutral · 1 location-aware · 2 firmographic · 3 firmographic + competitive.

The agent graph

Deterministic spineClassify → Score → Route → (Cache) → Design → Render. Pure, reproducible, provable.
LLM leavesCompetitor research (Tier 3) and the copy agent — they only propose.
The GateDisposes — verifies every line vs source + surface policy; a blocked line never ships.
Open /graph for the full exhibit; it’s grounded in pipeline/personalization/scene.py + creative.py.
Related