Skip to main content

​
Edge Provenance

Introd’s moat depends on whether the graph can distinguish a real relationship from a convenient inference. A relationship edge is only useful when the system knows where it came from, how confident it is, and whether it is safe to use as introduction evidence.

​
Runtime Contract

The backend path scorer models provenance through RelationshipEdgeProvenance in server/services/graph/pathIntelligence.ts. 
type RelationshipProvenanceSource =
  | "linkedin_import"
  | "email_interaction"
  | "calendar_overlap"
  | "manual_confirmation"
  | "intro_history"
  | "unknown";

type RelationshipVerificationType =
  | "verified"
  | "observed"
  | "imported"
  | "inferred";
Each edge must resolve to:
  • source: where the evidence came from.
  • confidence: normalized confidence from 0 to 1.
  • evidence: short explainability lines shown to the user or used in diagnostics.
  • inferred: whether the relationship is inferred rather than verified or observed.
  • verificationType: whether this is verified, observed, imported, or inferred.

​
Evidence Priority

Strongest evidence:
  • manual_confirmation: a person explicitly confirmed the relationship.
  • intro_history: an intro was requested, accepted, completed, or otherwise progressed.
  • email_interaction: communication history exists.
  • linkedin_import: first-degree LinkedIn import exists.
Weaker evidence:
  • calendar_overlap: people appeared in the same meeting or event, but the relationship may be weaker than direct communication.
  • unknown: the system lacks enough metadata to verify the edge.

​
Inferred Edges

Inferred edges are allowed as signals, but not as fake trust. Allowed:
  • Show as low-confidence context.
  • Use in weak-link diagnostics.
  • Use to suggest “warm this route first” actions.
Not allowed:
  • Treat inferred edges as equivalent to verified edges.
  • Use same-company or shared-tag inference as proof of relationship.
  • Inflate a route into a high-confidence warm intro path.

​
Scoring Implications

scoreEdgeQuality() combines:
  • relationship strength
  • provenance confidence
  • freshness
  • interaction evidence
  • provenance weight
  • inferred-edge penalty
This means a shorter path with weak evidence should lose to a longer path with stronger trust evidence.

​
Production Rule

If the graph cannot prove a route, Introd should say so clearly. Correct behavior:
We do not have enough verified relationship evidence to recommend a warm intro path yet.
Incorrect behavior:
Here is a warm path because two people worked at the same company.