Skip to main content

​
Trust-Aware Pathfinding

Introd should optimize for the highest probability of a successful introduction, not the fewest hops. The legacy GitLab docs described BFS/shortest-path style traversal. That is not the production principle. GitHub now preserves the safer direction: trust-aware path scoring, provenance checks, weak-link diagnostics, and tenant-scoped graph queries.

​
Ranking Principle

A path is only useful if the connector is likely to help and the route can be explained. The scorer favors:
  • evidence-backed relationship edges
  • higher connector trust
  • relationship freshness
  • recent interaction history
  • intro-history evidence
  • stronger first-hop connector quality
  • fewer weak links
  • clearer explainability
The scorer penalizes:
  • inferred edges
  • stale relationships
  • low-confidence evidence
  • little interaction history
  • excessive hops
  • unknown provenance

​
Why Shortest Path Is Wrong

A weak two-hop path can be worse than a trusted three-hop path. Example:
  • Weak 2-hop route: imported LinkedIn edge plus unknown connector relationship.
  • Strong 3-hop route: prior intro history, recent email interaction, and a manually confirmed connector.
The second path should rank higher because it is more likely to produce a real introduction.

​
Backend Implementation

Primary scoring file:
  • server/services/graph/pathIntelligence.ts
Key outputs:
  • path_score: route quality score.
  • path_confidence: confidence in the complete path.
  • weakest_link_score: lowest quality relationship in the path.
  • weak_link_penalty: penalty applied when one relationship is weak.
  • freshness_score: relationship freshness across the route.
  • avg_node_trust: average trust quality of people in the route.
  • explainability: evidence lines explaining the route.
  • weak_link_diagnostics: warnings explaining weak confidence.
Graph query service:
  • graph-intel/app/pathfinder.py
Graph-intel must keep queries tenant-scoped with org_id, must filter out unsafe inferred edges, and must rank by relationship confidence before hop count.

​
Result Semantics

High confidence means:
  • Introd has verified or observed evidence for the route.
  • The route has no severe weak link.
  • The connector relationship is fresh or historically reliable.
Low confidence means:
  • The route may be relevant, but the graph cannot prove trust strongly enough.
  • The user should nurture, verify, or add context before asking for an intro.
No result means:
  • The target is not represented in the user’s scoped graph.
  • Or no path has enough evidence to recommend responsibly.

​
Production Rule

Never hide weak evidence behind confident product language. The product should say:
This route is possible, but confidence is limited because the connector relationship is stale.
Not:
Best warm intro path.