enhance (cost + performance) adding weights to graphs derived from confidence score to use weighted bfs

[akhiljns]

Weighted Graph Traversal and Pathfinding

Changes

build.py

Every edge now gets a cost attribute derived from confidence_score:

cost = 1.0 / confidence_score

Examples:

• EXTRACTED = 1.0
• INFERRED@0.8 = 1.25
• AMBIGUOUS@0.2 = 5.0

serve.py

Two new algorithms were added:

_weighted_bfs()

A priority-queue-based BFS using heapq that explores high-confidence edges first within the depth limit.

_weighted_shortest_path()

Uses nx.dijkstra_path() with the cost attribute.

MCP tools

• query_graph now accepts mode: "weighted" alongside "bfs" and "dfs".
• shortest_path now accepts weighted: true for Dijkstra-based traversal.

cluster.py

cluster() and _partition() now accept weighted=True.

When enabled:

• confidence_score is copied into the weight edge attribute for Leiden/Louvain clustering.
• EXTRACTED edges (1.0) bind communities tighter than INFERRED edges (0.6–0.9) or AMBIGUOUS edges (0.1–0.3).

main.py

New CLI flags:

• graphify query "..." --weighted — priority-queue BFS.
• graphify path "A" "B" --weighted — Dijkstra shortest path.

Help text has also been updated for both commands.

How It Works

graphify query "authentication" --weighted

Explores high-confidence edges first.

graphify path "UserService" "Database" --weighted

Finds the most trustworthy path, not necessarily the shortest.

The weighted path avoids low-confidence shortcuts. For example, an AMBIGUOUS 1-hop edge with cost 5.0 is skipped in favor of a 3-hop EXTRACTED path with total cost 3.0.