traffic_matrix_set:
  example:
    - source_path: region1/.*/rb
      sink_path: region2/.*/rb
      demand: 100000
      mode: combine  # <-- triggers the bug

Root Cause

In ngraph/exec/analysis/flow.py, both build_demand_context() and demand_placement_analysis() reconstruct TrafficDemand objects from serialized config but fail to pass the id field:

demand = TrafficDemand(
    source_path=config["source_path"],
    sink_path=config["sink_path"],
    demand=config["demand"],
    mode=config.get("mode", "pairwise"),
    flow_policy_config=config.get("flow_policy_config"),
    priority=config.get("priority", 0),
    # id=config.get("id"),  <-- MISSING
)

When mode: combine is used, pseudo-nodes are named using TrafficDemand.id (e.g., _src_{td.id}). Without preserving the ID, each reconstruction generates a new random UUID, causing pseudo-node name mismatches
between the cached context and the analysis phase.

Fix

Add id=config.get("id") to TrafficDemand construction in both:

- demand_placement_analysis() (line ~187)
- build_demand_context() (line ~433)

Testing Gap

This bug slipped through because:

1. No tests for mode: combine + context caching - The bug only occurs when both conditions are met
2. No TrafficMatrixPlacement integration test with mode: combine - Workflow tests likely only use mode: pairwise
3. No ID round-trip validation - No test verifies TrafficDemand.id survives serialization/deserialization

Suggested Test Coverage

| Test Type   | Description                                                       |
|-------------|-------------------------------------------------------------------|
| Unit        | TrafficDemand.id preserved through dict serialization round-trip  |
| Integration | demand_placement_analysis() with cached context and mode: combine |
| E2E         | TrafficMatrixPlacement step with mode: combine traffic matrix     |