Summary
In extropy network, strict topology gating can fail on modularity while the persisted network_metrics.metrics_json.modularity for the same network id is above the configured minimum.
This creates contradictory outcomes:
- run is quarantined/rejected due to modularity fail,
- but saved metrics for that artifact show modularity in-range.
Repro
- Run network with strict gate (default via balanced profile):
uv run extropy network -s <scenario> --validate --seed 42 --quality-profile balanced
- Inspect latest
network_runs.meta_json.quality.final_metrics.modularity and quality.bounds.modularity_min.
- Inspect latest
network_metrics.metrics_json.modularity for same network_id.
Observed (example)
network_runs.meta_json.quality.final_metrics.modularity = 0.406678...network_runs.meta_json.quality.bounds.modularity_min = 0.45- strict gate rejects (
accepted=false)
- same artifact in
network_metrics.metrics_json.modularity = 0.470425...
Expected
Modularity used for strict gate decision and persisted network metrics should match (or differences must be explicitly documented and both values persisted with labels indicating method/stage).
Why this matters
This makes network quality triage unreliable and repeatedly surfaces as an apparent modularity blocker even when reported metrics suggest pass.
Suspected area
extropy/population/network/generator.py gate path vs generate_network_with_metrics() + extropy/population/network/metrics.py modularity computation path/method consistency.
Summary
In
extropy network, strict topology gating can fail on modularity while the persistednetwork_metrics.metrics_json.modularityfor the same network id is above the configured minimum.This creates contradictory outcomes:
Repro
uv run extropy network -s <scenario> --validate --seed 42 --quality-profile balancednetwork_runs.meta_json.quality.final_metrics.modularityandquality.bounds.modularity_min.network_metrics.metrics_json.modularityfor samenetwork_id.Observed (example)
network_runs.meta_json.quality.final_metrics.modularity = 0.406678...network_runs.meta_json.quality.bounds.modularity_min = 0.45accepted=false)network_metrics.metrics_json.modularity = 0.470425...Expected
Modularity used for strict gate decision and persisted network metrics should match (or differences must be explicitly documented and both values persisted with labels indicating method/stage).
Why this matters
This makes network quality triage unreliable and repeatedly surfaces as an apparent modularity blocker even when reported metrics suggest pass.
Suspected area
extropy/population/network/generator.pygate path vsgenerate_network_with_metrics()+extropy/population/network/metrics.pymodularity computation path/method consistency.