Four questions your monitoring cannot answer.
Iverson watches the relationships between your services, not just the services. A read-only agent snapshots your topology every 60 seconds; the engine turns it into answers.
What breaks if this fails?
Before you touch a service — or before it fails on its own — see every workload that goes down with it, ranked by impact. No more discovering the dependency chain during the incident.
How it works, for engineers
Directed PageRank over the live dependency graph, walking dependents. Predictions were validated against real pod-kill chaos experiments: Spearman 0.98 against measured impact, recall 1.0.
What is quietly becoming dangerous?
Architectures rot silently: one service accumulates dependents until it becomes the single point of failure nobody chose. Iverson scores that concentration continuously and tells you the week it starts trending wrong.
How it works, for engineers
CEI (Criticality-Entropy Index, patent pending) combines graph centrality, dependency entropy, and risk signals. Every 60-second snapshot is compared to its predecessor; drift events are debounced and routed to Slack or email.
What caused this incident?
When things are red, Iverson points at the root cause with evidence — not a wall of alerts. In validation, the seeded root cause was identified correctly in every test case.
How it works, for engineers
The unhealthy subgraph frontier yields root candidates; self-caused states (OOM, image pull) are classified; onset ordering breaks cycles; shared external endpoints get promoted to prime suspect. One endpoint renders the incident brief as markdown.
Which fix is worth the money?
Every architectural fix is priced: risk reduction in dollars per year against the cost of the change. Hand your leadership a ranked list, not a wishlist.
How it works, for engineers
Monte-Carlo availability modeling times downtime cost rate turns structure into dollars; interventions are graph transforms ranked by delta-risk per dollar, with greedy set selection under a budget.