Research Insights Made Simple #25 · July 30, 2026Livestream · #SRE
Modeling Reliability from a Dependency Graph
Alexander Polomodov × Anatoly Krasnovsky
/ Research Insights Made Simple #25 · Graph reliability
- Alexander Polomodov — host
- Anatoly Krasnovsky — guest, engineer and distributed-systems reliability researcher
Slide contents
1. Modeling Reliability from a Dependency Graph
Alexander Polomodov × Anatoly Krasnovsky
2. The graph narrows the chaos search
3. Source version changes the conclusion
4. A hand-built model ages before the system
5. Telemetry becomes an executable model
6. The graph sees only the mandatory path
7. Version 1 fixes the failure count
8. Model and system measure one outcome
9. DeathStarBench is complex but controllable
10. At p=0.3 the model matched the system
11. Five rates preserve the overall trend
12. Bias changes across failure rates
13. Replicas help inside visible structure
14. Version 2 changes the failure model
15. Shared fate lies outside the call graph
16. Uncertainty forms the chaos backlog
17. Research becomes a platform pipeline
18. A pilot starts with one operation
19. Build a model before chaos engineering
Paper · ACM Digital Library
Code and data · GitHub
Author · Telegram
The graph selects hypotheses. Chaos checks reality.