Foundational research for reliable infrastructure
Classical algorithms, formal methods and reasoning systems for the networks that cannot afford to go down.
Incident intelligence
Causal models that turn alarm floods into incidents with a named origin.
Read papersRemediation
Optimal control for pacing fixes across fleets without causing the next incident.
Read papersFormal methods
Exact, explainable checks that an action is safe against live network state.
Read papersSystems
Architecture, compression and isolation for continuous cross-layer reasoning.
Read papersRecognized research
Papers and experiment notes from Raincurve Research
Turning Alarm Floods into Incidents with Hawkes Processes
Raincurve Research · 2026Experiment note · RemediationHow Fast Should a Network Fix Roll Out? Almgren-Chriss for Operations
Raincurve Research · 2026Experiment note · Formal methodsProving a Network Action Is Safe Before It Runs
Raincurve Research · 2026Paper · SystemsCurve-1: A Reasoning Architecture for Cross-Layer Infrastructure Intelligence
Raincurve Research · 2026Paper · SystemsCompact Contracts: Compressing Telemetry for Cross-Layer Reasoning
Raincurve Research · 2026Paper · SecurityFirecracker Sandboxing for Secure Execution
Raincurve Research · 2026Research blog
View allWhy alarm correlation needs a causal model, not a time window
Rule engines group alarms by closeness in time and space. We look at what that misses when background noise rises, and what a self-exciting model does differently.
What trading desks can teach a NOC about rollout pacing
Optimal execution and network remediation share the same trade-off: move fast and pay in churn, move slowly and pay in exposure.
Runbooks are not proofs: checking actions against live state
Most unsafe maintenance actions look safe in isolation. Here is why the check has to see in-flight work and lost backup paths.
Inside Curve-1: from raw telemetry to ranked hypotheses
A walkthrough of the four stages that turn fragmented operational signals into an explainable root-cause ranking.