Declarative Recovery-Correctness Analysis
for Distributed Microservices
ChaosLab verifies whether distributed applications recover to a
correct business state after controlled failures —
not just whether the services come back up.
After a chaos experiment, a distributed system can report every service as up, every health check green, and every pod running — while the underlying business state is quietly wrong. Liveness checks don't catch this. ChaosLab does.
A retried request after a network partition charges the customer twice, even though every service reports success.
An order is confirmed, but a crash during recovery leaves stock levels untouched — silently overselling later.
The order service and payment service disagree about what happened, and nothing surfaces the contradiction.
ChaosLab sits between a chaos-engineering tool and the application under test, correlating what actually happened against a declarative definition of what should have happened.
Fault injection is supporting infrastructure. These are the parts that make recovery correctness verifiable.
Business-level invariants — "an order's payment and inventory state must always agree" — expressed declaratively, not hard-coded per test.
Logs, database snapshots, and OpenTelemetry traces are correlated across services to reconstruct exactly what happened during recovery.
Checks outcomes against real business rules, not just HTTP status codes or pod readiness.
Reproducible failure scenarios — network partitions, pod kills, latency injection — driven through Chaos Mesh.
Deliberately broken recovery paths validate that ChaosLab's oracles actually detect incorrect state, not just healthy ones.
A structured PASS/FAIL verdict backed by the evidence trail that produced it — auditable, not a black box.
A birds-eye view of the ChaosLab system. See the full breakdown on the Architecture page.
The recovery-correctness engine (Oracle Engine, Evidence Collector, Correlation Engine, Verdict Generator) is under active development and not yet implemented. This website currently reflects the approved project plan and design — not a working analysis engine.