◇ Introduction
Frequently asked questions
Most-asked questions about scope, accuracy, and what Kendall does and doesn't do.
How is this different from running a real load test?
A load test measures what your current production system does under a synthetic workload — useful, but expensive to set up, slow to run, and risky against shared environments. Kendall simulates a designed system against an analytical queueing model, returns results in seconds, and lets you ask hypothetical questions (“what if I add a Redis cache here? what if the database's p99 service time doubles?”). It complements load tests, it doesn't replace them.
How accurate is the simulation?
Kendall's engine is checked against worked examples from Harchol-Balter, Performance Modeling and Design of Computer Systems (2013). Closed-form formulas (Little's Law, Erlang C, Pollaczek–Khinchine) match to numerical precision. Discrete- event simulations of M/M/c and M/G/c networks fall within ~3% of the analytical mean wait time over a 5-minute simulated window. Where the model is known to diverge from reality (retries with backoff, batched service, correlated arrivals), the math reference says so.
Does Kendall know about my actual production data?
No. Designs you build, simulate, and store are entirely synthetic — you pick the topology, the traffic rate, the service-time distributions. Nothing is read from your real infrastructure or APM. The LLM advice layer also never sees your raw production data; it only sees the synthetic simulation results.
Can I import an existing architecture diagram?
Not yet. Today you build designs by hand on the canvas. Import from Terraform / Pulumi / CloudFormation plans is on the roadmap.
Can I export the design as Terraform / Pulumi?
Not yet. Export to IaC is on the roadmap. The catalog already conforms to a normalised ServiceDefinition contract, which is the prerequisite for generating provider-specific output.
Why is AWS Aurora in the catalog but not its read replicas?
Catalog entries are top-level service types. Replication, cluster topology, instance sizing, and other per-service configuration will land in the inspector's capacity tab — designed but not yet shipped. Today you can place an Aurora node and label it; the simulator side is what consumes the deeper config when the engine is live.
How is my data stored?
Designs live in Postgres as a single JSON blob per design (nodes + edges + annotations + containers). Sessions are managed by NextAuth with the Prisma database-session strategy. We never store passwords; magic-link verification tokens are single-use and expire after 10 minutes.