capacity-planninglisted

# Capacity Planning ## Purpose Produce a decision that states **which workload and failure scenarios a concrete configuration can support, with what evidence and uncertainty**. Capacity is not one QPS number: it depends on request mix, payload distribution, cache state, dependency behavior, resource limits, placement, software version, admission policy and the SLO. The authoritative result is a measured feasible envelope. Analytical models interpolate, forecast and expose sensitivities; they do not manufacture tail latency from throughput. In particular, do not add p99 values, treat a whole fleet as M/M/1, or multiply a single-replica latency limit by a USL efficiency coefficient. Quantiles are not additive, and throughput scaling does not determine a latency distribution. This skill has no fixed Java API baseline or executable Java examples. Its authoring contract is version-neutral measurement; the target project's build toolchain, deployed JDK/vendor, flags and dependency versions determine which runtime behaviors apply. Inspect those before recommending JVM or container changes. Applying this skill does not authorize a Java/framework upgrade, preview features or new infrastructure dependencies. ## Required output: capacity decision record Record: - decision owner, date, review date and affected service/version; - demand unit: admitted requests, messages, bytes, tenant operations, or another useful-work measure; distinguish offered, admitted, attempted and success