cache-engineeringlisted

# Cache Engineering Use this skill for cache decisions and failures across request-local memoization, process caches, Redis or other distributed caches, HTTP caches, and CDNs. Preserve the application's source of truth, authorization model, availability target, and chosen infrastructure unless the user asks to change them. A cache is a bounded copy, not a second authority. Every accepted design must define value semantics, complete key scope, freshness, fill ownership, invalidation, degraded behavior, observability, and a condition that would disprove its value. ## Establish scope and evidence Identify the user-visible operation, authoritative data source, cache layers, principals and tenants, mutation paths, consistency limit, request deadline, origin capacity, target environment, and representative workload. Inspect existing code, configuration, metrics, and deployment topology before proposing a mechanism. Measure a comparable baseline: end-to-end latency distribution, throughput and errors; origin request rate, concurrency, latency and limiting resource; and repeated work the proposed cache can actually remove. Do not subtract unrelated percentile values or call a percentile a maximum. If evidence is unavailable, return a bounded measurement plan rather than claiming an improvement. For the full value test, read [references/value.md](references/value.md). ## Define the cached-value contract Record: - authority and revision or generation; - positive, negative, and