Replication package: Kubernetes resource requests and limits in public Helm charts (dataset and code)

Replication package for an empirical study of Kubernetes CPU and memory requests and limits in public Helm charts: the mined dataset (17,076 Artifact Hub packages, 10,537 charts, 23,028 declared container resource blocks), normalization and RQ1-RQ3 analysis, helm template rendered-manifest validation, a signal-aware linter with a precision/recall accuracy evaluation against established analyzers, and a controlled runtime experiment (1,088 runs across three workload classes - a synthetic service, Redis, and PostgreSQL - on a local cluster and five Amazon EKS instance types) establishing that the resource-specification signals map to observable CPU throttling, out-of-memory termination, and eviction outcomes. Code under MIT, data under CC-BY-4.0; the underlying charts are not redistributed. Version 3.0.0 adds the multi-workload runtime campaign, the linter accuracy evaluation, and the recast ecosystem-prevalence analysis.