LAFC-Evict: Learning-Augmented Cache Eviction Dataset

LAFC-Evict v0.2 is a public real-data preview dataset for learning-augmented cache eviction research. It contains derived candidate-row features and finite-horizon counterfactual supervision labels for cache-eviction decisions. This v0.2 preview contains 4,800,000 rows across two Parquet configurations: cross_family_evict_value_v1 and objective_ablation_scalar. It includes only Wikimedia pageview-derived rows from the wiki2018 family. It does not include raw trace rows, raw Wikimedia page titles, model files, experiment logs, or machine-local filesystem paths. Object identifiers are deterministic public pseudonyms. Hugging Face remains the interactive ML-ready distribution endpoint. This Zenodo record is intended as a frozen archival copy with DOI/versioning support. Wiki2018 provenance: Wikimedia pageview data is made available by the Wikimedia Foundation at https://dumps.wikimedia.org/other/pageviews/ under the Creative Commons CC0 1.0 public domain dedication. This LAFC-Evict preview does not redistribute raw pageview dump rows or raw page titles. Wikimedia and the Wikimedia Foundation do not endorse this derived dataset.