AnomaLog: Curated Artefacts for Reproducible Log Anomaly Detection Benchmark Evaluation

This record contains the curated evaluation artefacts for AnomaLog, a framework for reconstructing, executing, and auditing log anomaly detection and security-event analysis benchmarks under explicit experimental assumptions.

The archive contains retained run outputs corresponding to the dissertation evaluation. Artefacts are organised by claim role: paper-facing reproductions, best-effort reconstructions, benchmark extensions, diagnostic baselines, and sensitivity analyses. The retained experiments cover DeepLog and DeepCASE case-study reconstructions as well as benchmark-extension and diagnostic runs across datasets including HDFS, BGL, OpenStack, Thunderbird, and AIT-ADS.

Each retained run directory includes run-local provenance and evaluation artefacts, including files such as `environment.json`, `dataset_manifest.json`, and `metrics.json`. This upload is not an exhaustive dump of every run produced during development. Exploratory, legacy, partial, or non-retained runs discussed in intermediate audit notes are not included. The package should therefore be interpreted as a curated evidence bundle for the dissertation’s reported evaluation, rather than as a complete experiment cache.

The corresponding AnomaLog software/code archive is available separately at https://zenodo.org/records/20541303.