cid 966625262f122283a0b4f01f58297d53bcb1f41a7e9264d84c77f9cd0fcad234 · original · open access · via openalex
Title: AI Bill of Materials CycloneDX Evaluation Framework - Results Authors: Wiebe Vandendriessche; Merlijn Sebrechts; Bruno Volckaert Venue: Zenodo (CERN European Organization for Nuclear Research) Year: 2026 Canonical: doi:10.5281/zenodo.23602099 URL: https://doi.org/10.5281/zenodo.23602099 Collector: openalex OA: true ABSTRACT AIBOM-CDX-Eval: results of run eval-003 Raw outputs, scores and aggregated results of run **eval-003** of AIBOM-CDX-Eval, a benchmark of open-source AI Bill of Materials (AIBOM) generators. It measures performance (T1), reliability (T2), model discovery in source code (T3) and CycloneDX quality (T4). - Inputs: the 100 most-downloaded Hugging Face models (10 repeats per tool) and 16 synthetic source-code targets.- Tools: aibomgen-cli, OWASP aibom-generator, ALOHA, lab700xorg-aisbom, cdxgen, HTS AiBOM, akumar-aibom, jorisbot-aibom, trusera-aibom. Each is pinned to a commit; versions are in `run_meta.json`.- Run: 6,096 jobs on 2026-10-20, framework commit `9fc720a`. Files - `eval-003-runs.zip`: the raw output, logs and `scores.json` of every job.- `eval-003-aggregated.zip`: the tables (CSV/JSON) and figures.- `eval-003_report.md`, `eval-003_summary.json`: the report and the numbers behind it.- `aibom-cdx-eval-9fc720a.zip`: the framework code at the commit used (includes the README and methodology).- `SHA256SUMS`: checksums. Reproduce Put the unzipped runs at `results/runs/eval-003/` in the code, then run `uv run eval score --run-id eval-003` and `uv run eval aggregate --run-id eval-003`. To run the tools again, use https://github.com/idlab-discover/aibom-cdx-eval. Licence: data CC-BY-4.0, code AGPL-3.0.