haruspex-bench / README.md
TyrusRC's picture
Haruspex in-distribution benchmark splits (web, mobile, cloud, network, GUIDE triage)
2159550 verified
|
Raw History Blame Contribute Delete
889 Bytes
metadata
license: other
task_categories:
  - text-classification
tags:
  - security
  - benchmark

Haruspex security-decision benchmark (in-distribution splits)

Held-out test splits for haruspex. Each .jsonl row is {type, state, question, criteria/candidates/levels, target}. These are in-distribution splits of the training data; independent results are reported in the model card against public datasets (e.g. deepset/prompt-injections, the OWASP Benchmark).

  • data/security_test.jsonl — web/offensive security decisions (own synthetic)
  • data/mobile_test.jsonl — mobile app security (own synthetic)
  • data/cloud_test.jsonl — cloud posture (own synthetic)
  • data/network_test.jsonl — network security (own synthetic)
  • data/guide_test.jsonl — Microsoft GUIDE incident triage (CDLA-Permissive-2.0 (Microsoft GUIDE))