{
 "license": "CC-BY-4.0",
 "attribution": "Fide AI, Agentic Cyber Explorer",
 "url": "https://agentic-cyber-explorer.pages.dev/findings/pipeline-choices-move-cyber-scores/",
 "asOf": "2026-09-26",
 "id": "pipeline-choices-move-cyber-scores",
 "claim": "Evaluation pipeline choices alone can move a model's cybersecurity benchmark score by more than 80 points and reorder models.",
 "evidenceKind": "measured",
 "scope": "One preprint audit of public cyber benchmarks.",
 "topics": [
  "eval-validity"
 ],
 "atlas": [],
 "evidence": [
  {
   "event": "benchmark-scores-pipeline-dependent-cyber-2026"
  }
 ],
 "relations": [],
 "statusHistory": [
  {
   "status": "reported",
   "on": "2026-09-08",
   "why": "Audit of pipeline dependence in cyber benchmarks.",
   "event": "benchmark-scores-pipeline-dependent-cyber-2026",
   "kind": "evidence"
  }
 ],
 "halfLifeDays": 365,
 "wouldChange": "Standardized pipelines that make scores comparable across evaluators.",
 "fideQuestions": [
  "FID-075"
 ],
 "methods": [
  "ctf-benchmarks"
 ],
 "review": "assistant-drafted",
 "addedOn": "2026-09-25"
}