reports/eval/comparison.json

{
  "base": {
    "vuln_detection": "acc=0.68 f1=0.579",
    "mcq": "acc=0.62",
    "humaneval": "pass@1=0.939"
  },
  "merged": {
    "vuln_detection": "acc=0.64 f1=0.625",
    "mcq": "acc=0.72",
    "humaneval": "pass@1=0.8171"
  }
}