Blame for data/calibration.json

{
 "rel": {
  "R1": [
   29,
   6
  ],
  "R2": [
   54,
   11
  ]
 },
 "ans": {
  "R1": [
   34,
   58
  ],
  "R2": [
   64,
   28
  ]
 },
 "kap": {
  "1": [
   2,
   30
  ],
  "2": [
   15,
   24
  ],
  "3": [
   14,
   11
  ]
 },
 "pg": {
  "1": 0.3333333333333333,
  "2": 0.40860215053763443,
  "3": 0.25806451612903225
 },
 "eta": 0.5909090909090909,
 "source": "Calibrated on 90 held-out HotpotQA/2Wiki/MuSiQue questions with Claude Haiku as the LLM (BRAG benchmark)."
}