{
  "alt": "Dot chart with five rows. Raw contacting sits just right of zero and recalibrated contacting clearly right of it. Raw routing sits left of zero and recalibrated routing right of it with a range that crosses zero. Accepting an agent's finish sits right of zero with a range that reaches it.",
  "caption": "Right of the line, Jev's policy does better than the simple alternative; left of it, worse. Compare each decision's raw and recalibrated rows. Recalibrating needs labeled records, which these figures do not charge for.",
  "data": [
    {
      "conf_high": 3.2,
      "conf_low": -0.9,
      "decision": "Contacting reviewers, raw probabilities (against a classifier)",
      "gap": 1.2
    },
    {
      "conf_high": 8.2,
      "conf_low": 4.5,
      "decision": "Contacting reviewers, Jev recalibrated (against a classifier)",
      "gap": 6.4
    },
    {
      "conf_high": -2.6,
      "conf_low": -17.8,
      "decision": "Routing complaints, raw probabilities (against a classifier)",
      "gap": -10
    },
    {
      "conf_high": 11.4,
      "conf_low": -0.7,
      "decision": "Routing complaints, both recalibrated (against a classifier)",
      "gap": 5.3
    },
    {
      "conf_high": 34.6,
      "conf_low": 0,
      "decision": "Accepting an agent's finish (against a person checking all)",
      "gap": 17.5
    }
  ],
  "id": "policy-gaps",
  "source": "Case chapters of this booklet: coffee reviews and CFPB complaints, weighted; SWE-rebench patches, weighted; constructed patches and agent tasks, unweighted; engine calls October 2026; costs assumed",
  "spec": {
    "hi": "conf_high",
    "lo": "conf_low",
    "reference": 0,
    "referenceLabel": "No gap",
    "sort": "data",
    "x": "gap",
    "xDomain": [
      -30,
      50
    ],
    "xFormat": {
      "digits": 0,
      "signed": true
    },
    "xLabel": "Jev's gap against the simple alternative (%)",
    "y": "decision"
  },
  "subtitle": "Jev's policy against the simple alternative, percent of that alternative's value or cost, with 95% intervals",
  "title": "Whether Jev's policy beats the simple alternative turns on calibrating its probabilities",
  "type": "dot"
}