{
  "alt": "Dot chart of seven contact policies on a dollar scale. Perfect foresight is highest. Luna, the Claude Haiku adapter, Jev and the TF-IDF classifier sit close together just below it. The keyword rule is far lower, and contacting every reviewer is lowest, just above zero.",
  "caption": "The four model-based dots sit a little below the ceiling set by perfect foresight, with Luna and the adapter ahead of Jev and the classifier. The keyword rule and contacting every reviewer earn much less. Each dot combines the measured mix of right and wrong contacts with the assumed costs ({fact:contact_cost} per contact, {fact:expected_recovery} expected recovery). Whiskers are 95% intervals from a bootstrap that resamples products within star ratings; the ceiling and contacting everyone are fixed by the star mix.",
  "data": [
    {
      "conf_high": null,
      "conf_low": null,
      "estimate": 5978,
      "group": "Ceiling",
      "n": 5000,
      "policy": "Perfect foresight"
    },
    {
      "conf_high": 5253,
      "conf_low": 5135,
      "estimate": 5195,
      "group": "Decision engine",
      "n": 5000,
      "policy": "OpenAI Luna"
    },
    {
      "conf_high": 5202,
      "conf_low": 5083,
      "estimate": 5144,
      "group": "General language model",
      "n": 5000,
      "policy": "Claude Haiku adapter"
    },
    {
      "conf_high": 5144,
      "conf_low": 5004,
      "estimate": 5076,
      "group": "Decision engine",
      "n": 5000,
      "policy": "Jev 1.14.0"
    },
    {
      "conf_high": 5104,
      "conf_low": 4928,
      "estimate": 5015,
      "group": "Trained classifier",
      "n": 5000,
      "policy": "TF-IDF + logistic"
    },
    {
      "conf_high": 2309,
      "conf_low": 1978,
      "estimate": 2148,
      "group": "Simple rule",
      "n": 5000,
      "policy": "Keyword rule"
    },
    {
      "conf_high": null,
      "conf_low": null,
      "estimate": 832,
      "group": "Simple rule",
      "n": 5000,
      "policy": "Contact every reviewer"
    }
  ],
  "id": "net-value-by-policy",
  "source": "Amazon Reviews 2023, coffee products; 5,000 sampled reviews, weighted; engine calls October 2026; assumed costs",
  "spec": {
    "colors": {
      "Ceiling": "muted",
      "Decision engine": "cat1",
      "General language model": "cat3",
      "Simple rule": "cat2",
      "Trained classifier": "cat5"
    },
    "group": "group",
    "hi": "conf_high",
    "lo": "conf_low",
    "n": "n",
    "reference": 0,
    "referenceLabel": "Contact nobody",
    "sort": "data",
    "valueLabels": true,
    "x": "estimate",
    "xDomain": [
      0,
      7000
    ],
    "xFormat": {
      "digits": 0,
      "prefix": "$",
      "thousands": true
    },
    "xLabel": "Net value (US dollars per thousand reviews)",
    "y": "policy"
  },
  "subtitle": "Net value of contacts per thousand reviews against contacting nobody, at the cost cutoff p*, by policy, weighted",
  "title": "Model-based contact policies earn far more than a keyword rule or contacting everyone, and Jev's raw probabilities earn no clear premium over the classifier",
  "type": "dot",
  "width": "body"
}