{
  "url": "https://www.moonstonehq.com/articles/painter-or-pokemon",
  "slug": "painter-or-pokemon",
  "title": "Painter or Pokemon: the Artist Predicts 14% of a Card's Rank, the Creature Predicts 40%",
  "summary": "Knowing only who painted a card explains 12% of where it lands in its own set, out of sample, against a placebo of minus 2%. On the 12,671 test rows the species model also sees, the artist figure is 14% and the species figure is 40%: the creature wins by roughly three to one on identical rows. And the concentration of chase work in a few hands has collapsed from 81.7% to 15.5% across eight eras.",
  "section": "Data forensics",
  "published": "2026-09-04",
  "readMinutes": 9,
  "publisher": "moonstone",
  "findings": [
    {
      "value": "0.14",
      "measures": "artist out-of-sample R2, on the shared test rows",
      "basis": null
    },
    {
      "value": "0.40",
      "measures": "species R2, same rows",
      "basis": "the creature wins by roughly 3x"
    },
    {
      "value": "0.79",
      "measures": "artist split-half reliability",
      "basis": "placebo minus 0.04"
    },
    {
      "value": "18.5%",
      "measures": "of attributed top-decile cards belong to one production house",
      "basis": null
    }
  ],
  "tables": [
    {
      "title": "Where an artist's cards land inside their own sets",
      "columns": [
        "Artist",
        "Priced cards",
        "Sets",
        "Mean within-set percentile",
        "Share in their set's top decile"
      ],
      "rows": [
        {
          "Artist": "Hikaru Koike",
          "Priced cards": "46",
          "Sets": "15",
          "Mean within-set percentile": "0.85",
          "Share in their set's top decile": "56.5%"
        },
        {
          "Artist": "Kimiya Masago",
          "Priced cards": "25",
          "Sets": "5",
          "Mean within-set percentile": "0.83",
          "Share in their set's top decile": "44.0%"
        },
        {
          "Artist": "Shizurow",
          "Priced cards": "81",
          "Sets": "21",
          "Mean within-set percentile": "0.76",
          "Share in their set's top decile": "43.2%"
        },
        {
          "Artist": "aky CG Works",
          "Priced cards": "184",
          "Sets": "42",
          "Mean within-set percentile": "0.75",
          "Share in their set's top decile": "25.5%"
        },
        {
          "Artist": "Ryo Ueda",
          "Priced cards": "442",
          "Sets": "84",
          "Mean within-set percentile": "0.60",
          "Share in their set's top decile": "31.7%"
        },
        {
          "Artist": "Mitsuhiro Arita",
          "Priced cards": "709",
          "Sets": "123",
          "Mean within-set percentile": "0.57",
          "Share in their set's top decile": "15.8%"
        },
        {
          "Artist": "Ken Sugimori",
          "Priced cards": "1,059",
          "Sets": "90",
          "Mean within-set percentile": "0.43",
          "Share in their set's top decile": "7.1%"
        },
        {
          "Artist": "Miki Tanaka",
          "Priced cards": "196",
          "Sets": "78",
          "Mean within-set percentile": "0.32",
          "Share in their set's top decile": "1.9%"
        }
      ]
    }
  ],
  "charts": [],
  "sections": [
    "The design that keeps this honest",
    "The ladder, and the man at the bottom of it",
    "The collapse nobody announced"
  ],
  "method": [
    "Population: 21,071 priced English cards in the 171 sets holding at least 20 priced cards; 19,582 carry an illustrator, across 395 artists. Prices resolve through the shared canonical printing rule. The dependent variable is a card's price percentile within its own set. Sets are split odd and even into disjoint halves; an artist's score is estimated on the training half only and evaluated on cards in the held-out half, so the two are computed from disjoint samples in disjoint sets and cannot be mechanically linked. A label shuffled placebo scores minus 0.02 at card level and minus 0.04 at artist level. Artist-level reliability is a split-half correlation across the 141 artists with at least ten cards in each half. The species comparison is run on the identical 12,671 test rows. Top decile membership is computed within each set. Illustrator attribution comes from the catalog's illustrator field, which is backfilled from TCGdex and is missing on roughly 7% of priced English cards; those cards are excluded rather than imputed."
  ],
  "terms": "Figures are moonstone's own recorded measurements, dated as published. Quote them with the population stated beside them. The article at `url` is the human form of this file."
}