{
  "id": "cold-start-prior-value",
  "claim": "For cold-start landing pages, brand-mean Beta priors (~160 pseudo-observations) beat uninformative Beta(1,1); text-embedding kNN does not beat brand mean once same-brand neighbours are removed.",
  "definition": "Leave-one-page-out measurement of prior information value across multi-page races.",
  "numbers": [
    {
      "label": "n_pages",
      "value": 83
    },
    {
      "label": "impressions",
      "value": 20341
    },
    {
      "label": "conversions_24h",
      "value": 343
    },
    {
      "label": "brand_mean_tau_star",
      "value": 160
    },
    {
      "label": "brand_mean_delta_ll_vs_beta11",
      "value": 199.6
    },
    {
      "label": "knn_vs_brand_p_no_improvement",
      "value": 0.517
    }
  ],
  "canonical_url": "https://lutiq.com/research/cold-start-prior-value-loo/",
  "methods_url": "https://lutiq.com/research/cold-start-prior-value-loo/",
  "open_data_url": "https://lutiq.com/data/research/cold-start-prior-loo-2026-08.json",
  "sources": [
    "https://lutiq.com/research/cold-start-prior-value-loo/"
  ],
  "last_verified": "2026-08-06",
  "caveats": [
    "Retrospective LOO, not online prior A/B.",
    "Signal is brand identity in this sample, not design transfer via text embeddings."
  ],
  "_meta": {
    "schema": "https://lutiq.com/data/answer-packets/_schema.json",
    "publisher": "Lutiq"
  }
}