{
  "case": "consumer",
  "question": "Did weighting or exclusion move the finding?",
  "domain": "research integrity · consumer research",
  "register": "detection",
  "status": "method-demonstration",
  "verdict": "valid",
  "geometry_digest_v3": "f7d31d9e945971bd26076cf9b7c18901e9801cdaa373d45f2751f251a84ecc8b",
  "note": "Non-influence on three real public consumer-research datasets (Ecdat::Yogurt, GSS 2018, NHANES). A declared exclusion — or the survey's own design weights — rings 22–53× the median of the data's own churn and 7–14× its 99th percentile; both quantiles are declared because both are properties of the corpus. An equal-size random placebo moves 21–32× less under either floor, and that separation — not a pass/fail — is the claim. LOCALIZATION IS WITHDRAWN (2026-07-15): this receipt previously reported the named coupling as independently nominated. The exclusion selects rows by the coupling it then measures, so the rank restated the operator's choice, not a property of the field; a decoy coupling self-nominates the same way. No rank is published. Whether a non-circular localization holds is untested and unclaimed. Detection, not adjudication; non-influence is not validity."
}