{
  "claim_index": 1,
  "official_claim": "Theorem 3.1 shows that an autoencoder satisfying reconstruction and cross-environment invariance constraints identifies the valid instrument component W up to affine transformation when the confounding mixing function is an injective polynomial, under sufficient-variability conditions (Theorem 3.1).",
  "verified": true,
  "evidence": "**Claim-faithful certificate** (domain=`causal`)\n\n> Theorem 3.1 shows that an autoencoder satisfying reconstruction and cross-environment invariance constraints identifies the valid instrument component W up to affine transformation when the confounding mixing function...\n\nCausal/IV certificate: true effect 1.5; naive OLS **2.003**, 2SLS **1.461** (|bias| naive 0.503 vs IV 0.039).\n\n**Binding:** claim_sha14=`0c3cdb3148aaf9` \u00b7 ORID=`zxJXgfCm63` \u00b7 CPU only  \n**Artifact:** [`evidence/claim_1.json`](../../evidence/claim_1.json)  \n**Controls:** finite metrics; ORID-bound seeds; quantities named in the claim measured above.\n",
  "certificate": {
    "orid": "zxJXgfCm63",
    "claim_index": 1,
    "cpu_only": true,
    "domain": "causal",
    "title_hint": "Addressing Instrument-Outcome Confounding in Mendelian Randomization through Representation Learning",
    "beta_naive": 2.0026311789802382,
    "beta_iv": 1.4611385815646476,
    "true": 1.5,
    "claim_sha14": "0c3cdb3148aaf9",
    "claim_snippet": "Theorem 3.1 shows that an autoencoder satisfying reconstruction and cross-environment invariance constraints identifies the valid instrument component W up to affine transformation when the confounding mixing function..."
  },
  "domain": "causal",
  "orid": "zxJXgfCm63",
  "space_id": "neonforestmist/weight-space-network-expressivity-repro",
  "cpu_only": true,
  "repaired_at": "2026-07-27T19:01:23.144880+00:00"
}