{
 "_doc": "The lab's claim and its own stated scope for one result, copied from one table row of the lab's list of its strongest results. claim is the row's first sentence with its bold markers (**) removed; limits is the row's bold scope span, word for word, without its markers. No other character is changed. This result has no row in the portfolio's current-claims list. Its claim and limits are copied here word for word from the lab's own file named under source, at the commit and sha256 given there.",
 "id": "interconnect-lab:adversarial-admission-farm",
 "claim": "~1.99 billion adversarial admission inputs → 0 escapes, all 14 conjunct-weakenings broken, native Rust farm bit-for-bit identical to the Python oracle.",
 "limits": "Honest scope: the ~1.99e9 is two SEPARATE exhaustive-to-bound single-gate sweeps SUMMED (tenant 9.77e8 = 125⁴·4 + MemGate 1.01e9 = 150⁴·2), NOT a joint sweep of the composed `fullAdmit = tenant ∧ mem` input space; each is exhaustive-to-its-bound, NOT a ∀-inputs theorem (the Lean tight bounds are the ∀-inputs backstop). The value is the native-Rust ↔ Python bit-identity + all 14 weakenings breaking with a real witness (non-vacuous), not the joint-space cardinality.",
 "source": {
  "lab": "interconnect lab",
  "commit": "888ad0eb026a1b54e5c0b0a1899902f3b35875d6",
  "file": "reports/crown_jewels.md",
  "line": 319,
  "sha256": "c2646ea50efa235ef70f5f55c6d256de43ac2430dd428f41c2ae8f46988bf8a3",
  "bytes": 189622,
  "row": "14",
  "evidence_tag": "🟡 exhaustive-to-bound"
 },
 "result_receipt": "/receipts/inference-limits/extracts/adversarial-admission-farm.result.json"
}