{
  "abstract": "Language models can propose useful actions without being reliable authorities over the systems that execute\n\nthem. We define Artificial Deterministic Intelligence (ADI) as an architectural approach in which probabilistic\n\nmodels provide bounded inference while identity, authority, memory admission, policy, execution rights, state\n\ntransitions, verification, and release are governed by deterministic systems external to the model. Determinism\n\nrefers to specified control decisions under explicit state and inputs; it does not imply deterministic generation,\n\ncorrect policies, truthful memory, or universal safety. We present a layered reference model and separate\n\nconceptual requirements from observations in a governed research evidence ledger. In a first-party synthetic\n\nmemory benchmark of 30 designs repeated three times per arm, scoped 8B and 35B configurations each met\n\nthe recorded acceptance contract on 90/90 jobs, while scoped 4B and broader-context 35B configurations\n\naccepted 72/90 and 46/90; 43 broader-context responses were truncated under the fixed output budget. A\n\nseparate eight-task context-admission diagnostic retained PARTIAL overall parity despite matching the\n\ncorrected context condition. Historical memory tests provide bounded implementation evidence; worker-\n\nconstruction performance records with incomplete artifact lineage remain excluded. These observations\n\nmotivate, but do not establish, the complete ADI architecture. The proposal synthesizes established work on\n\nreference monitors, capability security, formal methods, runtime assurance, neuro-symbolic systems, and\n\ngoverned agents. Its contribution is an explicit lifecycle-wide authority contract, an evidence taxonomy, and\n\nfalsifiable evaluation questions. Independent replication, comparative evaluation, and validation of the full\n\nenforcement boundary remain open.",
  "abstract_shown": true,
  "affiliations": [
    "Cortex Agentics Global"
  ],
  "article_type": null,
  "arxiv": null,
  "author_total": 1,
  "authors": [
    {
      "affs": [
        0
      ],
      "name": "Branden Laskowski"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/2vwb0hjznnm3n.bib",
  "cited_by": 0,
  "doi": "10.5281/zenodo.23510050",
  "fulltext_url": null,
  "id": "2vwb0hjznnm3n",
  "issns": [],
  "landing_url": "https://zenodo.org/records/23510050",
  "language": "eng",
  "licence": "CC-BY-4.0",
  "licence_label": "CC BY 4.0",
  "licence_source": "article",
  "licence_url": "https://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": null,
  "native_subjects": [],
  "openalex": null,
  "pdf_url": "https://zenodo.org/api/records/23510050/files/ADI_v1.0_PUBLIC_RELEASE.pdf/content",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-20",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/2vwb0hjznnm3n.ris",
  "sources": [
    {
      "first_seen": "2026-11-03T12:20:33Z",
      "native_id": "23510050",
      "source": "zenodo",
      "url": "https://zenodo.org/records/23510050"
    }
  ],
  "subjects": [],
  "title": "Artificial Deterministic Intelligence: Probabilistic Reasoning, Deterministic Authority",
  "title_text": "Artificial Deterministic Intelligence: Probabilistic Reasoning, Deterministic Authority",
  "type": "preprint",
  "updated_at": "2026-11-03T12:20:33Z",
  "url": "https://research.draconitic.com/paper/2vwb0hjznnm3n",
  "venue": "Zenodo",
  "versions": []
}