{
  "title": "AlphaCOGANT: Recursive Corporate Self-Improvement as Active Inference",
  "version": "1.0.1",
  "doi": "10.5281/zenodo.20976824",
  "doi_url": "https://doi.org/10.5281/zenodo.20976824",
  "zenodo_record": "https://zenodo.org/records/20977774",
  "record_id": "20977774",
  "publication_date": "2026",
  "resource_type": {
    "title": "Journal article",
    "type": "publication",
    "subtype": "article"
  },
  "creators": [
    {
      "name": "Daniel Ari Friedman",
      "affiliation": "Atta Labs",
      "orcid": "0000-0001-6232-9096"
    },
    {
      "name": "Tucker Cahill Chambers",
      "affiliation": "Atta Labs"
    }
  ],
  "description": "The AlphaFund whitepaper reframes recursive self-improvement (RSI) as a portfolio\noptimization problem:  a corporation recursively improves when realized economic\ngains finance the next cycle of better prediction and deployment, and the firm's\nstanding is summarized by t-RSI, a standardized gap between alpha-creation and\nalpha-decay rates. AlphaCOGANT observes that this construction is, term for\nterm, an Active Inference agent  — and makes the correspondence executable.\n\nWe render AlphaFund's Economic World Model (EWM) as a generative model written\nin Generalized Notation Notation (GNN), produced by the COGANT\ncodebase-to-GNN translation pattern. The firm's five capital channels —\nInvestments, Sensors, Actuators, Parameters, and R&D — become the hidden-state\nfactors of a partially-observed model; capital allocation becomes the control\nvector; and the portfolio optimizer's marginal-return objective becomes\nExpected Free Energy (EFE) minimization. The EFE decomposition supplies a\nprincipled reading of AlphaFund's own categories: its pragmatic value is\nexpected log-equity growth (the alpha-creation rate, read off the broker ledger),\nand its epistemic value is the information gain about the EWM that Sensors and\nR&D purchase (the data-scaling and forecast-sharpening laws). t-RSI is recovered\nas the standardized distance between the create-rate and decay-rate posteriors —\nthe thresholded EFE-improvement certificate that admits a self-improvement commit\nonly when creation confidently exceeds decay.\n\nWe give the technical and computational realization: a GNN model file for the\nfive-channel firm, a tested NumPy Active Inference engine that performs state\ninference, computes the epistemic/pragmatic EFE split and the marginal-return\nvector, and evaluates the t-RSI certificate. We argue that GNN-via-COGANT brings\ntwo things AlphaFund's program needs and Active Inference already enforces:\nfiltration integrity (the model may condition only on information available\nat decision time — the same \"no-peeking\" discipline that separates an EWM from a\nlanguage model) and auditable capital allocation (every admissible funding\nmove has a negative-EFE score under a single, legible objective). This is not\nfinancial advice; it is a demonstration that this reduced\nrecursive-corporate-self-improvement model has a direct Active Inference\nrepresentation supported by source-owning methods and artifact checks .\n\n---\nAssociated artifacts\nGitHub release: v1.0.1 (https://github.com/docxology/alphacogant/releases/tag/v1.0.1)\nDOI: https://doi.org/10.5281/zenodo.20976824\nZenodo: https://zenodo.org/records/20976824\nPDF SHA-256: 41efa7a8a98e6a67cece68377b8f0c5f19304e85c664ec9516353ee24eb0421f",
  "keywords": [
    "active inference",
    "expected free energy",
    "recursive self-improvement",
    "Generalized Notation Notation",
    "economic world model",
    "portfolio optimization",
    "epistemic value",
    "reproducible research"
  ],
  "files": [
    {
      "name": "Friedman_2026_Alphacogant_41efa7a8.pdf",
      "size_bytes": 2888068,
      "checksum": "md5:5d57b61baf89f039a601f0668a327691",
      "download_url": "https://zenodo.org/api/records/20977774/files/Friedman_2026_Alphacogant_41efa7a8.pdf/content"
    }
  ],
  "related_resources": [
    {
      "type": "repository",
      "url": "https://github.com/docxology/alphacogant"
    }
  ],
  "github_repo": "docxology/alphacogant",
  "github_release_url": "https://github.com/docxology/alphacogant/releases/tag/v1.0.0",
  "release_tag": "v1.0.0",
  "release_name": "AlphaCOGANT v1.0.0",
  "pdf_sha256": "",
  "pairing_confidence": "strong",
  "pairing_evidence": [
    "github_release_mentions_doi",
    "github_repo_self_linked"
  ],
  "checked_at": "2026-06-27T22:20:00Z"
}