{
  "authored_package": {
    "attribution": "Zerone contributors (Yu and Ai)",
    "license": "Apache-2.0",
    "scope": "Our experiment code, generated data/results, analysis, figures and reproduction wrapper. The cited paper is not included or relicensed."
  },
  "ledger_scope": "This package contains no chain state and is not itself a verdict. A prior separate local exercise accepted a paper source-report claim using same-operator accounts. This new computational candidate was not submitted in that exercise. Any later public claim or review must be identified separately by its own receipt.",
  "literature_source": {
    "authors": [
      "John Duchi",
      "Elad Hazan",
      "Yoram Singer"
    ],
    "downloaded_pdf_sha256": "691043712c05c19c98223fba52507beff0480afb11fa17ea40a2a032d6f0bf3c",
    "title": "Adaptive Subgradient Methods for Online Learning and Stochastic Optimization",
    "url": "https://jmlr.org/papers/volume12/duchi11a/duchi11a.pdf",
    "use": "Eq. 1 motivates the diagonal update. This package's epsilon conventions, zero-step convention, synthetic data and finite experiment are our specified follow-up, not the paper authors' experimental result.",
    "venue": "Journal of Machine Learning Research 12 (2011), 2121\u20132159"
  },
  "original_execution": {
    "computational_content_sha256": "55f0f8d77b07f81b218ffe808342a6c6c89b80f434bd04cb4f2baffa73e0b787",
    "environment": {
      "implementation": "CPython",
      "platform": "macOS-15.3-arm64-arm-64bit-Mach-O",
      "python": "3.14.3 (main, Feb  3 2026, 15:32:20) [Clang 17.0.0 (clang-1700.6.3.2)]",
      "runtime_dependencies": "Python standard library only"
    },
    "original_private_input_manifest_sha256": "f2987beac3bd6fa74e227e38ce7a02e677d3a089cbed9746d73a2d2535506c11",
    "original_private_results_file_sha256": "3a7685353fe829f6003a8cc38bc81fe2e8d49f4bb1a6b1dd1135c970c572da6a",
    "review_scope": "One operator and that operator's agents; parent rerun matched computational content. No independent external reviewer or paper-author endorsement."
  },
  "original_input_sha256": {
    "dataset.json": "fea3ad27a3b53b8de7842ed701da442881fb3762214f34f1d3b1118516b70efc",
    "experiment.py": "58636f175a0b0aa984270c7ea660a46a2979e657076a293b149da0fbc4024941",
    "protocol.json": "7a4b83c2747b1a974847d130a005f11f3dcf9f6ace4f21a15eca9ea6f672c476"
  },
  "packaged_on": "2026-09-13",
  "packaging_transformations": [
    "protocol.json, dataset.json and experiment.py are byte-identical to the frozen original inputs.",
    "results.json is exactly the canonical object containing original pass, runs and trajectory_comparisons. Every float, checkpoint, full-trajectory hash and comparison is unchanged. Absolute-path execution metadata is excluded.",
    "reproduce.py verifies package-relative files and calls the frozen module's execute function; it does not invoke the original path-bearing CLI manifest mechanism.",
    "report.md retains the scientific account with public reproduction instructions and links. claim.json retains the scoped computational statement with public evidence references.",
    "Figures are exact copies of the corrected v2 render; matplotlib3.11.0 was used only for plotting. Numerical reproduction has no matplotlib dependency.",
    "Private receipts/logs/manifests, paper PDF, all chain fixtures, identities, homes and keys are excluded."
  ],
  "preparation": {
    "original_input_manifest_saved_utc": "2026-09-13T11:41:17.892900Z",
    "protocol_saved_utc": "2026-09-13T11:35:25.544573Z",
    "scope": "Local same-operator pre-specification before execution; not independently timestamped preregistration."
  },
  "schema": "zerone-adagrad-provenance/v1"
}