[
  {
    "section": "s1",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "1. A forecast is defined by what was available"
  },
  {
    "section": "s2",
    "kind": "source_based_review",
    "sources": [
      "dl3-c04-1",
      "dl3-c04-2",
      "dl3-c04-3",
      "dl3-c04-4",
      "dl3-c04-5",
      "dl3-c04-6"
    ],
    "claim": "2. Architecture papers answer a different question"
  },
  {
    "section": "s3",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "3. Event time and availability time are not identical"
  },
  {
    "section": "s4",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "4. An autoregressive control makes the information gap exact"
  },
  {
    "section": "s5",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "5. A future proxy can look like a powerful representation"
  },
  {
    "section": "s6",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "6. Window construction can cross the split"
  },
  {
    "section": "s7",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "7. Normalization can carry future information"
  },
  {
    "section": "s8",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "8. Hyperparameter selection has its own clock"
  },
  {
    "section": "s9",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "9. Global models introduce cross-series availability"
  },
  {
    "section": "s10",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "10. Forecast error should be reported by horizon and regime"
  },
  {
    "section": "s11",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "11. Missingness is sometimes a message"
  },
  {
    "section": "s12",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "12. A prospective replay is stronger than a retrospective table"
  },
  {
    "section": "s13",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "13. What the constructed checks establish"
  },
  {
    "section": "s14",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "14. A timestamp audit can be tested without a model"
  },
  {
    "section": "s15",
    "kind": "author_analysis_or_verification_proposal",
    "sources": [],
    "claim": "15. The benchmark is a simulation of a decision"
  }
]