{
  "_meta": {
    "schema_version": "1.0",
    "site_url": "https://calibrationledger.com",
    "feed_url": "https://calibrationledger.com/api/forecasts.json",
    "human_readable_url": "https://calibrationledger.com/track-record/",
    "creator": {
      "name": "Paulo de Vries",
      "url": "https://calibrationledger.com/about/"
    },
    "license": "https://creativecommons.org/licenses/by/4.0/",
    "methodology_url": "https://calibrationledger.com/methodology/",
    "discipline": [
      "Probability assigned at posting is locked.",
      "Resolution date stated at posting; no moving goalposts.",
      "Public source URL required for every resolution.",
      "No retroactive deletion of resolved forecasts.",
      "Aggregate Brier and Murphy decomposition recomputes on every resolution."
    ],
    "schema": {
      "forecast": {
        "id": "stable identifier (slug-or-uuid)",
        "question": "the resolvable question, stated unambiguously",
        "probability": "probability assigned to YES at posting (0.0-1.0)",
        "posted_at": "ISO 8601 timestamp of posting",
        "resolution_at": "ISO 8601 expected resolution date (may be approximate)",
        "domain": "topical category (geopolitics, ai_benchmarks, markets, weather, sports, technology, other)",
        "source_url": "URL where resolution will be verified",
        "status": "open | resolved | unresolvable",
        "outcome": "0 (no) | 1 (yes) | null (still open)",
        "resolved_at": "ISO 8601 timestamp of when outcome was recorded (null if open)",
        "resolution_source_url": "URL of the verifiable resolution evidence (null if open)",
        "brier_score": "(probability - outcome)^2 (null if open)",
        "notes": "optional notes; appends-only after posting"
      },
      "aggregate": {
        "total_posted": "count of all forecasts ever posted",
        "total_resolved": "count of forecasts with status=resolved",
        "total_unresolvable": "count of forecasts marked unresolvable",
        "mean_brier": "average Brier across resolved forecasts (null if total_resolved < 1)",
        "reliability": "Murphy 1973 reliability term (null if total_resolved < 10)",
        "resolution": "Murphy 1973 resolution term (null if total_resolved < 10)",
        "uncertainty": "Murphy 1973 uncertainty term (null if total_resolved < 10)",
        "calibration_curve": "array of (predicted_bin_midpoint, observed_frequency, count) tuples (null if total_resolved < 10)",
        "computed_at": "ISO 8601 timestamp of last aggregate recomputation"
      }
    }
  },
  "aggregate": {
    "total_posted": 0,
    "total_resolved": 0,
    "total_unresolvable": 0,
    "mean_brier": null,
    "reliability": null,
    "resolution": null,
    "uncertainty": null,
    "calibration_curve": null,
    "computed_at": "2026-04-27T17:00:00Z"
  },
  "forecasts": [
    {
      "id": "eu-ai-act-art50-2026-08-02",
      "question": "EU AI Act Article 50 (transparency obligations for providers and deployers of certain AI systems) becomes enforceable on schedule on 2026-08-02.",
      "probability": 0.85,
      "posted_at": "2026-04-27T17:30:00Z",
      "resolution_at": "2026-08-02T00:00:00Z",
      "domain": "geopolitics",
      "source_url": "https://artificialintelligenceact.eu/article/50/",
      "status": "open",
      "outcome": null,
      "resolved_at": null,
      "resolution_source_url": null,
      "brier_score": null,
      "notes": "Calibration Ledger's positioning explicitly cites Article 50 as a buyer-education catalyst (per IDENTITY.md). Resolution: official EU AI Office announcement or EUR-Lex publication confirming the date held."
    },
    {
      "id": "gpt5-gpqa-diamond-85pct-2026-12-31",
      "question": "A frontier large language model (any vendor — OpenAI, Anthropic, Google DeepMind, Meta, etc.) scores ≥85% on GPQA-Diamond by 2026-12-31, with the score reported in an official model card or peer-reviewed evaluation.",
      "probability": 0.55,
      "posted_at": "2026-04-27T17:30:00Z",
      "resolution_at": "2026-12-31T23:59:59Z",
      "domain": "ai_benchmarks",
      "source_url": "https://github.com/idavidrein/gpqa",
      "status": "open",
      "outcome": null,
      "resolved_at": null,
      "resolution_source_url": null,
      "brier_score": null,
      "notes": "GPQA-Diamond is the hardest 198-question subset of the Graduate-level Google-Proof Q&A benchmark (Rein et al. 2023). Frontier models trended ~50-78% through 2025-Q4. Resolution: any vendor's official model-card or arXiv eval reporting ≥85% on the diamond split."
    },
    {
      "id": "sp500-up-yoy-2026-12-31",
      "question": "S&P 500 index closing value on 2026-12-31 (last trading day) is higher than its closing value on 2026-01-02 (first trading day of 2026).",
      "probability": 0.62,
      "posted_at": "2026-04-27T17:30:00Z",
      "resolution_at": "2026-12-31T22:00:00Z",
      "domain": "markets",
      "source_url": "https://www.spglobal.com/spdji/en/indices/equity/sp-500/",
      "status": "open",
      "outcome": null,
      "resolved_at": null,
      "resolution_source_url": null,
      "brier_score": null,
      "notes": "S&P 500 historical base rate of positive calendar years ≈70% (1928-2025). Probability shaded down for 2026-specific uncertainty (rate environment, elections, geopolitics). Resolution: official S&P Dow Jones Indices closing value."
    },
    {
      "id": "anthropic-next-claude-2027-q1",
      "question": "Anthropic publicly announces a successor model to Claude Opus 4 (named e.g. 'Claude 5', 'Claude Opus 5', or any next-major-tier model) by 2027-03-31.",
      "probability": 0.70,
      "posted_at": "2026-04-27T17:30:00Z",
      "resolution_at": "2027-03-31T23:59:59Z",
      "domain": "technology",
      "source_url": "https://www.anthropic.com/news",
      "status": "open",
      "outcome": null,
      "resolved_at": null,
      "resolution_source_url": null,
      "brier_score": null,
      "notes": "Anthropic release cadence ≈ 6-12 months between major model lines through 2024-2025. Probability assumes pattern holds. Resolution: official anthropic.com announcement or model card."
    },
    {
      "id": "calibrationledger-track-record-10forecasts-2027-04-27",
      "question": "calibrationledger.com/track-record/ has at least 10 publicly posted dated probabilistic forecasts (status open OR resolved) on 2027-04-27.",
      "probability": 0.50,
      "posted_at": "2026-04-27T17:30:00Z",
      "resolution_at": "2027-04-27T23:59:59Z",
      "domain": "other",
      "source_url": "https://calibrationledger.com/api/forecasts.json",
      "status": "open",
      "outcome": null,
      "resolved_at": null,
      "resolution_source_url": null,
      "brier_score": null,
      "notes": "Self-referential forecast on operator execution. Probability calibrated to honest current state: this is the 5th forecast posted today; reaching 10 within 12 months requires the operator posts ≥1 every ~2.5 months on average. 50% reflects genuine uncertainty about whether operator sustains the cadence vs. project drift."
    }
  ]
}