# Dataset manifest for MKT-006 — two roads to the same champion: a homegrown
# World Cup model side by side with Goldman Sachs.
# Schema: schemas/data-manifest.schema.json (EPIC §8.2).
#
# The Kickoff and devigged-market figures are the frozen pre-kickoff board of
# record from the world-cup engine (snapshot c7f3f5afa805, seed 20260611),
# recorded in world-cup/reports/2026_dry_run.md, engine_choice.md,
# market_comparison.md and backtests.md. The Goldman Sachs champion figures are
# from Exhibit 7 of the 29 May 2026 GS Global Investment Research note, with the
# top-eight teams independently verified against the note and reputable
# secondary coverage. market_deviation is rebuilt by notebook.ipynb running top
# to bottom with no network access. The build ledger is derived from the
# world-cup and poly git histories and from public facts about the GS note.
paper_id: MKT-006
license: cc-by-4.0

datasets:
  - id: champion-marginal
    description: |
      Head-to-head champion (tournament-winner) probabilities for the 2026 World
      Cup for the nine teams that anchor the comparison: the Kickoff Elo bracket,
      the Goldman Sachs model, and the devigged Polymarket baseline. Kickoff and
      market columns are the frozen pre-kickoff board (world-cup/reports/
      2026_dry_run.md); the Goldman Sachs column is Exhibit 7 of the 29 May 2026
      note. Probabilities, not percentages.
    file: champion_marginal.csv
    rows: 9
    schema:
      team: string
      kickoff_p: float
      gs_p: float
      market_p: float
    source: world-cup-report+gs-note
    collection_window: [2026-05-29, 2026-06-11]
    license: cc-by-4.0
    sha256: 898234e3db91cfbe157b92edb6a51f5c855a801af31ee7a98ce5654076ddd16f

  - id: market-deviation
    description: |
      Each model probability minus the common devigged-market probability, in
      percentage points. Positive is overweight versus the market, negative is
      underweight. Rebuilt by notebook.ipynb from champion-marginal.
    file: market_deviation.csv
    rows: 9
    schema:
      team: string
      market_p: float
      kickoff_dev: float
      gs_dev: float
    source: derived
    derived_from: [champion-marginal]
    collection_window: [2026-05-29, 2026-06-11]
    license: cc-by-4.0
    sha256: 31400e0dbc6a555f69339da7d816ab0368533f004b952f744e031c16a84d8f53

  - id: champion-backtest
    description: |
      Beat-the-closing-line test on the realized champion of the 2010, 2014,
      2018 and 2022 World Cups. model_p is the Kickoff Elo-bracket pre-tournament
      champion probability (no leakage); market_p is the devigged closing
      outright price for the same champion. Source: world-cup/reports/
      backtests.md and market_comparison.md. Identical to the MKT-004 backtest
      table by construction.
    file: champion_backtest.csv
    rows: 4
    schema:
      year: int
      champion: string
      model_p: float
      market_p: float
      model_log_loss: float
      market_log_loss: float
      model_beats_market: bool
    source: world-cup-backtest
    collection_window: [2010-06-11, 2022-12-18]
    license: cc-by-4.0
    sha256: 224b5f3428284c6c7cc3755b58233a0256ab1acc8acc861064cb1983e5df3caf

  - id: match-calibration
    description: |
      Held-out per-match calibration of the two Kickoff team-strength engines
      (Elo and Dixon-Coles) on 1,767 international matches. Brier, log-loss and
      expected calibration error on the home-win/draw/away outcome. Source:
      world-cup/reports/engine_choice.md and match_calibration.md. Establishes
      that Dixon-Coles is the better per-match model even though Elo is selected
      for the champion marginal by calibration.
    file: match_calibration.csv
    rows: 2
    schema:
      engine: string
      brier: float
      log_loss: float
      ece: float
      n: int
    source: world-cup-backtest
    collection_window: [2014-01-01, 2026-06-02]
    license: cc-by-4.0
    sha256: c256ec93b1b26a122d5721c750f32599eb36275db625e2201239bb34e96cae42

  - id: engine-selection
    description: |
      Held-out champion-marginal log-loss of the two Kickoff engines. Elo scores
      1.889 and Dixon-Coles 2.504, which is why Elo is selected for the champion
      forecast even though Dixon-Coles is the better per-match model. Transcribed
      from world-cup/reports/engine_choice.md.
    file: engine_selection.csv
    rows: 2
    schema:
      engine: string
      champion_log_loss: float
    source: world-cup-backtest
    collection_window: [2014-01-01, 2026-06-02]
    license: cc-by-4.0
    sha256: df7a7f311124ccb205b8c5bf25bfef9906b124ad9fee3a0d74f2c8297836316c

  - id: assessment-rubric
    description: |
      The seven-dimension assessment of the two systems as champion-marginal
      forecasters, scored one to five per dimension with the edge attributed.
      Transcribed from the head-to-head comparison report
      (world-cup/reports/gs_vs_kickoff_winner_comparison.md, section 8). The
      dimensions are orthogonal goods and are not a summable score.
    file: assessment_rubric.csv
    rows: 7
    schema:
      dimension: string
      goldman_sachs: int
      kickoff: int
      edge: string
    source: manual
    collection_window: [2026-06-10, 2026-06-12]
    license: cc-by-4.0
    sha256: ecc1730a40fa1ff2373bddb6047f4388c0acdbdd0ea35a1ebd4e97c0f3cca214

  - id: build-ledger
    description: |
      The construction contrast across seven dimensions: builders, build span,
      codebase, match corpus, data cost, reproducibility and validation. The
      Kickoff column is derived from the world-cup and poly git histories
      (commit dates, commit counts, lines of Python) and the frozen snapshot
      date; the Goldman Sachs column states only public facts about the note.
    file: build_ledger.csv
    rows: 7
    schema:
      dimension: string
      goldman_sachs: string
      kickoff: string
    source: git-and-public
    collection_window: [2026-06-02, 2026-06-12]
    license: cc-by-4.0
    sha256: aa6acebd051d350f8f533eafeb73e8cd3ab0462982fbb1c390dd00a7c485a342