Skip to content
Use Core Analytics with your model

Use Core Analytics with your model

Integrate Foresportia analytics, compare probabilities, rank upcoming matches, or enrich an existing model.

Use Core Analytics as features

Core Analytics is designed to be a structured analytics source, not a finished model. Use Starter to integrate Foresportia outputs, rank upcoming matches, compare probabilities, or enrich an existing model.

# Feature extraction only — bring your own labels and model.
payload = match_core_analytics
ratings = payload["ratings"]
probabilities = payload["probabilities"]

features = {
    "home_elo": ratings.get("elo_home"),
    "away_elo": ratings.get("elo_away"),
    "elo_difference": ratings.get("elo_difference"),
    "elo_reliable": ratings.get("elo_reliable"),
    "home_probability": probabilities["one_x_two"]["home"],
    "draw_probability": probabilities["one_x_two"]["draw"],
    "away_probability": probabilities["one_x_two"]["away"],
    "confidence": probabilities.get("confidence"),
    "entropy_bits": probabilities.get("entropy_bits"),
    "expected_total_goals": payload["forecast"].get("total_goals"),
    "home_position": payload["standings"]["home"].get("position"),
    "away_position": payload["standings"]["away"].get("position"),
}

# Drop the row when Foresportia flags it, instead of comparing or enriching with noise.
if payload["quality"]["warnings"] or ratings.get("elo_reliable") is False:
    features = None
Read the real shape. ratings and forecast are flat objects: use ratings["elo_home"], not ratings["home"]["elo"]. A wrong path silently yields None for every row and quietly destroys the feature. See Ratings.

Compare and evaluate without the usual traps

  • Evaluate comparisons chronologically. Never mix observations across the kickoff boundary.
  • Exclude any post-kickoff information. Each payload is produced before kickoff, at its as_of.
  • Keep the Foresportia quality.warnings and drop or down-weight flagged rows.
  • Measure calibration, not only accuracy. A confident wrong model is worse than an honest one.
Historical scope. Starter supports historical analysis, chronological evaluation and backtesting over up to 90 rolling days of verified pre-match snapshots. This bounded window is not a multi-season training corpus; the upcoming Pro offer is intended for long-history training, calibration and backtesting.

Related

Probabilities · Ratings · Bulk and units to collect many matches cheaply.

↑ Back to top Back to contents