Limitations

The current advanced models (expected duel probability, round win probability, and the value-added metrics built on them) are interpretable baselines. Their coefficients are sanity-calibrated on a synthetic corpus, not fitted on production match data. They are labeled baseline throughout and their confidence is capped accordingly. Directional comparisons within a game are meaningful; absolute values will shift once the models are refit on real data. Every model's calibration status is on its model card.

Percentiles are shown only when a subject meets the metric's minimum sample within the selected window and peer group; below that, the raw value is shown without a rank rather than implying a precision the data doesn't support. Sample sizes accompany every ranked value.

Advanced metrics require event-level telemetry, which exists only for matches with a parsed replay/demo or an equivalent source. Matches with official basic data only show conventional statistics. Coverage is labeled on every match, tournament, and player page, and the coverage & sources page summarizes it per game.

Official basic statistics and model-derived advanced statistics are never mixed silently. Each value carries a provenance label; the provenance page explains the labels and the source-authority order used when sources disagree.

While in preview, the site is populated with synthetic data — fictional teams and players — to demonstrate the analytical tools. Synthetic records are marked, excluded from search indexing, and will be removed before production coverage begins. Nothing synthetic represents a real competitor or event.