Find the signal.
Build the sample. Explore the trade-offs. Make the next decision.
INTELLIGENCE
Loading workspace…
Expected-completion model audit
Ambition, execution and delivered progress
Default axes: harder pass mix →, more completed forward gain ↑. Select a player for completion above expectation. Bubble size reflects minutes.
Comparison studio 0 / 3
Choose up to three players. Bars share the visible cohort scale for each metric; these are raw rates, not percentiles or overall ratings.
Players in view
| Player | Position | Compare |
|---|
Metric definitions & interpretation
Expected completion uses regularised logistic regression on pass origins, destinations, length, direction, lateral position and supplied long-ball, cross, header, through-ball and restart tags. Each fixture is excluded from its own training fold; at least three other fixtures and 300 training passes are required. Training uses both teams in the selected competition/date sample. Player estimates require 30 scored passes and 80% coverage. This controls for observed pass mix, not pressure, available alternatives or opposition quality. Positive completion above expectation does not by itself establish attacking value. Through-ball tags are supplied; no defensive-line break is inferred. Forward gain is a coordinate measure, not xT.
Event rates require minutes and recorded event data for every player fixture in the sample. Source availability is not proof of complete capture. xA requires a supplied value in every fixture. Different positions, opponents and competitions create different opportunities. All competitions is an exploratory option, not a like-for-like benchmark.