Game Control

Both parts are shares between 0 and 1, from the home team’s side. Above 0.50 means the home team is controlling play; below 0.50 means the away team is. If a part is missing, the remaining weights are scaled up instead of substituting a bad fallback.

85%True-5v5 xG share

xGF / (xGF + xGA) — 5v5 only

The backbone. True five-on-five expected-goal share removes power plays and penalty kills, isolating the even-strength battle.

15%All-situations xG share

xGF / (xGF + xGA) — all strengths

A smaller correction that keeps special teams and full-game chances in view without letting them outweigh the five-on-five signal. Pulled-goalie play is left out of this part.

Live games

Answers who is controlling the game right now based on the tracked process signals accumulated so far.

Final games

Becomes a recap view of which team carried more underlying play over the full 60 minutes — independent of how the goals fell.

Not a win probability. Game Control does not estimate the likelihood of winning. A team can control the process and still lose.

Not a result simulation. It does not forecast how a game would end if replayed. It summarizes the process signals that actually occurred.

Not score-state adjusted. Raw xG is used intentionally. Score-state adjustments can mask real differences in team quality by assuming teams coast with leads.

Not driven by empty-net chaos. The light all-situations term excludes pulled-goalie states so late desperation sequences do not distort the read on who played better.

On live games, the card adds a confidence label based on elapsed game time and shot-event volume from the unfiltered game-control feed.

Early reads are intentionally framed as volatile. As the game builds a larger process sample, the label steps through developing, solid, and stable reads.

Model version

current xG composite release

How it performs

Uses the NHL-native xG output with fixed 85% true five-on-five and 15% all-situations weights.

What it doesn’t do

Process share is not win probability and does not claim the team with the better share deserved the final score. More than one version of the expected-goals model can feed it, so the label does not name a version; which version supplied a given value is documented and is being verified.

Validation evidence

Game Control fixes the mix at 85% true five-on-five xG share and 15% all-situations xG share, leaves out pulled-goalie play, and reweights the rest when one part is unavailable. Accepted model versions and their validation evidence are documented.

Game Control is a summary stat, not a full audit. It will not capture every edge case — particularly short, high-variance games where a handful of chances dominate the xG totals. For a deeper picture, consult the raw xGF%, special-teams results, goalie value, and shot-danger breakdowns on the game card.

Open a game and start in Control for the process edge, then use Shot Quality and Event Timeline to audit the result.