Matchup lab · assembled from players
Updated after every game. Green is better than today's league average, red is worse, and ± is one standard deviation · as of 2026-11-15.
Pick any two teams and the page scores the game in your browser with the same ratings the call on every game page uses — except that here every team state is built from the players in front of you: the dressed roster's filter states at their trailing usage shares, plus the coach term and the franchise conversion state, the same roster sum the engine's team table is. So the explanation runs down to people: scratch a skater, name the starting goalie, hand a team another club's coach, give a player more or fewer minutes, take home ice away, send a team back in time, change the rest, and the call, the margin's bars and the scorelines all move.
Away · tonight
Home · tonight
Lab numbers are interactive hypotheticals — never logged, never a pregame row in any ledger; the bundle's own calls are the append-only live file the shadow read scores. The rosters are each team's latest dressed lineup and the player states are the filters' current checkpoints (forecast file 2026-11-15); the coach and conversion terms and the comparison column come from the engine's stored team-state row (newest 2026-11-15) — on the Pi one game apart from the lab's sum, on a PC whose table has not rolled forward, a season. The shares are the engine's own trailing deployment (5v5, shots, starts) and the forecast's projected special-teams hours over the league's team hours (a power play's shares sum to about five skaters, a kill's to about four). A rest of 0 means a back-to-back (played yesterday); rest is capped at the engine's 7-day ceiling. Scoreline odds carry the step's fitted tie adjustment (ρ = 0.35); the environment level is 3.03 goals per side entering the next unplayed game.