In-Season Projections

GP is his club's games inside the window, read off the schedule, and it is the column that makes a weekly number worth anything: the same player is a different projection in a two-game week and a four-game one. A man who is out with no return date shows as out on every horizon; one with a reported date is projected from that date on; a game-time decision is projected in full and marked with a question mark, because a game-time decision plays more often than not and zeroing him is the bigger error.

The rate behind every column is the nightly blend of the preseason projection and this season's play, weighted by games with per-stat weights backtested on last season. Conf says how much of that rate is his own season rather than ours. Role change is where the changepoint model feeds in: where it finds his power-play share or ice time has genuinely moved, the stats that scale with each are moved in proportion, weighted by the probability, and capped. The blend on its own believes a promotion slowly on purpose; this is what stops it being fifty games late.

Nothing here is a forecast of form: no streaks, no opponent, no goalie faced. That per-game layer is next, and it will sit on top of these numbers rather than replace them.

rate x games x availability. The rate is the nightly blend of the preseason projection and this season's play, adjusted where the changepoint model finds a genuine role change (see `role`). Games are the club's schedule inside the horizon; weeks run Monday to Sunday. Nothing here is a forecast of form.