Goalie Intel

Smoothed percentile of conditional P/start against prior 28 days of forecast references; initial cohort bootstrap when history is missing. Reference: 1716 values. Building reference: numerical 0–10 values are hidden until 7 reference days and 100 forecasts are available. The 0–10 value is conditional on starting and is not comparable across scoring profiles, seasons or model versions.

NHL observations fetched 2026-10-25 18:24:12 · schedule snapshot 2026-10-25 18:21:45. No personal Fantrax scoring integration or confirmed-starter feed is connected.

A quality start is an observed start with SV% strictly above .900. Exactly .900 does not qualify. Relief and zero-shot observations are excluded from observed QS%. Match QS chance uses the same simulated shot/goal outcomes as win and fantasy forecasts.

Expected contribution = model start probability × simulated conditional profile points. Saves, goals against, team scoring, wins and shutouts share each simulated scoreline. Scoring weights apply to every draw. Shown intervals are 10th–90th percentile simulation outcomes, not calibrated confidence guarantees. Start-probability ranges are heuristic uncertainty bands.

Season remaining xP sums game-level contributions. xP/G divides by the same future team-game count. Full season total adds already earned points only when the full remaining schedule is available. Missing future games are not silently filled with neutral opponents. Relief appearances count in observed points, not actual P/start.

Start share blends smoothed current-team season starts (65%) and recent ten-team-game starts (35%). Observed rest, availability and forecast back-to-back rotation adjust probabilities separately from conditional match value. Single-goalie rosters leave 15% unallocated. Current availability is an assumption for future role, not a recovery-date forecast.

When available, MoneyPuck GSAx blends the current and two preceding seasons (weights 1 / 0.65 / 0.35) with 1200 neutral prior minutes. Team xGA feeds the goal environment. Missing GSAx uses labelled NHL save-performance/league priors, with optional fresh Edge. Team rates use 15 prior games. Missing reports never establish medical clearance; previous-day team schedule and a goalie's own observed usage remain separate.

Trend uses the intersection of future game IDs in two immutable snapshots with the same season, model and scoring weights. The displayed comparable period and delta exclude actual points and games that have already elapsed. The stable threshold is max(0.05 points, 3% of prior comparable xP/G, twice the aggregate Monte Carlo sampling error). This beta threshold is not empirically calibrated yet.

Observed inputs: canonical NHL roster, NHL Stats game reports, NHL season schedules/Live, availability records, optional Edge and attributed MoneyPuck CSV data. Model outputs: start probability, outcome bands, conditional points, win support, contribution and season projection. No confirmed starters, bookmaker odds or personal Fantrax integration.