The Estimated Net Rating
One number per skater per season, made before the first game and moved by every game after it, printed with a band that has been checked.
Two numbers, named apart
The table player_season_nr_forecast holds two quantities per skater and season, and they are kept under different names because they answer different questions. Talent (talent_net, with talent_sd) is the posterior mean of his per-game Net Rating, times 84 as published (the column itself is stored times 82): the rate we think he produces at, today. The year-end projection (ye_net, with ye_lo80 / ye_hi80) is what his season Net Rating will read when the schedule is done: the games already played, as they happened, plus the talent for the games that remain. The number on a card is the year-end projection, and the band under it is that projection's 80% predictive band; the talent's own sd is never drawn as a band. Before opening night the two are the same thing; after the last game the projection is the realised season exactly, which is the first check the stage runs on itself.
Both are built from the per-game offensive and defensive components that Net Rating stores (o_rating, d_rating: goals above the positional average per game, all strengths), regular season only, and each component is forecast on its own. Goalies are not in this table; the lineup card takes their number from the goalie forecast.
The season-start prior
Before a game is played we start from a player's last three seasons. The most recent counts fully and the two before it count for less (half to seven-tenths as much for the second season back, three-tenths to a half for the third; the table below has the exact weights). A season in which he played more games counts for more. Each past season is first moved to the player's current age, then the whole is nudged toward the average for players with a similar amount of history, as if he arrived with some extra games of average play (ten to 120, depending on the component). That nudge is large for a player with little history and small for a veteran. This is a version of the Marcel projection, Tom Tango's simple system (recent seasons weighted, regressed to the average, adjusted for age); the formula is below for readers who want it. Two things are unusual about ours.
m = ( Σ_k w_k · n_k · a_k + P0 · μ[bucket] ) / ( Σ_k w_k · n_k + P0 ) a_k = mean_k + A(age_now) − A(age_k) k = 1, 2, 3 seasons back
Here n_k is games played in season k, w_k its season weight, mean_k the per-game rating that season, A the aging curve, and P0 the number of average games the nudge is worth.
The reference mean depends on how much history the player brings. A skater with fifteen games behind him is usually a fringe player, not a regular, and pulling him toward the average regular was a measured bias in the first fit. So the average he is pulled toward (μ) is that of players with a similar amount of history, grouped by games in the last three seasons: entrant, 1–40, 41–100, 101–200, 201 and up. An entrant has nothing but his bucket's mean and P0 games of information. The entrant mean itself weights each past rookie season by 1 − exp(−n/τ), with τ of 30 games for forwards and 10 for defencemen: weighting rookies by their games let the ones who played full seasons, the good ones, dominate the mean, and a fresh entrant inherited a survivor-weighted number (defence entrants ran +1.37 high in the first fit). The τ pair was chosen walk-forward on entrant error with each position's bias held inside half a goal per 82 games (0.51 per 84); a parameter-free alternative, equal weights on each rookie's shrunk mean, left defence at +1.19 and was not taken. One caveat: τ was chosen per position on the same walk-forward folds it is evaluated on, one parameter per position, so the entrant cells are mildly in-sample; the league-level numbers are not affected.
A player who has not yet played an NHL game has no forecast at all: the points projection counted him at the entrant mean and the team card showed a dash. A forward now gets a prior from his draft record instead: a regression on the log of his overall pick, whether he was drafted, his age and his pre-draft league (CHL, NCAA, Europe or other), fitted on every debutant's first season since 2011-12 with the same 1 − exp(−n/τ) weights. Walk-forward over 2013-14 to 2025-26, it cut forwards' error from 4.58 to 4.26 goals per 84 and took the entrant mean's −3.86 bias on top-ten picks to −0.39. Draft position told us nothing about defencemen (calibration slope 0.14), so they keep the entrant mean. The prior describes a rookie who dresses and stays: only 24% to 61% of debutants, depending on position and where they were picked, reach twenty games.
Aging is applied to each past season before it is averaged. A is a cumulative curve fitted on within-player year-over-year changes of each component (a polynomial in age, chosen by leave-one-season-out), so a 32-year-old's season at 29 is moved to what it would read at 32 before it counts. It is fitted here, on these components; it is not the shooter and goalie aging curve the xG walker uses. Survivorship is stated rather than corrected: a declining player drops below the pairs' 20-game floor and out of the fit, so the old end of the curve is flatter than the truth, and that shows up in the level check below.
On top of that starting estimate there is a small second stage, a penalised regression (the arm the holdout chose is …): age band and history bucket unpenalised, prior-season 5v5 minutes per game and power-play share, and the prior's square and its interaction with age band, penalised. It is fitted only on players with a prior season; entrants keep the starting estimate. It was kept because it beat the starting estimate alone for both position groups on held-out seasons without materially widening the worst decile bias (a twentieth of a goal of slack), which was the rule set before the fit.
Badges were tried and are not in the model. Prior-season badge posteriors as extra features could only be tested on the one held-out season where a prior season of badges exists, and on it they did not help. The arm stays in the research script and reruns when a second season is available.
How fast the season takes over
The starting estimate treats past games as evidence about his current talent. They are not quite that: talent moves between seasons. The uncertainty around the estimate is therefore the game-to-game noise in the games behind it plus a fitted season-to-season drift, τ², estimated by maximum likelihood on season-start residuals. That drift is what sets the in-season weighting:
v0 = s² / I0 + τ² I_eff = s² / v0 (the prior's worth, in games) post = ( I_eff · m + n · run ) / ( I_eff + n ) var = s² / ( I_eff + n ) ye = ( n · run + (N − n) · post ) / N N = expected games this season
s² is the measured game-to-game variance of the component for the position group, n the games played, run the running mean over them. The offensive component drifts; the defensive one barely does, so a three-season history is worth a few dozen games of current offence and a couple of hundred of current defence.N is the games played plus the team's remaining games at the player's availability so far, and before opening night it is the length of the schedule (84 games from 2026–27).
Does it beat the obvious?
Walk-forward, …: every parameter used for a season was fitted on the seasons before it, and the target is the realised season Net Rating of skaters with at least 20 games. Goals per 84, as the holdout was scored.
Level check, and the known misses
Bias is mean(forecast − realised) at season start, by position and age band. The gate we set ourselves was half a goal per 82 games (0.51 per 84, the size it keeps), and the table is printed whole rather than trimmed to the cells that pass.
Known misses, quoted from the holdout memo (backend/research/nr_forecast/holdout.md, the fit this page's data file comes from; goals per 84 at season start, as the memo gives them). Defencemen aged 34 and over are over-forecast by +0.98: the aging pairs only contain players who kept playing, and halving the pairs' game floor did not move it. Defencemen with 41 to 100 games of history are over-forecast by +0.61; entrants, after the change to the entrant mean above, sit at −0.35 (forwards) and −0.07 (defencemen), inside the gate. Across prior deciles there is a hump: the middle deciles run +0.5 to +1.0 high and the top decile −0.6 low, which is not explained and is recorded as open rather than corrected after the fact. Read all of it against the typical error of about four goals per 84.
The band, and the roster total
The band printed with a year-end number is the predictive 80% band: the talent's posterior variance plus the noise of the games still to come. The talent band alone covers only six or seven in ten realised seasons at season start, as it should, since a realised season carries its own noise; it is exposed under its own name and never drawn as the band. Coverage of the predictive band by games seen, and the scale applied where it fell short:
The roster total on the team card is the expected roster's: the lineup's skaters, the players on its absence list and any injured player with a forecast whom the injury report lists and the lineup page leaves off, each counted at his expected share of the remaining games (from last season's record, whether he is absent today, and for a listed injury the reported return date; a player with no forecast yet counts at the entrant prior above, not at replacement level), the twelve forwards and six defencemen of highest projection among those available dressing in each simulated game, and any gap filled at replacement level. It replaced the plain sum of the dressed eighteen on 2026-09-30, when a team with its best player on injured reserve was found priced with him at zero games for the season. The goalie part counts both listed goalies, each at his expected share of starts: his recent starts, shrunk, with the pair scaled to the 92.7% of a team's starts its top two goalies take (the rest count at league average). Tested on 2022-23 to 2025-26 it beat a fixed starter/backup split, but it runs high at the top: goalies forecast for 65% to 80% of starts averaged a 68.0% forecast and started 64.3% (53 forecasts). On a game night the named starter replaces the split for that game. The availability model was checked bucket by bucket against nine seasons of who actually dressed; one known limit: skaters with under 82 career games dress about three points more often than it says. Their errors are not independent: teammates share the ice and the on-ice expected goals in the components. Over … real team-games, the variance of the summed residual was … times the sum of the individual variances (a pairwise correlation of about …), and the total's band is widened by that factor rather than printed on an independence assumption.
Where it comes from, and what checks it
The fit is backend/research/nr_forecast_fit.py; its tables are research/nr_forecast/holdout.md and this page's data file. The parameters it chose go to scraper/nr_forecast_params.json with a fingerprint of the compute_net_rating.py they were fitted against, and the stage that writes the table, scraper.compute_net_rating_forecast, refuses to run against a different one. The arithmetic both use is scraper/nr_forecast_lib.py, one implementation, so what was validated is what runs.
Two checks run on every write: with no games in the season the projection must equal the talent, and with the season complete it must equal the realised mean; the stage raises if either gap is not zero. On the finished 2025–26 season the stage's year-end number matched player_ratings.net_rating to 4 × 10⁻¹⁰ on 715 skaters with twenty or more games, which is the identity holding, and the season-start prior for that season missed the realised value by 4.07 per 84 on average (3.97 per 82), in line with the holdout.