Ratings and grades

Every rating in the game is a real season, scaled so a year in the 1980s and a year in the 2010s can sit on the same board. The letter at the end of a run is not that. It is your record against the club's real one.

What a player rating is

The file has 3,979 rated players, built from Wikipedia box scores. A season's raw value starts from per-36-minute production, so a starter and a bench man are compared on what they did with the minutes, not on who was simply given more of them. Points, rebounds, assists, steals and blocks all go in. The weights on the counting stats were fitted so that a club's ratings track real winning, not chosen because a round number looked tidy.

Shooting is not a flat bonus for a high field-goal percentage. A man who scores his points more efficiently than the league that season is credited for the points a league-average shooter would have missed with the same shots, and the credit has a ceiling. A very high percentage on a small number of shots, which is what a centre who almost never shoots threes can post, is not allowed to explode into a superstar season. Free throws are a separate term, because the file does not have free-throw attempts.

Availability is folded in after that. Total minutes are compared with 1,900, and the fraction is square-rooted, so a 55-game star stays clearly ahead of a 78-game journeyman instead of being flattened together, while a man who played a handful of games cannot rate like a full season. This is why an injured year looks bad. The rating is what that season was worth, not a plaque for his career.

Pace is the other problem. Teams scored 110.7 points a game in the file's 1985 baseline and 91 in 1998. A raw points-per-game figure from one of those years is not the same object as the other. Each season is scaled on its own: the ratings in that year are turned into a z-score, then placed on a 15 to 99 line with the formula 50 plus 13 times the z-score, clamped at both ends. A league-average rotation season lands near 50. An all-time season lands in the mid-90s. The highest peaks in the file, among men drafted in a year you can actually start, are 97: Giannis Antetokounmpo in 2019 (taken 15th in 2013, by Milwaukee) and Russell Westbrook in 2016 (taken 4th in 2008, by Oklahoma City).

Offence and defence are split so a club has a shape, not one opaque number. Offence is the scoring, with the shooting adjustment, plus assists. Defence is rebounds, steals and blocks. A box score cannot see team defence. Steals and blocks are thin proxies, and rebounds are only partly defensive, so the defence number ranks activity and rim protection. It is not a scouting report on a scheme. Both splits are scaled the same way as the overall, inside that season.

The draft year he did not play

A man is in a season if that season has a rating. No rating means he is not in the league, which is how retirement and a lost year both work. There is one exception, and it is only for his own draft class.

416 of the 2,861 drafted players in the file have no rating for the year they were drafted, about 14.5 percent. They did not play it. Under a plain reading they would vanish from their own board. Larry Bird, pick 6 to Boston in 1978, would not be in the 1978 draft. His first number on file is 72, in 1979, and his peak is 85 in 1987. Nikola Jokić, pick 41 to Denver in 2014, first rates at 70 in 2015 and peaks at 95 in 2021. David Robinson, the first pick to San Antonio in 1987, first rates at 80 in 1989. Joel Embiid, pick 3 to Philadelphia in 2014, first rates at 67 in 2016. Arvydas Sabonis was drafted 24th in 1986 and first rates at 78 in 1995, which is also his peak in the file.

For that one draft season, and only then, the board uses his earliest rating. He can play the season you drafted him into. That is a fiction, and a smaller one than deleting him. The exception does not spread. Sabonis is on the 1986 board at his 1995 number, then away until 1995, which the squad screen can show as the year he is back. Widening the rule to "no rating this year, so use a nearby one" would put retired players back in the league. It is not done.

The age on the card is still his age in the draft year. The rating is what he later was. Both are true, and they are allowed to come from different seasons.

TEAM, and the men behind the five

TEAM is not the average of whoever happens to be good. It is five starting spots, always divided by five. A 90 with four empty spots is 18, not 90. The spots are two guards, two forwards and a centre, then a bench. Empty starters used to be left out of the division, which made a hole look like an improvement. They count as zero now, which matches the way a club's strength is actually scored.

The words on that number, from the top: ELITE at 70 or better, STRONG at 66, SOLID at 63, AVERAGE at 60, WEAK at 57, and POOR below that. A group's own average, such as the bench, is a different figure. It only includes the men who are there. It is not the team rating, and the screens that mean one do not print the other.

When the game scores a club for a simulated night, the rotation is weighted. The five starters count in full. The next three count at 0.45 each, and the next two at 0.2. Those weights come from how minutes were really shared: across hundreds of club-seasons, the top five took about 60 percent of the minutes, the next three about 21 percent, and the ninth and tenth under 9 percent together. Men past the tenth still count for a little, because an empty end of the roster being worth more than a real player was a bad model. A star past the tenth simply takes a higher spot. He is not bonus production on top of a full rotation.

The letter at the end

The grade answers one question: did you beat what this club really did over the seasons you played? It is not a rank of how entertaining the draft was, and it is not a comparison with other people's runs until the summary says so, separately, from the anonymous tally of finished grades.

Real win totals are put on an 82-game scale first. The game always plays 82. The league did not. The file's 1998-99 records are 50 games, 2011-12 records are 66, 2020-21 records are 72, and 2019-20 varies by club, in a band from the mid-60s to the mid-70s. Comparing a full simulated season with a lockout total raw would call an ordinary year a miracle. Same rate, same length, then the gap means something.

Titles are counted as a difference: yours minus theirs. Wins are counted the same way, then divided by the number of seasons, so a long run is not graded on the raw sum. A title is worth 6 on that scale, and it is not divided by the length. One title is a fixed jump whether you played one year or twenty. The score is six times the title gap, plus the per-season win gap.

The bands, on that score: A+ at 16 or better, A at 8, B at 2, C at -7, D at -16, and F below that. With the titles level, two extra wins a season is the line for a B. One extra title and the same wins scores 6, which is a B, not yet an A. A C runs from just short of the B line down to seven wins a season worse. The words are FAR BETTER, BETTER, MUCH THE SAME, WORSE, and then the two that change with length: A+ is A DYNASTY, or A SEASON FOR THE AGES if you only played one year. F is A LOST DECADE, or A LOST SEASON.

If the file has no real baseline for that club over those years, there is no letter. The screen says so, and the tally records it as NA rather than inventing a grade.

The bands were set by playing a large number of runs with the same policy the computer clubs use, then putting a competent run near the middle instead of failing it for ordinary variance. They are the same bands at one season and at twenty. A one-season grade is mostly the roster you were given. A policy that always drafts the worst man available fails far more often over five seasons than over one, which is the draft finally being measured. If you want the drafting judged, play five.

What the numbers are not

They are not a forecast, and they are not a moral ranking. A low year can be an injury, a small role, or a man who really was that bad for a year. A high year is a box-score season, scaled against that year's peers. Defence in particular is only what a box score can see. The how to play page covers what you are allowed to do with the numbers, and the guides apply them to a few classes without smoothing the awkward picks out.