The rating is the claim. The grade is the check.
A rating says what a player or a team is worth. A game grade says what they actually did in one game, on the same universal scale, so the two sit side by side and the gap between them is legible. KPG is the single-game player grade, KTG the single-game team grade, both role-weighted, context-adjusted, and level-normalised so they read the performance and not the circumstances. A single game is a wide-confidence point estimate and the trend is the signal, and when the grade and the rating disagree over a run the engine flags it to investigate, never quietly rewriting the rating. It grades what happened. It never moves a rating.
What you are rated, and what you did, measured the same way.
KPG is built on the same 0 to 100 scale as KR, on purpose, so the two are directly comparable. KR is the standing claim about how good a player is; KPG is what he actually did in this game. Because they share a scale, the distance between them is a number, not a feeling, and a run of KPGs against the standing KR is the honesty check the whole metric exists for.
Because the grade lives on the same scale as the rating, a divergence is a number and not a vibe: rated 85, grading around 78, so either the rating is high or the role is wrong, and the engine says which to check rather than silently moving the number. A single game is noise, so the trend is the signal. A rating you can never check against performance is a claim with no receipt.
Illustrative engine read on the real KR-KPG cross-reference (KPG on the universal KR scale, the trend as the signal, convergence versus sustained divergence, no auto-mutation). Composite player, demonstration figures.
It rewards doing your job well, not filling a box score.
A grade built on raw counting stats rewards volume and the players who chase it. KPG is built on possession value instead: each action the player is involved in is scored by its effect on scoring minus conceding, expected points per possession, then weighted by his role and archetype, scaled by the quality of what he did against its difficulty, and corrected for his teammates and the game state.
Empty volume rewarded, padded lines inflated.
A quiet defensive game read as "did nothing."
A garbage-time flurry read as production.
Weighted by the position and archetype (the same OPF as KR).
Quality scales the base: a hard shot made beats an open one.
Corrected for the supporting cast and the game state.
The grade is tuned to be right rather than flattering: a player who did his job well in a hard, live game scores well with modest counting numbers, and a padder in a dead game does not. It reads the contribution the role existed to make, at the difficulty it was made, corrected for who he played with and the state of the game. Believable over impressive, one game at a time.
Illustrative engine read on the real KPG stages (possession-value action base weighted by position and archetype, quality scaling the base, supporting-cast and game-state adjustment, minutes credibility, level normalisation through KLVN, the KVision tracking upgrade for the off-ball actions). Composite players, demonstration figures.
Players roll up into the team grade. They do not add up to it.
KTG grades a team's single game against what the team was trying to do, its scheme and its game model, combining outcome and process, corrected for opponent and stakes, on the same Team KR scale. It reads four components, and the individual KPGs roll up into it as one input, but the team grade is never the sum of the player grades, because shape, scheme, and game state matter beyond any individual line.
A team is graded against what it was trying to do, corrected for opponent and stakes, on the process as much as the result, so a side that dominates and loses and a side that wins against the run of play both read honestly. The seam back to the players stays open, one bright individual performance in a poor team result preserved as the true useful thing it is. The team grade explains who drove it without erasing what they did.
Illustrative engine read on the real KTG structure (the four components RVE, EFF, CTRL, and CTX, and the KPG roll-up that is one input and not a sum). Composite team, demonstration figures.
A rating is a standing claim about how good a player or a team is; a grade is what they actually did in one game, measured on the same scale so the two can be held up against each other. The grade reads the job the role existed to do, at the difficulty it was done, corrected for the supporting cast and the game state, and tuned to be believable rather than impressive, because a possession defended well is worth more than a stat-sheet line filled in a decided game. A single game is noise and the trend is the signal, and the trend against the standing rating is the honesty check the whole thing exists for; when they disagree over a run the engine flags it to investigate on the proper evidence and never once reaches back to quietly rewrite the rating. The team grade rolls the players up without being their sum, and preserves the gap between a bright player and a poor result rather than reconciling it away. The claim and the check are two different objects, kept separate on purpose, because a rating nobody ever checks is just an opinion that stopped listening.
Rate what they are. Grade what they did.
Game Grades put a single game on the same scale as the standing rating, so the gap between claim and check is legible and always investigated, never buried, and never mutating the rating.