The rating is the durable claim. The game grade is the single-game check on it.
KPG grades a player's one game on the same KR scale his rating lives on, KTG grades a team's one game on the Team KR scale. They are not the rating and never overwrite it: they are the game-level evidence the pipeline re-derives on, built on expected points because in football the honest unit of a play's value is its change in expected points, not a raw box-score count. They read the process, not the scoreboard, they carry their confidence, and a persistent gap between the grades and the rating is a flag that the rating has gone stale.
A single game, on the scale the rating lives on.
A composite receiver, one game. His durable rating is a fact from Player Intelligence. His KPG is a single-game grade on the same KR scale, and it never touches the rating.
A KPG is single-phase and weightless, and a handful-of-snaps grade is a lead, not a fact, read against the position's workload. The grade is a game spoken in the rating's units, with its confidence attached.
Illustrative on the real KPG pipeline (the expected-points base action value, the quality interaction, the role and scheme context, the supporting-cast and game-state adjustment, the snap-count credibility, the level normalization, the football confidence tiers, the KVision upgrade). Composite player, demonstration figures.
How well the team played, whether or not the ball bounced its way.
A composite team, one game. KTG grades how well it played, not only whether it won, on the same Team KR scale, rolled up across the three phases the Team KR weights.
The phase split shows where the game was won and lost, and the decomposition is the diagnose-before-optimize output at the team-game level. KTG measures the game the team actually played.
Illustrative on the real KTG components (result versus the simulation's expectation, efficiency margin by phase, control factors with the turnover luck split, context stakes), the three-phase roll-up, the process-over-result flags, the lite-mode fallback. Composite team and game, demonstration figures.
A game never overwrites the rating. A run of them re-derives it.
A single KPG is high-variance and widely banded, so a loud game never moves the durable number. A run of them, converging, does, and only by triggering the pipeline, never by editing the rating directly.
The grades wire into the rest of the system as its game-level evidence: the development engine reads the trend, Game and Week Ops reads the KPGs against the plan, and a rising trend on a cost-controlled player is surplus forming. Each grade carries the confidence of its games. The rating stays the durable claim, and the grades are the running check.
Illustrative on the real cross-references (the KR-versus-KPG-trend honesty check and re-evaluation trigger, the single-game-versus-season convergence, the KTG-versus-KPG consistency check, the downstream wiring). Composite player and team, demonstration figures.
A rating is a durable, multi-game claim about a player or a team, and a single game is not that, it is one high-variance sample, so the engine grades it as a check on the claim and never as a replacement for it. KPG and KTG speak the rating's own language, on the same scales, built on expected points because the honest unit of a football play is its change in expected points and not a box-score count, and they read the process rather than the scoreboard, so a team that outplayed an opponent and lost a coin flip grades above its result and a team that won on a lucky bounce grades below it. Every grade carries its confidence and its tier, because a game read from the box and a game read from tracking are different objects, and a grade is only ever as firm as its worst load-bearing evidence. And when the grades and the rating disagree for long enough, the grades do not win by force, they trigger the pipeline to re-derive the rating on the accumulated evidence, because the check flags the claim and the pipeline settles it. Grade the game honestly, in the rating's units, at its real confidence, and let a run of honest grades, not any single loud one, move the durable number.
Grade the game in the rating's units. Let the run of them check the rating.
Game Grades reads a single game on the KR and Team KR scales, on expected points, process over result, with its confidence attached, and never overwrites the durable rating, it flags it.