Game Grades

A single game, read in the same currency as the rating. Without ever becoming the rating.

KPG and KTG grade one game on the KR and Team KR scales, built on run expectancy and win probability, so a game and a rating speak the same language. The defining property is that they grade the process, not the box score, so a hard-hit out grades well and a bloop grades modestly. The pitcher is read on the DIPS split, the defense and the park removed; every grade is matchup-resolved and carries its confidence and data tier; every parameter is a versioned v0 calibration target; and the grade feeds the form trend without ever overwriting the OVERALL KR or the Team KR.

Case 01 · the hitter KPG

Grade the swing, not the scorer.

The engine values each event in run-expectancy and win-probability terms, then splits the result from the process: the actual RE24 read against the expected contribution from quality of contact, the exit velocity, the launch angle, and the expected outcome of each batted ball. A game of hard-hit outs grades well; a game of lucky bloops grades modestly.

68KPG64% confidenceproxyv0 calibration target
0 for 4, and a good game
Lineout to centerout104 mph, 12 deg.640 xwOBArobbed
Flyout to the trackout101 mph, 30 deg.560 xwOBAbarrel
Groundout, hardout99 mph, -2 deg.300 xwOBAsolid
Strikeoutoutno ball in playno BIPweak
Result line: 0 for 4, hitless
Process line: two barrels, all hard contact
The luck gap runs in his favor here: the box score says quiet, the contact says loud, and the grade reads the contact.
This is believable-over-impressive at the game level: the grade never rewards a bloop or punishes a robbed line drive. The luck gap is surfaced, not hidden, so a hitless game of barrels grades well and a three-hit game of bloops grades modestly.

He went 0 for 4 and graded a 68, because two of the four outs were barrels and all four balls were struck hard. The grade reads the process the box score buried, and it surfaces the luck gap instead of letting the hit column tell the story. A hard-hit out grades well; a bloop grades modestly. The grade reads the swing, not the scorer.

Illustrative engine read on the real hitter KPG (each event in RE24 and WPA terms, the result split from the process by quality of contact, the luck gap surfaced). Composite hitter, demonstration figures; parameters are v0 calibration targets.

Case 02 · the pitcher KPG, the DIPS split

Graded for the arm, not the glove behind it.

The pitcher KPG reads his actual runs against what he controlled: strikeouts, walks, and quality of contact allowed, the stuff and command, the DIPS split at the game level. A pitcher betrayed by his defense or a bandbox is not graded for the runs they gave up, and one bailed out by a great defense is read for what he actually did.

74KPG61% confidenceproxyv0 calibration target
On the line5 earned runs on the line
What he controlled
Strikeouts8
Walks1
Quality of contact allowed.290 xwOBA against, soft
Removed: not his to answer for
Two runs on a misplayed fly ballThe defense gave those up, not the arm.
One run on a cheap wall-scraperThe bandbox park, not the pitch.
About 2 runs on what he actually controlled
Matchup-resolved: this was a strong lineup in a hitter's park, so the events are read against what he faced, not the raw line. A quiet start against an ace in a big park is not scored as a poor one either.
Read the actual runs against what he controlled, and the grade comes apart from the box score cleanly. The defense and the park are removed on both sides, so a pitcher is neither punished for a misplay nor rewarded for a robbery.

Five earned runs on the line, but eight strikeouts, one walk, and soft contact underneath, with two runs on a defensive misplay and one on a bandbox removed. On what he controlled it is about a two-run start, and the KPG grades that, not the scoreboard. The DIPS split at the game level: graded for the arm, not for the glove behind it or the wall behind that.

Illustrative engine read on the real pitcher KPG (actual runs against the DIPS-controlled events, defense and park removed, matchup-resolved). Composite starter, demonstration figures; parameters are v0 calibration targets.

Case 03 · KTG and the honesty check

The check on the standing number, never the replacement.

KTG grades the team's single game on the Team KR scale, built from the same RE24-and-WPA foundation. The honesty check is the point: a run of KPG and KTG grades that diverges from the standing KR is the signal the Pipeline re-evaluates on. The game grade is the form read that tests the durable rating, never the thing that replaces it.

58KTG66% confidenceproxyv0 calibration target
The single-game team grade, on the Team KR scale
Standing Team KR 84.0the durable rating, unchanged
Recent single-game grades, all below the standing KR
6158635760
Five straight grades below the standing 84.0 is a form read worth weighing. The divergence is the signal the Pipeline re-evaluates on, not a number that rewrites the rating on the spot.
The game grade feeds the form trend and the Pipeline weighs it, and it never overwrites the OVERALL KR or the Team KR. One game is evidence; the rating is the verdict; and the check on the durable number is a run of grades, not a single line.
A two-way player is graded twice
As a hitter: 66 KPG
As a pitcher: 79 KPG
Two grades, reported separately and never averaged, the same rule the rating follows.

KTG puts the team's game on the Team KR scale, and a run of soft grades below the standing 84.0 is exactly the divergence the Pipeline is built to weigh, without ever letting one game or one week overwrite the rating. The two-way player is graded twice and never blended, because averaging would hide the very thing that makes him rare. The game grade is the honest check on the standing number, never the replacement.

Illustrative engine read on the real KTG and the honesty check (the single-game team grade on the Team KR scale, the form-trend divergence as the Pipeline's re-evaluation signal, the two-way player graded twice and never averaged). Composite club, demonstration figures; parameters are v0 calibration targets.

The law underneath
Evidence, never the verdict.

A game grade speaks the same language as the rating and stays subordinate to it. It reads process over result, resolves the matchup and the park, carries its confidence and data tier, and feeds the form trend. It is the honest check on the standing number, never the replacement, because one game is evidence and the rating is the verdict. A hard-hit out grades well and a bloop grades modestly, a pitcher is graded for what he controlled and not for the glove behind him, and a run of grades that diverges from the standing KR is the signal the Pipeline weighs, not a number that overwrites it.