Team Intelligence

Baseball is a sport of individual duels. So the honest team model adds them up.

Team value is overwhelmingly additive, and the engine adds honestly instead of manufacturing a synergy the sport does not have. The additive core, a replacement baseline plus the summed run creation and run prevention, carries about 90% of the answer and reconciles against the Pythagorean from the runs. The construction margin sits on top as the thin edge it actually is. And run prevention is the one genuine interaction, the arm split cleanly from the glove. Throughout, the team layer consumes finalized player KRs and never re-evaluates a player, because reading the player is Player Intelligence's job, and adding them up is this one.

Case 01 · the additive core

Roughly 90% of the answer is just the sum.

Team KR is built additively: a replacement baseline plus the sum of individual run-creation and run-prevention value, then reconciled against the run-based Pythagorean win expectation. The team is very close to the sum of its individual duels, so the engine adds honestly and then checks the sum against the runs, rather than manufacturing a synergy the sport does not have.

Replacement baselinea 48-win floor
+Sum of run creation+27 wins of individual batting value
+Sum of run prevention+16 wins of individual pitching and fielding value
=Team KR91 projected wins82% confidence
Reconciled to the Pythagorean
Additive projection: 91 winsPythagorean from runs scored and allowed: 89 wins96% agreement
Summed individual value predicts team wins nearly one to one; the Pythagorean from runs scored and allowed predicts it at about 96%. The engine adds, then reconciles, and the small gap is the honesty check, not a synergy term.
What the engine refuses to invent
Batting-order optimizationabout 1 win a season, 3 at the ceiling
Clubhouse chemistrynever shown to move wins at the scale talent does
The engine does not invent a synergy layer, because optimizing the order is worth about a win and chemistry has never been shown to move wins the way talent does. The additive core carries roughly 90% of the answer.

The replacement baseline plus the summed run creation and run prevention lands at 91 projected wins, and the Pythagorean from the runs agrees at 96%. That is the whole core: the engine adds the individual value, reconciles it against the runs, and does not reach for an emergent synergy the sport has never shown. The team is very close to the sum of its duels, so the honest model adds them up and checks the runs.

Illustrative engine read on the real additive core (replacement baseline plus summed run creation and run prevention, reconciled to the Pythagorean win expectation), each shown with confidence. Composite club, demonstration figures.

Case 02 · the construction margin

A thin edge on top, never a replacement for the core.

On top of the additive core sit the adjustments a naive sum misses, priced as the thin edge they are, never as a replacement for the core. The run-to-win conversion is geometric, positional scarcity makes the same value worth more up the middle, depth changes the full-season worth, bullpen value is leverage-weighted, and platoon structure matters.

Run-to-win conversiongeometric, not linearThe tenth marginal win costs more runs than the first, so the conversion curves instead of running flat.
Positional scarcityworth more up the middleThe same batting value is worth more at shortstop or catcher than at a corner, because the position is scarce.
Depth and injury resiliencethe full-season worthWhat the roster is actually worth across 162 games, not just the starters listed on paper in March.
Bullpen leverageleverage-weightedA reliever's value is weighted by the leverage of the innings he throws, not by raw innings.
Platoon and handednessstructure mattersThe handedness balance of the roster changes the matchup value the sum alone cannot see.
Each of these is a thin adjustment built on top of the additive core, never a replacement for it. The margin is the edge; the core is the answer.
Positional value is a flat additive term (Decision 5)
Elite catcherPlayer KR 88+2.5 for catching
Fringe catcherPlayer KR 61+2.5 for catching
Every player at a position gets the same positional adjustment, scaled only by how much he plays it, because the credit is for playing a hard, scarce position, not for being good at it. A flat catcher adjustment is added for catching, elite or replacement.
Never a multiplier, which would double-reward a good player at a premium spot by scaling his positional credit with his quality.

The two catchers carry very different player KRs and the identical +2.5 positional term, because the credit is for standing behind the plate, not for being good once he is there. Positional value is added flat and scaled only by playing time, so a premium spot never double-rewards a good player. The margin is real and thin; positional value is flat, additive, and never a multiplier.

Illustrative engine read on the real construction margin (run-to-win geometry, positional scarcity, depth, bullpen leverage, platoon structure) and flat additive positional value scaled by playing time. Composite club, demonstration figures.

Case 03 · the run-prevention split (DIPS)

The one place two players share an outcome, split cleanly.

Run prevention is co-produced by the pitcher and the fielders, the single place two players' outcomes are truly entangled. So the engine attributes it: the pitcher read through what he controls, and the defense credited for range independent of him, so a Pitching KR and a Fielding KR come apart cleanly instead of one man wearing the other's outcome.

Run preventionco-produced, the one genuine interaction on the roster
Pitching KR
82
The pitcher read through what he controls: strikeouts, walks, home runs allowed, and the weak contact he induces. The defensive-independent share.
Fielding KR
76
The defense credited for range and conversion independent of the pitcher, so a good defense does not inflate the arm and a bad one does not bury it.
Everything else on the roster sums cleanly; this is the one place it does not. The engine makes the boundary explicit, splitting the arm from the glove, because that is the only place two players truly share an outcome.

The combined run prevention comes apart into an 82 Pitching KR and a 76 Fielding KR, the defensive-independent share and the range share separated, so neither man wears the other's result. It is the single interaction the engine models, and it models it on purpose rather than smearing it across the sum. One genuine interaction, split cleanly; everything else on the roster adds.

Illustrative engine read on the real pitcher-fielder DIPS attribution (the pitcher on what he controls, the defense on range independent of him, the two KRs coming apart cleanly). Composite club, demonstration figures.

The law underneath
Add them up, reconcile the runs, and stop.

Baseball is a sport of individual duels, so the honest team model adds them up, reconciles the sum against the runs, and stops. The engine refuses to manufacture the emergent synergy the interacting sports have and baseball does not, prices the construction margin as the thin edge it actually is, and separates the arm from the glove because that is the only place two players truly share an outcome. And it consumes finalized player KRs without ever re-evaluating one, because the team is the sum of reads it did not get to redo.

Sum the duels. Check the runs.

Team Intelligence builds Team KR from the additive core, reconciles it against the Pythagorean, prices the construction margin as the thin edge it is, and splits run prevention between the arm and the glove, without ever re-evaluating a player the engine already read.

Get access