The best roster is not always the best team. The gap between them is the truth.
Talent on paper is one number. What five players produce together is another. The engine reads both, and the distance between them is fit: scheme, role, spacing, and the fragility a single rating hides. It reads the system a team actually runs from its own data, grades how the roster fits it, and flags exactly where the team would break.
Talent adds up on paper. Fit decides what they produce.
Sum the roster and you get raw talent. That is not the Team KR. The engine reads what the group produces together, and the gap between the two is fit, in either direction. Here is a real full read: the Duke Blue Devils, built around a National-Player-of-the-Year hub.
The roster aggregates to 86.4. They produce 90.6 together, a +4.2 lift, because the superstar-plus-spacing construction fits: the hub creates, the shooters punish the help, the rotation holds its level. Build the same talent wrong, three ball-dominant scorers and no spacing, and the same math runs the other way: the team produces below its paper number. The gap is the read no list of ratings can give.
Real KaNeXT engine eval, Duke (D1 High-Major), public-box layer, 38 games (demo render). Greyed reads resolve at KVision.
It reads the scheme from the data, then grades the fit and names what is missing.
No one tells the engine the system. It infers the offense and defense from the team's own play data, grades how the roster fits, and maps demand by demand what is covered and what is not. Same Duke read.
A team built for what it runs: the offense and defense both grade as built for this, and the one soft spot, half-court creation against length, is named, not buried. The engine does not just score the fit; it hands you the coverage map.
Real engine eval on the OSIE and DSIE system-inference and coverage-map structure. Greyed and watch rows firm up at KVision.
The variance bands, and the fault lines a record hides.
The engine reads the team's range, its floor, its ceiling, and the version that travels in March, and runs structural fragility checks underneath. For a dominant team the variance is margin, not fragility. For a fragile one, a single flag names what breaks it.
Duke's floor sits at 87.5, so its variance is schedule and matchup, not collapse. Now the same checks on a composite team the record calls fine:
That composite team can be winning and still be built on a fault line. The record says fine; the engine says one injury from falling apart. Naming that before it happens is the point of reading the team, not just rating it.
Duke variance real (engine eval). The fragility contrast is an illustrative composite on the real Fragility Flags structure.
A list of ratings tells you how much talent is in the room. It cannot tell you whether the talent fits, whether the scheme suits it, whether one injury ends the season. Those live in the space between the roster and the team, and that space is where games are won and lost. The engine measures it: the produced number against the paper number, the system read off the floor, the coverage map, the fault lines named before they open. A team is not five ratings. It is what the five become together.
Do not just rate the roster. Read the team.
Team Intelligence is the starting point for everything the front office and the coaching staff do next.