#434 — what the shipped roles should weight

Every figure here is your save, scored by the app's own arithmetic — RoleScoring reimplemented line for line, priors and shrinkage and declared scales included, so the order on the left is the order the depth grid would show. Turn a weight and watch it move. 47 of the 80 metric and statistic columns are weighted by no role, and all 80 are captured, so this is about which of them earn a place rather than which have data.

The order, as you change it

Scored on every weight the role holds, attributes included, so this is the total the grid ranks by. Movement is against the shipped weights.

Against all 96 captured players, not against a comparison. The attribute column will match the app exactly — 20 is 20 whoever posted it. The form column will read LOWER than the same player's number on a comparison of eight, and that is the squad-relative rule working (#342): a metric score divides by the best figure anyone in the compared set posted, so widening the set from eight to ninety-six raises the divisor. It is the right set for reviewing a default, which has to be right for the squad rather than for one screen.

Player AttFormTotal±

Do not just add. A role is a short list on purpose — eight columns at 3 say less than four at 5 and 2. Anything you add should displace something or come in low. The ± column on the left is the test: a weight that moves nobody is decoration whatever it costs.

What it weights now

ColumnCategorySpreadImpactWeight

Everything else it could

ColumnCategorySpreadImpactWeight

Spread is how far apart the squad actually sits on that column, measured in the scorer's own 0–1 share rather than in raw units — the gap between the 10th and 90th percentile. A column everyone is level on cannot change an order whatever weight it carries, and eight of them are flatter than 0.10.

Impact is measured, not guessed: it adds the column at weight 3, re-ranks your squad, and counts the rank positions that move. Zero means the column is decoration for this role.

When the two disagree, read it carefully. A column can be flat and still move the order a long way — save percentage on a centre back moves 180 positions on a spread of 0.09. That is not the column separating anybody. It is DILUTION: a near-identical share added at weight 3 pulls every player toward the same number and compresses the gaps that were already there, so whoever was close swaps places. High impact on a flat column is a reason to leave it out, not to put it in, which is why those are hidden by default.

Volume half of a weighted ratio is a rule rather than a list — a column in a category where the role already weights a ratio (a name ending %, R or /A) and this one does not. The roles chose ratios in five categories, so most of the unweighted list is the other half of a pair rather than an oversight. The chip names which ratio it paired with; check a few before trusting it.