How the board is built, and how the pick line is chosen.

1. Projection

Points per 90 minutes across last season, pulled toward the position's average by a prior worth 900 minutes of evidence, then multiplied by projected minutes and an availability factor. A player with 200 good minutes has not proved what a player with 2,000 has, so the shrinkage matters most exactly where the evidence is thinnest.

This reads Premier League history, not your league's history, which is why it works for a first-ever draft with nobody's picks recorded yet.

2. Value over replacement

VORP is the projection minus what the replacement player at the same position is projected to score, and replacement is the first player past the starters in a league your size. A thin position's players climb the board on their own: a defender's 140 points and a striker's 140 points are not the same asset, and this is the number that says so.

3. Tiers

A grade inside each position, S down to F. A player under 270 minutes of Premier League evidence gets ? instead of a grade, because there is nothing there to grade, and ? is treated as a real answer rather than filled with a guess. A player who has left the league carries X, stays visible on the board for honesty, and is excluded from every pool the tool picks from.

4. The pick line

The strategy that ships is best value at a position you still need: the biggest VORP among the positions your squad shape has not filled. It was chosen by a tournament, not by argument.

160 paired snake drafts (20 seeds across 8 draft slots), each strategy taking every slot in turn, scored on the final squad's best legal starting XI. Needs-aware best value scored 1505.9 mean best-XI projected points. Two more elaborate formulas we built on purpose to try to beat it lost: one-step regret of waiting scored 1435.4, and a full-rollout planner that played out the rest of every draft scored 1464.8. Both lost in every one of the 8 slots. A 40-seed rerun agreed: 1436.6 and 1463.9 against 1508.3.

The full story of the two challengers, including why each one lost, is on what we got wrong.

5. Where it is weakest

• The tournament's rivals are simulated strategies, not humans. Seven rivals rotate through the four baseline strategies above; a real room of people does not necessarily draft like any of them.
• Survive odds are a labelled heuristic, not a measured probability. The chance a player lasts to your next pick is clamped between 2 percent and 98 percent and built from list order, not from real drafts.
• The board's finishing-luck flag is a flag you read, never an input to the ranking. It marks a player whose goals ran ahead of, or behind, the quality of his chances; it never changes his projection or his tier.
• A two-season prior for this board's own points-per-90 was tested and killed. Every blend of two seasons beat the single-season baseline on the point estimate, but no confidence interval cleared the pre-registered bar. The board's prior stays single-season. Full numbers on what we got wrong.

How to cite this page

DraftScout.ai, "How it works", draftscout.ai/how-it-works, last changed 2026-08-02.