The public research that grounds how Milly Metrics scores a book — roster construction, stacking, and portfolio exposure. Each entry is one paragraph of the specific findings, tagged with which score it feeds: Build, Upside, or Exposure.
Aggregates all five Best Ball Mania seasons to find the position counts that reliably advance: early WR capital, RBs backfilled a little later, and 2–3 TEs total — never the double-early “bully TE.” QB1 should land before Round 14, and reaching well above ADP repeatedly costs points every year. This is the backbone of the Build score’s round windows and shape ladder.
A re-analysis of advance rates by build: two QBs beat three (first QB best in Rounds 6–10), 4–6 RBs total but rarely 3+ early, and 7–10 WRs with 3–4 taken inside the first five rounds. Confirms the early-WR-capital, late-RB, and QB-timing rules the Build score rewards.
A large injury-and-pricing study (79k player-games) showing the RB/WR risk calculus has flipped — RB injury rates hit a 17-year low while WR hit a high, making early RB safer than it used to be. The strongest signal is fragility as a persistent trait: a player’s injury history predicts risk more than workload or age, and receiving backs are safer than carry-heavy ones. Feeds the RB-lean shapes and the exposure caution against over-stacking injury-flagged players.
Lays out the DraftKings build archetypes — Robust RB, Hero RB, Zero RB, Elite QB/TE and more — and grades each against the 8.3% baseline win rate. No single build dominates; each wins when its premises hold. This is the DK archetype set behind the DraftKings shape ladder.
Confirms DraftKings’ format — 20-man rosters, full PPR with passing and yardage bonuses — and how it shifts allocation. Its key rule: take two QBs if you land an elite one early, three if you wait, since DK’s rushing bonuses make dual-threat QBs worth more. Sets the roster and scoring parameters the DK model has to match.
A pick-frequency study of the DK Millionaire Maker field: every one of the ten most common builds carries 5–6 RBs. It’s field frequency, not advance rate — what the field does, not what wins — but it validates 5–6 RBs as the DraftKings norm the shape ladder is built around.
The foundational simulation study: two QBs beat three and each should be stacked hard, spending less capital on RB correlated with better advance, and finals rosters want 6–9 game-stacked players (game-stacking lifts finals win probability ~50%). Directly informs the Upside score’s correlation weighting and the stacking targets surfaced in your book.
A finals-simulation study where correlation dominates: 3 stacked QBs reached the finals 23% more than random, the target is 6–10 game-stacked players, and carrying 14+ “live” players into the playoffs roughly doubled finals EV. Top-10% closing-line-value teams also made finals at nearly double the base rate. Feeds the Upside correlation model, the Exposure score’s CLV component, and the live-player logic.
Five principles for winning best ball, headlined by stacking beyond just QB+WR (any correlated teammates), playing for upside over depth, and being contrarian on low-owned players in tournaments. Backs the Upside score’s broadened same-team correlation and the contrarian read in the Exposure score.
A re-run of the Manifesto on newer data: raising your regular-season advance rate lifts both floor and ceiling, so early-week edge compounds, while 6–9 game-stacked players stays the finals sweet spot. Closing-line ADP value and more live players again drove big EV gains, and on build, early WR capital stayed king with reduced-RB shapes (full Zero RB, Hero RB) still advancing best. Feeds the Upside correlation weighting, the Build shape logic, and the Exposure CLV signal.
A portfolio-management playbook that reads like a spec for the Exposure score: track exposure by both team share and fee share, hunt for stack blind spots, use CLV to trim decliners and load risers, and diversify your Week 17 pairings. This is the direct model for the exposure board, the stacking pop-out, and the CLV signal.
A worked example of reading a real 50-draft portfolio, flagging 12 players owned in 20%+ of teams — more than double the neutral baseline — as deliberate concentrations. It models how to separate intentional edges from accidental over-exposure. Grounds the Exposure score’s over/under-vs-baseline reads and the exposure board.
These studies are compiled into one body of evidence, not followed one at a time: roster-construction findings set the shapes the Build score grades, the stacking work drives the Upside score’s ceiling weighting, and the portfolio research shapes how the Exposure score reads your book against the field. Where sources disagree, the model leans on the most recent, largest-sample evidence and is tuned per platform.