Checking for this week's published slate…
Average Points is the expected score. Correct for 50/50s and double-ups.
Upload one file per position. The chosen window supplies BOTH the mean and the sigma — there is no separate sigma window, because mu and sigma must come from the same games. This REPLACES the projection before lineups are built, so it changes which lineups exist.
Drives Sigma, Floor and Ceiling. These are ASSUMPTIONS, not measurements — replace them once you have game logs. Higher = more volatile. Sigma = mean × CV.
Tick any combination. All three is the DraftKings default; untick to force the FLEX into a narrower set.
IR and OUT are always removed.
Pruning keeps the cheap efficient plays, so lineups lean on minimum-salary punts. Top N keeps only the best N per position, removing those punts and forcing the last slots up a price tier. Pruning is lossless for the single best lineup, but NOT necessarily for a portfolio under exposure limits — use Tuning to compare them on your slate.
Removes players who are provably useless (someone cheaper AND higher scoring). Cannot change the best lineup, only speed.
Proxy for "is he actually starting". 0 = no floor. Locked players ignore floors.
Talent overlay. Adjusts the selected mean up OR down before pruning. Measured player sigma stays fixed; a positional-CV fallback tracks the adjusted mean so its configured volatility ratio stays fixed. Influence QB 0.20 RB 0.30 WR 0.30 TE 0.25; X-Factor adds 0.06×(index/100)2 of the player's own mean. Coefficients are PHASE 1 assumptions fitted in-sample.
Drops anyone projected below this. Unlike a salary floor it targets low producers directly, whatever they cost. 0 = off.
Overlap = most players two lineups may share. Exposure = ceiling on how often one player appears. Set overlap 9 and exposure 100 to turn both off. If the full count cannot be found, the largest set is shown with an exact warning; DraftKings export stays disabled until the set is complete and compliant.
Maximum 100 lineups per generated set.
none
none
Set these in the Player pool tab.
Portfolio is normally the right choice. Full Pool may contain tens of thousands of candidates and is intentionally much slower.
EXPERIMENTAL. p controls the whole shape: p→1 is spikier with more zeros, p→2 becomes gamma. Its zero mass is also constant within a position until real usage data replaces the CV.
Uses whatever games a player actually has. Under 2 games falls back to position and says so. No blending.
MEASURES the lineups you already have.
SAMPLES one lucky Sunday. Deliberately random — re-roll for a different draw. Not a projection, and that is the point.
Run the optimizer first.
Inspection tools. None of these pick lineups for you.
A single draw with per-player scores. At N=1 the ranking is pure noise — do NOT select lineups from it. It exists so you can see what one Sunday looks like and confirm the distributions are sane.
Same lineups, same sigma, same seed — only the shape changes.
Re-solves one lineup with PARETO and the exact ILP solver under the same player-pool and FLEX rules. ILP is a diagnostic proof, not a weekly portfolio generator.
Reruns the whole model at every cap and pools the results. The best lineup at $49,600 uses different players than at $50,000, so the sweep reaches rosters a single cap never can.
Maximum 100 lineups at each salary cap.
Always runs BOTH pruning and Top N — comparing them is the point, so there is no mode selector here.
FLEX can hold an RB, a WR or a TE. This tries every combination of which are allowed — RB+WR+TE, RB+WR, RB+TE, WR+TE, RB only, WR only, TE only — so you can see what each restriction costs before choosing one. It multiplies the work by 7.
The other positions hold at their Optimizer values. Pruning is solved ONCE as the baseline — it does not depend on the counts.
On, each pool AND the merged union pass through portfolio selection, so results respect your limits and are safe to export.