Fantasy football decision assistant · season 2026
XOCP answers the four questions that actually get asked during a season — who do I start, what do I bid, what do I change before kickoff, and who am I drafting — with every number computed under your league’s real scoring rules.
The thesis
Public rankings assume generic PPR. XOCP re-scores every projection through your league’s own settings — the real touchdown values, the real interception penalty, the real flex shape.
Proven, not asserted. Verified against Yahoo’s own reported points at a maximum delta of 0.000.
A favorite wants floor; an underdog wants ceiling. Optimizing total projected points gets that backwards. This was the centrepiece.
Cut on evidence, before a line of it was written. Across 4,400 simulated rosters the two objectives disagreed 0 times out of 400, and the largest gain it could ever produce landed inside the band the screen already labels as noise.
Why that cost a feature and not a phase
The lineup screen was built to work with no probability in it, and a gate pins that independence.
Written as the mitigation on 2026-08-24. Tested rather than intended. It held on 2026-08-27.
Measured, and written down
0.000
max scoring delta
v1 — recomputed against Yahoo’s own reported points for 50 players across three 2025 weeks, kickers and defenses included.
62,201
identity crosswalk rows
v1 — Yahoo, Sleeper, nflverse and ESPN ids resolved to one internal id. Exactly one manual override; name matching fails the build.
52
correlations, none significant
v1.5 — the milestone measured its own headline feature and cancelled it four days before it would have shipped.
0 / 400
win-probability disagreements
v1.6 — the centrepiece layer was cut on its own pre-registered trigger, before a line of it was written.
What exists today
Every projection is re-scored through your actual Yahoo settings. Josh Allen is 405.5 here and 361.5 under the defaults — that 44-point gap is the whole product, and it is on screen rather than in a footnote.
proven
Ranked, tiered and cross-positional, precomputed into a committed snapshot with zero live dependencies at render time. A paused database cannot take the board away on draft night.
shipped
Opponent, bye, home / away / neutral, and each opponent’s 2025 points allowed at that player’s position — scored through the same oracle, never a generic figure. Surfaced as information, deliberately not as a multiplier.
shipped
Not a ranked player list. Each slot carries what you would start instead and what the swap is worth — and the rows worth less than the noise floor say so rather than dressing a coin flip as a decision.
shipped
All 32 teams by carries plus targets rather than by published rank, with the 2025 room named wherever a player earned his number somewhere else.
shipped
The modal coverage shell as field geometry, placement stated as nominal, and eleven defenses that have no 2025 charted prior getting a real screen instead of a spinner. Reference only — a gate keeps it out of every ranking.
shipped
The signature element
Not a percentage bar. 50% is the ball on the 50; 68% is the ball on the opponent’s 32. A lineup change that moves you four points is shown as a four-yard gain, and deltas become legible in a way percentages never are.
Designed in v1, deferred because it renders a quantity that did not exist — and still waiting, because the probability layer behind it was measured and cut. It ships when there is an honest number to put in it.
What comes next, and what each item is waiting on
An irreversibility decision over the players who lock early — and it is not called “Thursday”, because Thursday is the wrong day in five weeks of eighteen. The league’s lock regime is read, never assumed.
next
What a waiver claim is worth against the specific player he displaces, emitted as a walk-away number. Blocked on Yahoo publishing the budget, the bid unit and whether $0 is legal — all three are null and refused a default.
blocked
Target share, air-yards share, WOPR and RACR, then touchdown-rate regression candidates in both directions. The weekly stat ingest that blocked all of them has now shipped.
unblocked
50% is the ball on the 50. A lineup change worth four points is four yards, not “+4.2%”. Designed, deferred, and waiting on a probability model honest enough to put in it — the first one was measured and cut.
designed
The part that generalizes
Roughly twenty measurements in this project were wrong in a way that looked like a finding. Every one was caught because a number contradicted something previously written down — never because anyone read the code and noticed.