Changelog

July 19, 2026

• The board now carries 29 hand-set usage adjustments (target, carry, and attempt shares) layered on top of the model, each with a written reason and a date.
• Sample calls: Brock Bowers priced as the Raiders' number one target, CeeDee Lamb back to alpha volume, and firm starter commitments at quarterback where the battles look settled.
• Every adjustment is versioned and stamped into the published run, so in January the pure model and the human-adjusted board get graded separately, in public.

July 19, 2026

• The public accuracy bars were re-scored under the roster-churn fix: WR ticked up from 85% to 86% against the 3-yr baseline.
• Against the expert consensus the model now wins WR rank ordering by its widest margin yet (0.782 vs 0.767) and newly edges TE.
• The RB concession stands: the crowd still out-orders the model there, now 0.774 vs 0.770, and the methodology page still says so.

July 18, 2026

• We caught the model handing 31% of every team's targets to deep-bench bodies, when real seasons settle around 12%, and that phantom volume was quietly shaving the top guys.
• The cause: summer depth charts list about four times more reserves than the charts the model learned on, and every listed body was collecting a full default share.
• The fix makes however many reserves a chart lists split that slot's historical share instead. Accuracy improves at every position in the four-year backtest, and top receivers gain roughly 15 to 25 projected points.
• Expect the top of the board to reshuffle when the next projection run publishes: elite target shares now look like elite target shares.

July 17, 2026

• Click any player row: the 2025-points-to-2026-value story now draws as a column waterfall instead of a list of numbers, so you can see where the value moves at a glance.
• Hover any column for its drivers: the share, volume, and rate splits behind each step, and what rank repricing means for that player.
• The axis starts at zero with no breaks, so a 2-point step looks exactly as small as it is, and the columns still sum exactly from his 2025 points to his 2026 value.

July 16, 2026

• Why the board projects Josh Allen for 8 rushing touchdowns after seasons of 12 and 14: the shrinkage math shown in full, five worked player examples, and three years of matched receipts.
• Also covers the smarter red-zone prior we built and benchmarked, and why it did not make the cut.

July 12, 2026

• The injury-prone label, measured: what two straight hurt seasons actually predict, why quarterbacks are the exception, and why the board projects health.
• The 2026 touchdown regression file: the players whose 2025 touchdown totals ran ahead of or behind their chances, with a promise to grade the lists in January.

July 11, 2026

• A new Research section publishes the data behind the model's decisions, starting with a study of which stats actually carry over year to year.
• First finding in plain terms: a receiver's target share explains about half of next season's, while his touchdown rate explains under 2 percent.

July 11, 2026

• The public accuracy numbers were re-scored under this week's model updates, and the RB accuracy bar ticked up from 83% to 84%.
• Against the expert consensus the model now scores 99.0% of the consensus points error across 2022 to 2025, a hair ahead over the four seasons.
• One honest concession: the consensus now edges us on TE rank ordering, and the methodology page says so.
• The named misses were refreshed too, including moving 2025 Kyler Murray from the wins column to the honest injury misses.

July 10, 2026

• Starting QBs are now projected at their real share of dropbacks, so Josh Allen throws his true 99% of Buffalo's passes instead of a hidden injury hedge.
• Drafted rookie QBs listed behind a veteran keep a sensible floor instead of projecting near zero.
• Running QBs like Allen, Lamar Jackson, Jalen Hurts, and Jayden Daniels now project at their team's proven rushing volume.
• Fixed a depth-chart quirk that handed fifth-string RBs phantom carries, which quietly shaved every starter's workload.
• Recent seasons now weigh more in usage priors, so Jahmyr Gibbs actually inherits David Montgomery's departed carries.
• Rankings shifted with these changes, and every one was backtested on 2022 to 2025 before shipping.

July 10, 2026

• Click any player row to expand a full detail panel without leaving the board.
• A per-lane value waterfall splits each player's value into Passing, Rushing, and Receiving.
• A value bridge shows how a player's projected points build up piece by piece.

July 9, 2026

• New driver breakdown answers why a player's projection is up or down versus last season.
• Fixed a brief signed-out flash in the landing page header for signed-in visitors.

July 8, 2026

• Player cards now carry headshots and a side-by-side comparison against the player's 2025 season.
• Removed the QB games-played haircut: boards show healthy-pace stat lines, and you price durability yourself.
• Coach chat now renders formatted markdown instead of plain text.