Research notes
The decisions behind the projections.
Every stance the model takes was earned against ten years of data, and these notes show the work: what we measured, what it said, and what it cost us when the answer was inconvenient. Published occasionally, in plain language. The compact version of the whole system lives on the methodology page.
July 16, 2026 · 6 min read
Why we project Josh Allen for 8 rushing touchdowns, not 14
He scored 12 and 14 the last two years, and our board still says 8. The exact math behind the number, the proof that touchdown rates barely repeat, and three years of receipts showing the group lands where the shrink says.
Read the note →
July 12, 2026 · 7 min read
The 2026 touchdown regression file
Expected touchdowns predict next season better than actual ones. The 2025 players who scored beyond their chances, the ones who got robbed, and a promise to grade both lists in January.
Read the note →
July 12, 2026 · 6 min read
The injury-prone label, measured
Games played is the least predictable stat we track. The label is worth about three games at its absolute strongest, quarterbacks are the one real exception, and that is why the board projects health.
Read the note →
July 11, 2026 · 7 min read
Sticky stats: what actually carries over from one season to the next
We correlated ten years of back-to-back player seasons, stat by stat. Roles repeat, efficiency barely does, and touchdown rates are closer to a coin flip than a skill.
Read the note →