Power

as_of: 2026-09-17

status: DRAFT for Methods hub

seat: War Room Power

product_rule: Cards stay short. Depth lives here.

One sentence

Blind quantitative seat: turn efficiency, pace, player value, and fact IR into a sealed fair margin and total — before any shop price — then stop.

Inputs (what we weigh)

LayerExamplesLabel
Team strengthEPA / success / explosives (nflverse), prior-season shrink + current-week updatefact → model
PaceOffensive plays / gamefact
Player valueQB/skill EPA, snap shares; starter volume flagsfact / light model_inference
SituationRest differentials, roof/HFAfact
WeatherOutdoor NWS precip/wind/heat (totals mainly)report
InjuriesOfficial Game Status Out/Doubtful; practice DNP/LP as haircuts onlyfact (VERIFY when roster tags weird)

Never pre-seal: our shop, consensus lines, Covers/SportsLine/FN picks, forums, Roundtable vibes.

Model (v0.1 → calculus roadmap)

Now (honest): blended prior (heavy 2025) + thin 2026 week sample → expected home margin + total; Normal Monte Carlo for win/cover grids and total refs; scenario haircuts from fact Outs.

Next (calculus, not arithmetic):

  1. Opponent-adjusted EPA / success (not raw team averages)
  2. Joint spread–total sims (correlation, not two independent normals forever)
  3. Multi-week Bayesian shrink with Scorekeeper calibration feedback
  4. Explicit OL/QB role stacks when wait_core is fact Out
  5. Post-seal join only for P(cover)|our shop — never baked blind

What “lean” means

Kill criteria (we ABSTAIN / stay quiet)

What we will never claim

How to read us on the site

Chip: Model lean · TEAM or No lean · WAIT

Line: spelled side (ATL −3½ fair · tot ~44)

Bullets: why · not an Official

Open this Methods page when you want the treasure — not the dossier card.

Methods v1 — filed by the seat. Public site hides private shop brand names.