Team Strength Lab

How far the standings actually spread out, season by season, versus the spread a coin-flip league (every team exactly .500, every game a 50/50) would produce purely from playing a finite 82-game schedule. That baseline noise is real — a perfectly equal league still wouldn't finish everyone at 41-41 — so it's subtracted out before calling anything "excess" spread. Lower = more compressed / competitive. Shown two ways, since "subtract out the noise" isn't a one-line-answer question — see below.

How far apart teams' actual scoring margins are (SRS = margin of victory, adjusted for schedule strength). This uses full margin instead of collapsing every game to a win/loss, so it's a lower-noise read on talent gaps directly — no luck-floor subtraction here, since there's no simple closed-form noise model for continuous point differential the way there is for a binary win/loss. Lower = more compressed talent distribution.

Two ways to combine "observed spread" and "luck floor" into one excess number, shown side by side because they disagree on how much: ● naive subtracts the standard deviations directly (observed − floor) — simple, but SDs don't subtract cleanly across independent noise sources. ● variance-based subtracts the variances first, the statistically standard way to remove one independent noise source from another (√(observed² − floor²)). Variance-based is always the larger number here (i.e. reads as less equitable) — subtracting SDs directly over-corrects for the luck floor. Naive is the more conservative/optimistic number; variance-based is the more defensible one.

Search for up to 8 teams to overlay their SRS across every season we've got.

Estimated directly from the schedule each season (mean home minus away score, over games with a known home team) — not a fixed constant.