The league had two methodologies: the original Google Sheet model (we’ll call it v1) and a new z-score model (v2). Both are computed from every season’s real data; you can toggle between them anywhere on the site.
The v2 z-score model in plain English
- Magnitude preservation. A z-score says how many standard deviations above or below the league mean a manager landed that season. +1σ on win % means exactly the same thing as +1σ on points, regardless of how spread out the league was.
- Comparability across metrics.Final standing, win %, share of points, and all-play win % have wildly different scales. Z-scoring puts them on the same axis so the weighted average is meaningful.
- Era-comparability. Yahoo era (2019–2022) had no weekly score data, so it uses 3 components renormalized to 1/3 each. The Sleeper era (2023+) adds the all-play component and uses 4 × 0.25. Both produce comparable career numbers because z-scoring strips out scale differences.
- Win % excludes the median game.2024 added the median bonus; counting it would silently inflate Sleeper-era win % relative to Yahoo. The v2 model uses head-to-head wins only, so the same season has the same meaning across eras.
- How to read the score.We take each manager’s z-score across the components, weight them, then rescale so the league average is 100 (score = 100 + 20·raw). A 120 is roughly a full standard deviation above the league average — elite territory. An 80is about a standard deviation below — a rebuild year. Think of it like baseball’s WRC+ or OPS+: 100 is exactly average, and every 20 points is one standard deviation.
- Low-confidence flag.A manager with fewer than 3 seasons of data has a career mean that’s very sensitive to a single outlier season, so those rows carry a small low sampletag on the leaderboard. An in-progress season counts fractionally toward the 3-season bar (at its weeks-played weight), so a 2-season manager isn’t mistaken for a steady performer just because a new year kicked off.
- Live in-season. During the NFL season the current year joins the model as a season in progress: its z-scores are computed from the games played so far (record, points share, all-play, and a provisional standing by record then points-for). It counts proportionally in the career average — through week N it carries N⁄14 of a season’s weight, so one hot September week can’t outweigh a full past season, and its influence grows every week until it’s a full season by week 14. Rankings and deltas update with each weekly sync and converge to their final values once the season is finalized.
The v1 spreadsheet model in plain English
The original model scores three things, then adds a loyalty/trophy bonus. It’s rank-based: within each category every manager is ordered against the whole field, and that rank — not the raw number — is converted to a 0–10 score, so each category is graded on a curve.
- Three weighted categories. Final standing is worth 60% (×0.6), share of points 30% (×0.3), and head-to-head win % just 10% (×0.1). The bias is deliberate — how you actually finished counts far more than a lucky or unlucky regular-season record.
- Value-table lookup.A manager’s rank in a category becomes a score of (N + 1 − rank) / N × 10 across the N-manager pool — 1st ≈ 10, last ≈ 0. The three category scores are blended by the weights above into a “weighted power” number out of 10.
- Legacy bonus. Added on top to reward longevity and hardware: +0.2 per season played, +0.2 per playoff appearance, and +0.6 per championship. This is why a long-tenured manager can edge out a flashier short-timer.
- Total score. Weighted power + legacy bonus — the single number behind the all-time v1 ranking.
- Live in-season. Like v2, v1 doesn’t wait for the year to finish: once weekly syncs land, the current season joins the model provisionally. Its wins and losses accrue in career win % game by game, while its share of points and provisional standing (by record, then points scored) enter the per-season averages at N⁄14 of a season’s weight through week N — the same proportional weighting v2 uses. The +0.2-per-season legacy credit accrues fractionally too; playoff and championship credit are only added once the season is finalized. Scores update with each weekly sync and converge to their final values when real standings land.
v1 ↔ v2 side-by-side
Same managers, same seasons, two different formulas. Δ is how far a manager moves when you switch from v1 to v2 (positive = climbs in v2).
Worked example: Kevin Hulsebosch (v2 rank 1 · 118.4)
Here’s how Kevin Hulsebosch’s 118.4is built, top to bottom. Each row is one season. The σ columns are that season’s per-component z-scores (higher = better); Weighted blends them by the era weights (¼ each in the Sleeper era, ⅓ each in the Yahoo era). Then two steps turn the seasons into the leaderboard number:
- Averagethe Weighted column across all seasons played (a season in progress counts at its weeks-played weight, N⁄14) → the career z-score.
- Rescale it: score = 100 + 20 × (career z-score).
| Year | Era | Final z | Win% z | Share z | All-play z | Weighted |
|---|
| 2019 | yahoo | 0.87 | 1.86 | 1.44 | — | 1.39 |
| 2020 | yahoo | -0.17 | 0.35 | 1.81 | — | 0.66 |
| 2021 | yahoo | 1.22 | 0.42 | 0.34 | — | 0.66 |
The v1 formula reproduces the original spreadsheet within ±0.01. Both models are recomputed from the live database every week — and both count a season in progress at its weeks-played weight — so the toggle stays apples-to-apples all year round.
| 2022 | yahoo | -0.53 | -0.37 | -0.62 | — | -0.51 |
| 2023 | sleeper | 1.57 | 1.79 | 1.24 | 1.66 | 1.56 |
| 2024 | sleeper | 1.22 | 1.43 | 1.23 | 1.42 | 1.33 |
| 2025 | sleeper | 1.22 | 1.05 | 1.59 | 1.54 | 1.35 |
| Career z-score = average of the 7 Weighted values | 0.92 |
| Power score = 100 + 20 × 0.92 | 118.4 |
So Kevin Hulsebosch’s 7 seasons average to a career z-score of 0.92, which rescales to 118.4 — the exact number at the top of the v2 leaderboard. Hover any σ cell for the percentile-rank intuition.