GAA.est. 2015

About

Data

Current-season data is pulled from ESPN’s fantasy API with fflr, an R package that wraps the undocumented v3 endpoints. League 252353 is public, so nothing here needs a login or a cookie.

The data refreshes on its own. During Sunday game windows the scoreboard updates about once a minute; standings and season charts refresh hourly; nothing is fetched overnight. If a fetch fails, the last good numbers stay on the page rather than being replaced by an error.

Archive

ESPN no longer serves this league’s early seasons. Requests for 2015 through 2017 return 404, and 2018 returns 401 — with or without credentials. Everything before 2019 on this site comes from a local copy captured while it was still available.

That copy stores each game as it was actually played, which matters for one season in particular.

2022

Before the 2023 season ESPN changed kicker scoring to per-yard and recomputed 2022 under the new rules. That flipped exactly one game: week 1, where KIER beat ANGS by 0.02 points as played, but loses by 0.08 under the revision.

One game, and it decided the season. ESPN seeded the playoff bracket from its own revised standings, which put CART in the field at the 4 seed and left KIER out. Under the scoring actually in force in 2022, that seed belongs to KIER — same 15 wins as CART, 32 more points for.

Replaying the bracket with every team’s real postseason totals: KIER beats BILL 200.14–195.38 in the semifinal, CHAR beats ANGS as actually played, and KIER beats CHAR 203.26–146.60 in the final. So this site records KIER as the 2022 champion.

That is an inference, and worth being plain about. What supports it: the seeding rule used here — head-to-head plus median wins, ties broken on points for — reproduces the bracket ESPN actually ran when it is fed ESPN’s numbers. Fed the as-played numbers, the same rule seeds KIER. Only one game in the reconstruction was never played, KIER against BILL; every other score is real, and both are two-week postseason totals set before anyone knew the matchup.

Playoff weeks

Playoff matchups span two NFL weeks and combine their points. Anything about scoring on this site is grouped by NFL week; anything about wins, records or the bracket is grouped by matchup. Mixing the two would make a playoff game look like two ordinary ones.

Expected wins and the median

Expected wins (xW) rank your score against the whole league that week rather than against one opponent, so a 0.9 means you would have beaten nine of ten teams. Comparing xW to real wins is a decent measure of schedule luck.

The median win is a second win each week for scoring above the league median. ESPN added it in 2022, so it is shown for those seasons only — and it is computed per week. During live games the median is based on projected scores and can move until the last game ends.

Terms

Regular season only unless a page says otherwise, including the current season through its last completed week. Lineup numbers start in 2018; ESPN kept no benches before that.

  • PF+ — points per game as a percentage of that season’s league average; 100 is average. σ is standard deviations from the same average. Both compare to the season, because league size (8 or 10) and scoring rules drift.
  • Optimal — the best legal lineup from the roster that week, in hindsight. Efficiency is points scored as a share of it; Coach+ is efficiency against that season’s league average.
  • vs Auto — your lineup against the one ESPN’s projections would have started, both scored on what happened. Overrode is how often you went your own way, Right how often that paid off.
  • Lineup loss — a game lost that the optimal lineup would have won.
  • FLEX (Positions) — the RB, WR or TE ESPN’s projections would have put there, bench included. Other positions count started players by position, not slot.
  • Schedule luck — actual wins minus the average over every team’s schedule that season. Different from Luck on Standings, which is wins minus xW. Points against splits into who you drew and when: the draw has been worth at most about three points a game in any season, the timing several times that.
  • 2022 lineup numbers use ESPN’s re-scored kicker points, the only version it still has.

Trades

The trade page is the one thing here worked out when you ask rather than ahead of time. It runs fflr’s evaluate_trade() against the league as it stands: for each week, it builds both teams’ best legal lineup from ESPN’s own projections under this league’s roster rules, with and without the trade, and reports the difference. That makes it a measure of projected starting points, not of player value in the abstract — a great backup who would never start for you is worth nothing to you here, which is the point.

By default a trade also has to beat the waiver wire: every lineup, before and after, may start the best available player at each position that week. Without that, a team whose only quarterback is on bye scores zero at the position, and any trade for a quarterback looks like a steal — when the team would really just pick one up. The baseline can be made stricter (the second- or third-best free agent) or switched off. Hindsight replays a trade over the weeks already played with actual points; it scores rosters alone, since in hindsight the “best available” player is simply whoever got lucky.

Last place

Last place — “The Griz” — isn’t listed. The constitution defines it through the consolation ladder, and recovering final placement from the stored bracket is guesswork, so it has been left out rather than guessed at.

Stack

R for the data, Observable Plot for the charts, and a static page served from a Raspberry Pi — plus one small R server for trades. No trackers, no cookies, no CDN — fonts and libraries are served from this domain.