Docs

All docs

Concepts

Guides

All-play record

  • Concept
  • Engine v0.18.0
  • Facts schema 0.5.0 and later
  • commishdesk.com/docs/all-play/

Definition

All-play asks how a team would have done if it had played everyone every week. For each week from 1 through the last regular-season week, every team that played is compared by points with every other team that played that week. The higher score wins that comparison, the lower score loses, and an exact tie is recorded as a tie. The results are added across the weeks into w, l and t.

pct is the decisive win rate, w / (w + l), rounded to three decimals. Ties are left out of the ratio. It is 0.0 before any decisive comparison exists.

All-play is the base for expected-wins-and-luck, and it is the largest input to power-rank-and-nudge.

See All-play record in a full league

Supported: Sleeper · See supported apps

See a full league while we open sign-ups.

Worked example

Fixture expected-weekly-facts-week10.json

Fixture excerpt
Rowwltpct
Flea Flicker Union882200.8
Two Minute Drill872210.798

Fixture: tests/fixtures/facts/expected-weekly-facts-week10.json (the anonymized league "Trench Warfare").

{
  "teams": [
    {
      "roster_id": "1",
      "team_name": "Flea Flicker Union",
      "season": {
        "all_play": {
          "w": 88,
          "l": 22,
          "t": 0,
          "pct": 0.8
        }
      }
    },
    {
      "roster_id": "2",
      "team_name": "Two Minute Drill",
      "season": {
        "all_play": {
          "w": 87,
          "l": 22,
          "t": 1,
          "pct": 0.798
        }
      }
    }
  ]
}

There are 12 teams, so each team makes 11 comparisons a week. Over ten weeks that is 110. Flea Flicker Union (roster 1) won 88 of them and lost 22, so pct is 88 / 110 = 0.800. Two Minute Drill (roster 2) went 87-22-1, one comparison being an exact tie, so its 109 decisive comparisons give 87 / 109 = 0.798. The tie raises neither w nor l, but it is still one of that team's 110 comparisons. Roster 2 is 9-1 in real games, while roster 1 is 7-3. Their all-play records are almost equal, which is why expected-wins-and-luck treats roster 1 as the less fortunate team.

See All-play record in a full league

Put in your Sleeper username or league ID, or look around a fictional league first.

Supported: Sleeper · See supported apps

See a full league while we open sign-ups.

Code

Show the code commishdesk/stats/weekly.py::_all_play_totals

Source: commishdesk/stats/weekly.py::_all_play_totals

def _all_play_totals(
    matchups_by_week: dict[int, dict[str, Matchup]], cutoff: int
) -> tuple[dict[str, int], dict[str, int], dict[str, int], dict[str, float]]:
    """Fold weeks ``1..cutoff`` into cumulative all-play wins/losses/ties plus
    the summed per-week win-equivalent fraction (``expected_wins`` before
    rounding), keyed by roster id. A roster with no counterpart in
    :attr:`WeekModel.rosters` (an orphan matchup row) still participates --
    it still affects every other roster's count -- it simply has no
    :class:`TeamWeekStats` entry to be reported on later."""
    wins: dict[str, int] = {}
    losses: dict[str, int] = {}
    ties: dict[str, int] = {}
    expected: dict[str, float] = {}

    for wk in range(1, cutoff + 1):
        rows = matchups_by_week.get(wk, {})
        participants = sorted(rows, key=_sort_key)
        decisions = len(participants) - 1
        if decisions < 1:
            continue
        week_win_equiv: dict[str, float] = dict.fromkeys(participants, 0.0)
        for i, roster_a in enumerate(participants):
            points_a = rows[roster_a].points
            for roster_b in participants[i + 1 :]:
                points_b = rows[roster_b].points
                if points_a > points_b:
                    wins[roster_a] = wins.get(roster_a, 0) + 1
                    losses[roster_b] = losses.get(roster_b, 0) + 1
                    week_win_equiv[roster_a] += 1.0
                elif points_b > points_a:
                    wins[roster_b] = wins.get(roster_b, 0) + 1
                    losses[roster_a] = losses.get(roster_a, 0) + 1
                    week_win_equiv[roster_b] += 1.0
                else:
                    ties[roster_a] = ties.get(roster_a, 0) + 1
                    ties[roster_b] = ties.get(roster_b, 0) + 1
                    week_win_equiv[roster_a] += 0.5
                    week_win_equiv[roster_b] += 0.5
        for roster_id in participants:
            expected[roster_id] = expected.get(roster_id, 0.0) + week_win_equiv[roster_id] / decisions

    return wins, losses, ties, expected

View this file on GitHub at v0.18.0

Edge cases

  • Week 1. A single week has no average week to compare against, so all_play is null for every team until week 2.
  • Playoff weeks. The fold stops at the week before the playoffs start. From then on all_play stays at its final regular-season value.
  • Byes. The comparisons per week are the number of teams that played minus one. A team on a bye is neither compared nor counted, and the other teams' totals reflect the smaller field that week.
  • Orphan rows. A matchup row for a roster that is not in the roster list still counts as an opponent for everyone else. It simply has no all-play row of its own.
  • Exact ties. Both teams get a tie. For expected_wins the tie is worth half a win to each.

See All-play record in a full league

Put in your Sleeper username or league ID, or look around a fictional league first.

Supported: Sleeper · See supported apps

See a full league while we open sign-ups.

On this page