Same breaks as the Bundesliga, far less carry-over

Capping home/away breaks, then evening out carry-over, gives an 18-team season with the Bundesliga's 48 breaks and 38 to 50% less carry-over.

A home-and-away league schedule has to do two things that pull against each other. It should have few breaks: a break is two home games in a row, or two away games in a row. Fans and grounds like a steady home, away, home rhythm. It should also have an even carry-over: no team should keep meeting the same opponent straight after the same other team. Carry-over is the subject of an earlier post.

The textbook circle method gets the fewest breaks there can be and close to the worst carry-over. The search on the league schedule generator went the other way. It put a price on each break and traded breaks for carry-over. Its weakness showed up when we raised that price: a 20-team season never got under 88 breaks. The Bundesliga plays 18 teams with 48 breaks, which is the least a mirrored home-and-away season can have.

Why the price never got there

We measured leg one (the first half of the season) on its own. With 20 teams it had 46 to 52 breaks, and the minimum is 18. With 18 teams it had 28 to 34, and the minimum is 16. Leg two can't undo that, so the season ends up with more than twice as many. The search started from low carry-over and tried to buy breaks back. Each move that removed a break cost more carry-over than the price of the break, so the search turned the move down.

Starting from the other end

The new search starts where the circle method does, at the minimum. It then lowers carry-over, but it refuses any move that takes the season over a set number of breaks, or that leaves a team more than one game off an even home and away split. Each move reshuffles who plays whom in some rounds. It can keep every team's home and away days as they are, which is only allowed if every match still has one home side, or it can move each match's ground with the match.

For leg two it tries three orders and keeps the best one that fits under the cap:

  • leg one backwards, which spreads carry-over well but adds a break for almost every team at halfway;
  • leg one again with the grounds swapped (the mirror), which has the fewest breaks;
  • the mirror with its first and last rounds swapped.

What it found

Three seeds per setting, every figure recounted from the fixtures it produced (eval/capped-breaks.txt in the repository):

TeamsBreaksCarry-overReal fixture lists for comparison
18482,154 to 2,674Bundesliga 2021-22 to 2024-25: 3,474 to 4,342 at 48
18601,696 to 1,858Ligue 1 2023-24: 1,840 at 74
18721,564 to 1,704Ligue 1 2024-25: 1,948 at 96
20544,734 to 5,080circle method 21,044 at 54; no top-flight season under 70

At 18 teams, a schedule with the same 48 breaks as the Bundesliga has 38 to 50% less carry-over than any of the four Bundesliga seasons we checked. Those schedules also carry derby, police and TV constraints that we don't model, so this doesn't show that the league's planners could have done it. It shows how much room the break minimum leaves.

The 20-team result is weaker, and we'd rather say so. With 20 teams the capped search barely improves leg one. It ends near 1,100 to 1,400 carry-over at 18 to 22 breaks, where the unconstrained search reaches about 500. Above about 60 breaks the older weighted search is still better. We also tried starting from the weighted result and lowering the cap step by step. It never got away from where it started, so we didn't ship it.

On the page

When you tick "Spread who you meet next", the page now runs ten searches: the eight it ran before, unchanged, and two capped ones. There is also a fourth choice for which one to keep: the fewest back-to-back home or away games possible, then the most even carry-over. Organisers who care most about rhythm get the minimum breaks without the circle method's carry-over. The other three choices almost never change. Across six test runs they changed once, at 18 teams, from 1,550 / 72 to 1,786 / 60. Leagues with an odd number of teams get two more weighted searches instead, because a bye already ends a run of home or away games and the capped search can't move one without unbalancing someone.

All of this runs in the browser and nothing is uploaded. The code is MIT licensed in ventures/league-schedule-generator/carryover.mjs. These are search results with made-up team names: no league has played one of these schedules yet.

All writing · More in Tools and projects · More on League Schedule Generator · All tools and research