Earlier today the league schedule generator learned to even out the carry-over effect: how often teams meet the same opponent straight after that opponent has played the same other side (the first post). For a home-and-away season it did what most people do by hand: write the first half, then play it again with the venues swapped.
That shortcut has a cost. Every "A, then C" step in the first half happens again in the second, so every carry-over count doubles, and Russell's carry-over value (the sum of the squared counts) roughly quadruples for that half.
What changed
Ticking "Spread who you meet next" on a double round robin now searches the second half as a round robin of its own. It starts from the first half played backwards, which turns every "A, then C" into "C, then A", so the counts spread out instead of doubling, and each team's home/away pattern is its first-half pattern reversed and flipped. Then a seeded search swaps rounds and partial rounds inside the second half, scoring carry-over plus the exact number of back-to-back home or away games. The rules every league expects still hold: each pair meets once in each half, the return game is at the other ground (so everyone has the same number of home games), and nobody meets the same team or sits out twice running across the halfway point.
The numbers
Seed 1, on the same comparison as the first post (published fixture lists from openfootball, CC0, 2021-22 to 2024-25):
| Schedule | Teams | Carry-over | Home/away breaks |
|---|---|---|---|
| Premier League, La Liga, Serie A, Ligue 1 (14 seasons) | 20 | 2,278 to 2,564 | 70 to 146 |
| Tool, first half mirrored | 20 | 2,136 | 114 |
| Tool, whole season searched | 20 | 1,678 | 124 |
| Bundesliga (4 seasons) | 18 | 3,474 to 4,342 | 48 |
| Tool, first half mirrored | 18 | 1,926 | 74 |
| Tool, whole season searched | 18 | 1,498 | 88 |
The least possible value for 20 teams is 1,460, so the search gets within 15% of the floor. It pays about ten more back-to-back home or away games for that, still inside what the professional leagues play. Those leagues also have derbies, police requests and television slots to fit, which the tool does not.
The code (carryover.mjs, lowCarryOverDoubleRounds) and the comparison (eval/carryover-real.mjs) are MIT licensed in the repository. Breaks are the next thing to work on: the search only moves the second half when it counts them, and moving both might get the extra ten back.
Estimates from code and public data; no warranty.