Carry-over is about who you play straight after whom. If the Lions are the strongest side, and the Hawks keep meeting teams the week after those teams played the Lions, the Hawks get tired opponents again and again. Russell's carry-over value (1980) squares and adds up how often each "A, then B" sequence happens across the league. Lower is fairer, and the fairest a schedule can be is every count at 0 or 1.
The league schedule generator has an option, "Spread who you meet next", that searches for a low value. It starts from the textbook circle method and anneals. For most sizes that works well. For 16 teams it was stuck at 316, against a floor of 224.
Sixteen is a special number
When the number of teams is a power of two, the best possible schedule can be written down directly. Label the 16 teams with the 16 four-bit strings. Adding two labels is XOR. Then pick an element g of the field GF(16) whose powers run through all 15 non-zero labels. In round r, team x plays team x XOR g^r.
That is a round robin: XORing twice gets you back where you started, and the pair {x, y} meets in the one round where g^r equals x XOR y. It is also perfectly fair. Team t plays t + g^r and then t + g^(r+1), and those two opponents always differ by g^r(1 + g). That difference tells you r, and r tells you t. So every "A, then B" sequence happens for exactly one team.
Leaving one team out as the bye keeps the property, so 15 teams works as well as 16 (and 7, 31).
What changed in the tool
At 4, 8, 16 and 32 teams (and 3, 7, 15, 31) the search now starts from this timetable. It tries every generator and every starting round (8 x 15 at 16 teams), which all have the same carry-over, and keeps the one with the fewest back-to-back home or away games. Measured on seed 1 (table):
| Teams | Season | Before (carry-over / breaks) | Now | Least possible |
|---|---|---|---|---|
| 16 | single | 316 / 32 | 224 / 24 | 224 |
| 16 | home and away | 1,048 / 110 | 912 / 64 | 912 |
| 15 | single | 259 / 43 | 195 / 41 | 195 |
| 32 | single | 1,538 / 140 | 960 / 106 | 960 |
On the page itself, which runs ten searches and keeps a balance of carry-over and breaks, a 16-team home-and-away season went from 1,220 carry-over with 54 breaks to 996 with 48. It is better on both counts.
There is a nice side effect for home-and-away seasons. Mirroring the first half usually doubles every count, which is why the tool searches the second half separately. With the field timetable, though, each team's last opponent of the first half followed by its first opponent of the second half is exactly the one "A, then B" step that half was missing. The plain mirror is already perfect (912 at 16 teams), and the venues just swap.
One thing broke along the way. The search's second pass trades carry-over for fewer breaks using a quick greedy estimate of the breaks. From a perfect start, that estimate sometimes gave away carry-over for breaks the real home/away assignment never delivered (224 / 24 became 256 / 24). The tool now scores both exactly and keeps the better one.
What it does not do
Other sizes have no construction this simple in the code yet. Twelve teams still search from the circle method and stop at 156 against a floor of 120; whether 120 is reachable for 12 is not something we have checked. And the practical question is still open: whether carry-over moves results in a rec league at all. No organiser has used this feature yet.