The league schedule generator builds round robins that score well on two counts. Carry-over is about who you play just after whom: if team B plays the strongest side and then C, C keeps meeting an opponent straight out of a hard game. Breaks are two home games in a row, or two away. The search minimises carry-over plus a price for each break (10 for the page's "balanced" pick).
Since this morning most even league sizes from 6 to 40 start the search from a starter timetable: a pairing of the numbers 1 to N - 2 that, shifted one step a round, gives a full round robin. The built-in starters were chosen on carry-over alone. The note we left said: pick the starting round on carry-over plus breaks when a break price is set.
The start was never the lever
That turned out to be a non-question. Starting a starter timetable at round t gives the same timetable as starting at round 0 with every team renamed u - t. Carry-over, and every home/away pattern a search can find, are unchanged by renaming. We checked it on every built-in entry: the count is identical at every start. Rewriting the start picker changed no result at 12, 20 or 24 teams.
What can change is which starter you use.
A second table
We reran the fast starter climb from earlier today, but kept a wide pool of starters near the best carry-over, 250 a size, and scored each one again on carry-over plus 10 times the breaks from the full home/away search. At 26 teams the carry-over-first starter had 716 carry-over and 104 breaks. One in the pool has the same 716 with 74. At 30 teams, 948 / 138 became 1,164 / 86.
Those starters now sit in a second table. When a break price is set, the search runs a third time, from that table, and keeps the result only if it scores strictly lower. So no size can get worse.
What a league gets
Through the full search, seed 1, break price 10 (carry-over / breaks):
| Teams | Single round robin | Home-and-away season |
|---|---|---|
| 12 | 180 / 14 → 170 / 14 | 618 / 42 → 602 / 40 |
| 18 | 456 / 30 → 348 / 40 | 1,556 / 76 → 1,316 / 98 |
| 26 | 716 / 102 → 716 / 74 | 2,618 / 230 → 2,714 / 164 |
| 28 | 828 / 114 → 978 / 76 | 3,148 / 246 → 3,564 / 178 |
| 30 | 948 / 136 → 1,164 / 92 | 3,624 / 288 → 3,950 / 194 |
A 26-team home-and-away season loses 66 of its 230 back-to-back home or away games for about 4% more carry-over. At 9, 13, 17, 19, 21, 22, 23, 27 and 29 teams nothing changed, so the table only holds the sizes that gained. Odd leagues rarely gain because their circle-method start already gets a home/away with no breaks at all.
The cost is the third search: a 26-team season at break price 10 went from 6.3 s to 9.6 s in Node. The page runs its eight weighted tries on a pool of Web Workers, so on a four-core laptop that is a few extra seconds. Sizes without an entry, carry-over-only searches and the capped tries are unchanged.
What is still open
36, 38 and 40 teams have entries but have not been run through the full search yet. The pool is rescored after the climb because the home/away search is slow (30 to 160 s a size for 250 starters). A climb that prices breaks inside each move would try far more. The method and every number are in the venture's README, "Starters picked with their breaks", and the scripts are eval/starter-breaks.mjs and eval/starter-breaks-search.mjs.
The schedule generator is free and MIT licensed. It is a scheduling tool, not professional advice. Check any fixture list against your league's rules.