Brand-new cups: a cutoff with no history
Every week brings cups nobody has played before: a collaboration cup with its own scoring, a new format, the first week of a series in a region. There is no previous edition to read. There is, however, a very close substitute: the same players, playing the same game in the same region, a few days earlier.
Replaying last week under this week's table
A finished leaderboard records, for every team, where it placed in each game and how many eliminations it made. Those games can be scored again under any table: this placement worth that many points, these eliminations at that rate, capped where the new table caps them, only the first games counted where the new cup allows fewer. The standings that come out are the ones the new table would have produced on those evenings.
That is how the forecast prices a cup it has never seen. It takes the six most recent leaderboards of the same region, team size, game mode and platform, preferably played to the same number of games, re-scores each under the new cup's table, and reads the threshold at each rank off the median of the six. Single-lobby formats – Arena tests, evaluation lobbies – are left out, because a lobby of twenty is not an open queue of thousands.
It answers the question an average cannot: two players with the same average score, one steady and one streaky, fare differently under a table that rewards wins more than eliminations. Replayed, each is scored as what they actually did.
What it is worth
Before the replay existed, a new cup was priced off its scoring table alone: the most a team could score, times the share of it that rank 20 usually reaches in cups of the same kind. On the 102 open-queue cups since mid-August that had no previous edition in their region, each forecast from leaderboards played before its day, the replay was about 4 to 5% off at ranks 1 to 250 and about 6% at ranks 250 to 1,000, where the scoring table alone was 8 to 11% off. On mobile cups, where the scoring table's shares fit worst, ranks 1 to 100 went from 14.5% to 5.7% median error.
Two limits remain, and the forecast says when they apply:
- Depth. The replayed leaderboards are read a few pages deep, so the replay is trusted down to a third of the teams loaded; past that, the forecast continues along the measured ladder of the field.
- A new format with no leaderboard to replay – an Arena cup whose number of games is not fixed, the very first cups of a new mode – still falls back to the scoring table, with a much wider range.
When the cup has run in another region
A new cup often runs in several regions over a few days. Once other regions have played it, their editions are evidence: the forecast combines them, scaled to the field, with the replay of the region's own recent leaderboards, half and half in log terms. On the day-two cups where both were available, the combination beat either alone. The model is rebuilt once a day, so a region that played the same evening counts from the next day.
What it does not do
A new cup is never priced off an older cup that merely has a similar name. The calendar names each cup the way the model does, from Epic's own event identifiers, so a cup the model has not seen is treated as new. Borrowing a neighbour by name once priced a solo Zero Build cup off a trio cup whose name shared three words: 463 points at rank 100, where its own format's leaderboards, replayed, said 294.
Updated 22 September 2026