Layer Cake

What you see here. The best three-brawler line-up that can be assembled for this one map — computed, not counted. On a single map hardly any line-up has enough games for a measured number of its own, so it is assembled from the individual strengths and the pair effects, then calibrated against real line-ups.

Best · on this map. All 6,455 combinations of this map's 40 best brawlers, scored from their strength here and their pair effects in this mode — the same calculation as the draft helper, only without opponents. How much of it comes through is not guessed but calibrated against 384 real line-ups from a different half of the battles, so the calculation cannot measure itself.

What the calculation cannot do: check itself. Measured against line-ups that actually exist, the projection errs low rather than high — across all modes the top six came out between 4.9 points too high and 9.1 too low, the top ten percent to within a point. For a line-up nobody plays there is no such check, and sometimes there is a reason nobody plays it. The list is therefore cut off at the measured boundary: it shows the best line-up for which any evidence still exists, not the best that can be computed. The calculation does produce higher ones (76 of them on this map) — they sit outside anything ever measured, which is why they are not shown here.