Table 1.

Errors of pairwise assembly rules are much lower than monoculture prediction errors

Prediction type	Mean error/max. error, trios	Mean error/max. error, quads	Overall
Pairwise outcomes	0.09 (0.02)	0.19 (0.02)	0.14 (0.01)
Carrying capacities (monocultures)	0.40 (0.03)	0.43 (0.03)	0.41 (0.02)
Random predictions	0.46 (0.03)	0.47 (0.03)	0.47 (0.02)

We made quantitative predictions of the relative fractions in multispecies competition outcomes using both monoculture carrying capacities as well as the pairwise assembly rules. Mean errors of both types of predictions are shown, as well as mean error for random predictions, for comparison. Errors of quantitative predictions are the L2 norm of the distance between predicted fixed point and observed fixed point (see “Methods” and Supplementary Fig. 2). The values shown are mean error normalized by the maximum error. Errors, shown in parentheses, are SEM of replicates (n = 118 for trios, n = 96 for quads; outcomes of different starting fractions of the same biological replicate were averaged before measuring error, and in the case of bistability, the smaller of the two errors was chosen after averaging outcomes from different starting fractions ending at the same final state)