Left: spectral energy distribution from the NIR to the X-ray at early epochs. The optical-to-X-ray SEDs are best described by the sum of an absorbed broken power-law (dashed lines) and a blackbody (dotted lines). Data excluded from the fits are shown as empty squares. If these data would be added, the fit statistic for an absorbed broken power-law (excluding the UV data) significantly worsens from χ²/ d.o.f. = 103.7 / 65 to χ²/ d.o.f. = 159.1 / 67. The interpolated values also significantly deviate from the best-fit model afterglow light curves (Fig. 2). Upper limits are shown by triangles. Right: extrapolation of the X-ray/NIR SED to the sub-mm-region compared to the available upper limits (Table 2). The NIR-to-X-ray SED from T₀ + 0.267 days was extrapolated to radio frequencies and evolved in time for a collimated and spherical expansion of the blast wave (for details see text). The AMI-LA measurement from T₀ + 2.38 was shifted to 2.6416 days, assuming the injection frequency to be blueward of the observed bandpass and using the scaling relations in Sari et al. (1998). This has no implications on our analysis.