English: Visual stimuli The first two iterations show a 3-s excerpt from the fighting sequence (twice the duration of the stimuli used in the experiments; see Supplementary video 1 in <ce:cross-ref refid="bib25">Neri et al. (2006)</ce:cross-ref> for a movie of the entire sequence). In the first iteration all trajectories are visible. The head trajectories (open circles) were not used. For each stimulus, only 3 limbs per agent were sampled; these are indicated by black dots (same convention as in <ce:cross-ref refid="fig1">figure 1</ce:cross-ref>a) for the example shown here (different limbs were selected randomly for different stimulus presentations). The resulting 18 trajectories (2 agents×3 limbs×3 joints) were sampled using a limited-lifetime procedure as demonstrated in the second iteration (grey background); this sequence provides an approximation to the actual stimulus that was used in the experiments. The first two iterations just described refer to the ‘target’ stimulus. The third and fourth iterations show the results of applying the limb-scrambling procedure to the same excerpt to obtain a ‘non-target’ stimulus (in the actual experiments target and non-target were generated from different randomly selected excerpts). In the third iteration, joints marked by the same colour undergo an equal temporal phase shift, but different phase shifts were applied to the different colours. The colour grouping is such that joints belonging to the same limb are dephased (scrambled) with respect to each other, but average limb trajectories are in-phase. The fifth and sixth iterations show the results of applying the body-scrambling procedure. Now the colour grouping is such that joints belonging to the same limb are in-phase, but the different limb trajectories are dephased with respect to each other. It can be seen that limb-scrambling (iterations 3–4) reduces the percept of agency in the stimulus much more than body-scrambling (iterations 5–6), even though the same set of phase shifts was applied in both cases. This effect is quantified by the data reported in <ce:cross-ref refid="fig2">figure 2</ce:cross-ref>a, where it is also shown that it is not a mere consequence of low-level differences between the two manipulations.