45behavioral complexity increased (Carbone and Narbonne, 2014; Seilacher et al., 2005), prompting 46 hypotheses about major shifts in ecological interactions and trophic structure during this period, due to 47 major changes such as widespread predation and active (vertical) burrowing, which may have facilitated 48 the first complex 'modern' food-webs (Dunne et al., 2008; Conway-Morris, 1986; Vannier and Chen, 49 2005, Erwin and Valentine, 2013; Mángano and Buatois, 2014). 'Conservation lagerstätten' sedimentary 50 deposits, featuring exceptional fossil preservation of both 'soft' and 'hard' body features (Orr et al., 51 2003), permit detailed studies from which species interactions can be deduced (Butterfield, 2003).
93organized to facilitate analyses at a finer depth scale (hereafter referred to as the running time-frame 94 analysis) (methods). These correlations, with relevant regularization, were then used to construct 95 correlation networks, for each sub-assemblage and each component of the running time-frame analysis. In 96 these networks, each node was a taxon and each edge between a node pair represented significant 97 correlation and thus possible interaction (methods). 98 3 99 Figure 1: (a) Location of Raymond Quarry (RQ) (Yoho National Park, British Columbia, Canada), 100 denoted by red dot; yellow region denotes extent of major Burgess Shale localities; Samples were 101 collected from the RQ member along the 'fossil ridge' connecting Mt. Field and Wapta Mountain. (Figure 102 S1, methods) (b) 'Preservation bias coefficient' for body type (with respect to soft, intermediate and hard 103 bodied categorization) (in red), habitat type (in black) and body size categories (in blue) calculated using 104 networks in running timeframe analysis, plotted along with estimated boundaries for the distinct sub-105 assemblages (A-D) in the 9.3 m shale section using variations of two different statistical approaches for 106 biofacies detection: ANOSIM and SHEBI (Figure S2, methods) and the associated average bias 107 coefficient for each sub-assemblage (in their respective index colors for each type of bias). The green 108 dotted line depicts the bias threshold we adopted for inclusion versus exclusion of sample layers.
109Preservation bias coefficients exceeding 0.5 translate into substantial changes in the structure of 110 interaction networks calculated from abundance correlations ( Figure S3b). Note that the 10 cm layers 111 comprising regions A' and D' were originally identified as belonging to sub-assemblages A and D, but 112 fossils from A' and D' were not used in the analyses presented here because of evidence for high levels of 113 preservation bias among taxa.
115Correlation networks can yield insights into possible interactions among taxa (Zhang, 2011; Carr et al., 116 2019; Barberán et al., 2012), but network features can be obscured by preservation and collection biases 117 (Dunne et al. 2008; Carr et al., 2019; Jordano, 2016). Intensive sampling and detailed annotation reduced 118 collection bi...