How can the perception of orientation be systematically wrong?

The target article by Vannuscorps, Galaburda and Caramazza (VGC hereafter) reports work that includes extensive investigations of their subject Davida’s performance on two sets of forcedchoice tasks that required correct perception of orientation: naming or copying orientationally-confusable letters (mainly b d p q), and identifying the direction in which an arrow is pointing (with arrows pointing up, down, right or left). This is the work I consider here. Table 1 (see Supplementary Information S1) reports all of the 17 datasets from these tasks that were obtained when the displays were two-dimensional with stationary black figures on a white ground and edges that had high luminance contrast and were unblurred or minimally blurred (high in spatial frequency). Davida was completely unable (more on this shortly) to perform any of these tasks under such presentation conditions. In contrast, when the stimuli were presented with low luminance contrast, or as three-dimensional objects, or were greatly blurred (i.e., with very low spatial frequencies), or were in motion, she performed all of these tasks very well – typically 100% correct. Figure 1 of the target article offered a model of the processes by which the retinotopic retinal representation evoked by a visually-presented object or scene is transformed into “a behaviourally relevant [spatiotopic and body-centred] frame of reference”. According to this model, the processing of retinotopic retinal representations is carried out by two pathways operating in parallel upon these representations. There is a parvocellular pathway characterized by sensitivity to colour and by relative insensitivity to low contrast, low spatial frequencies and motion; and there is a magnocellular pathway characterized by sensitivity to motion, insensitivity to colour, and relative insensitivity to high contrast and high spatial frequency stimuli. My Figure 1 is a redrawing of Figure 1 from the target article which perhaps makes the information-processing structure of the VGC model more transparent. A key concept in this model is the intermediate shape-centred representation (ISCR), defined by VGC as