Significant shape differences were identified between the replica and artefact assemblages in all instances. The Kruskal Wallis 2D shape comparisons returned p values of .0002 and < .0001 between the replica tools and artefacts, for PC1 and PC2 respectively. Similarly, the 3D data revealed PC1 and PC2 to vary significantly between samples, returning p values of < .0001 in each instance. All significant differences were mirrored by the repeat tests run with equal samples sizes and using Mann-Whitney U tests (Supplementary Information, Table 1).

Plots of PC1 against PC2 similarly reveal shape differences between the replica and artefact assemblages (Fig. 2 and Fig. 3). In both sets of analysis, the replica tools take up a greater amount of shape space than the artefacts, and thus exhibit greater form diversity. Comparison of the 2D and 3D plots, however, identify the former to have substantially greater shape space overlap between the artefacts and replica tools. Both the convex hulls (Fig. 2 and Fig. 3) and 95% ellipses (Fig. 2 and Fig. 3) highlight this greater overlapping of plan-view shape. Indeed, despite the 2D PC1 and PC2 median values being significantly different, the plots identify few tools (artefact or replica) laying outside of the shared shape space. In other words, the “highly variable” (Key and Lycett, 2017b: 517) replica handaxe forms are mirrored by the diversity of plan-view shapes produced by Acheulean hominins. Standard deviation values for PC1 and PC2 in the 2D analyses also indicate similar levels of shape variation between the tool-sets, with the replica tools displaying only marginally greater values. Thus, when viewed from a broad multi-assemblage level, there appear to be weak restrictions on the 2D plan-view shape of handaxes during the Acheulean techno-complex.

There is considerable overlap between the two tool-group's 3D shape spaces, with nearly all artefacts being subsumed within the convex hull of the 480 replica bifaces. In contrast to the 2D analyses, there are clear limits to an Acheulean handaxe's 3D form. Indeed, Figs. 3b and 3c identify numerous replica biface forms that have no equivalent in the artefact record (see also: Fig. 4, Fig. 5 and Fig. 6). Acheulean handaxes therefore display strong shape thresholds. Importantly, it is only through 3D analyses that this is revealed, suggesting that tool thickness and material volume drive these form limitations. Indeed, those bifaces not observed in the artefact record typically display high PC1 values, indicating low thickness to width ratios (i.e. high ‘refinement’ levels). Fig. 4 and Fig. 5 demonstrate that most bifaces found outside of the artefact's 3D shape space are relatively thin. Acheulean handaxes do, therefore, demonstrate strong shape thresholds defined by lower limits to their relative thinness.

There is some indication that tools with particularly low PC2 values, which point towards low elongation ratios (i.e. near circular tools), are also not reported in the archaeological record (Fig. 4, Fig. 5 and Fig. 6). Principal component standard deviation values indicate the 3D shape variation of Acheulean handaxes to be restricted relative to the replica tools in all instances (Table 2). In many instances artefact assemblages display only half the variation observed in the replica tools.

Mean PC1 and PC2 values, along with the shape space plots in Fig. 2 and Fig. 3, reveal subtle and not so subtle variation in the 2D and 3D shape of all 10 artefact assemblages. Although the present study is focused on Acheulean handaxes at a very broad technological level, it is clear there is also inter and intra-site variability in the shape of these tools. This variability is similarly reflected in each artefact assemblage's standard deviation values, with clear differences observed between some shape variation levels.

Replica stone tool assemblages are rarely used to understand morphological limits and variation levels in Palaeolithic artefacts. Here, this technique has been used to better understand the shape of Acheulean handaxes. It has been demonstrated that handaxe artefacts vary highly in their plan-view (2D) shape, with few limitations observed relative to two substantial and intentionally diverse replica tool-sets. Stricter morphological thresholds were, however, observed in the thickness and material volume of the Pleistocene tools. Indeed, the same handaxe artefacts occupied under 50% of the 3D shape space created by the replica bifaces. Thus, while the plan-view shape of handaxes appears to have been free to vary during the Lower Palaeolithic, stronger thresholds were observed regarding their 3D form. This does not mean that Acheulean handaxes are not variable in their 3D shape. Rather, the replica assemblage reveals a range of handaxe forms that could theoretically have been produced by Lower Palaeolithic hominins, but were not.

The replica handaxes not reciprocated in the archaeological record are, predominantly, relatively thin, flat and ‘tabular’ in form (Fig. 5). The use of principal component analysis makes it difficult to empirically define a threshold for any one morphological trait, but the tendency for artefacts to avoid high PC1 values in the 3D analysis, not going beyond 1.26, supports this conclusion. Avoidance of these handaxe forms was observed across the 371 artefacts in the 3D sample, despite their diverse origins. Their absence suggests that strong pressures limiting the production of highly thin or ‘refined’ tools were acting on Acheulean hominins. Consistent with traditional Palaeolithic terminology, ‘refinement’ refers only to a tool's thickness relative to its width.

The thickness of a handaxe relative to its width is a widely recorded shape trait. Unsurprisingly then, reports detailing the absence of highly ‘refined’ Acheulean bifaces is not novel in and of itself (e.g., Gowlett, 2011a, Gowlett, 2015, Li et al., 2014, Moncel et al., 2015, Norton et al., 2006 and Petraglia and Shipton, 2008). Despite considerable ‘refinement’ diversity being reported (Emery, 2010, Isaac, 1969, McNabb and Cole, 2015 and Shipton and Petraglia, 2010), rarely do artefacts display thickness to width ratios below 0.3, while mean values frequently range between 0.4 and 0.5.

Presented here, however, is evidence pertaining to the limits placed on Acheulean handaxe thickness within the context of what could theoretically have been produced by hominins. It has been demonstrated that while there is variation in the shape of handaxe artefacts, including their relative thinness and globular/tabular nature, when compared to the potential variability that could have been produced there are strong shape-space restrictions and central tendencies. Moreover, these predilections are pervasive across the cultural variability observed in the Acheulean techno-complex (Lycett and Gowlett, 2008), and exist in spite of any allometric size-shape shifts (Crompton and Gowlett, 1993). Thus, in spite of any observed 3D shape variation, this newly defined relative context reveals a strong central tendency in Acheulean handaxe 3D morphologies. Further evincing the ‘variable sameness’ observed by Glynn Isaac and others (Gowlett, 2011b, Gowlett, 2015, Isaac, 1977 and Lycett and Gowlett, 2008).

This tool-form propensity is likely due to several interacting factors. While handaxe shape, broadly speaking, has been shown to be weakly related to cutting performance (Key and Lycett, 2017b and Machin et al., 2007), the production of highly ‘refined’ tools (particularly in their proximal [base] portion) may limit the amount of force able to be exerted during cutting activities (Jones, 1980, Key and Lycett, 2019 and Key et al., 2016). Working edges also require support from sufficient volumes of material to resist loading, prevent breakage during use, and reduce torque (Gowlett, 2006 and Key, 2016). Relatedly, limited use-lives (Shipton and Clarkson, 2015) could have reduced reduction and flaking levels, preventing proximal shifts of a tool's centre of mass and maintaining thickness more generally across the tool (Archer and Braun, 2010 and McPherron, 2006).

Thickness and ‘refinement’ measures are also strongly linked to a tool's material volume, and in turn, weight. This is important because substantive weight decreases (beyond context-dependent thresholds) have potential to negatively influence a stone tool's ability to be applied during functional tasks (Gowlett, 2009 and Key and Lycett, 2017b). Moreover, specific functional tasks may have preferentially necessitated the production of thicker, heavier handaxes (Wynn and Gowlett, 2018). Additionally, handaxe shape (including ‘refinement’) is known to influence grip choice in tool users (Key et al., 2018), and the globular nature of Acheulean handaxe bases (i.e. a relatively thick base) has been argued to aid tool ergonomics and position its centre of gravity in the hand (Grosman et al., 2011, Jones, 1980, Jones, 1981 and Wynn and Gowlett, 2018). Combined, these factors would promote the production of relatively thicker mean forms. Thus, the tool-form limits identified here may represent ‘primary performance characteristic thresholds’, as proposed by Schiffer and Skibo (1997).

Equally, there may have been no, or only very weak, pressure selecting for highly ‘refined’ handaxes forms. Certainly, factors promoting highly thin forms in other bifacial technologies, such as hafting or their use as a projectile, are not frequently applied to Acheulean handaxes. Thus, irrespective of any pressures promoting thicker tools, there may have been little cause to thin bifaces beyond what was necessary to shape the tool and maintaining a sharp, relatively acute, cutting edge. Alternatively, hominins may also have been limited in their ability to produce relatively thinner and more ‘refined’ handaxes. Indeed, due to manipulative, cognitive or technological limitations (Key and Dunmore, 2018, Muller et al., 2017, Stout et al., 2014 and Wynn, 2002), it may not have been possible for early handaxe-producing hominins to produce highly ‘refined’ handaxes. Although restrictions would have reduced once populations displayed increasingly modern human-like anatomy. Relative thickness has also been linked to raw material availability (Ashton and McNabb, 1994 and White, 1998). For example, the production of handaxes from large stone flakes or bone, as opposed to stone nodules, has been suggested to aid the production of relatively thinner tools (Li et al., 2014, Moncel et al., 2015, Sharon, 2009 and Zutovski and Barkai, 2016). Although Shipton (2018) highlights how raw material and relative thickness relationships can vary between individual sites. Irrespective, the diversity of artefact raw materials examined here, along with their production on nodules and large flakes by multiple hominin species, means that raw material and evolutionary factors cannot alone explain the thresholds observed here.

It is, perhaps, more surprising that the 2D shape diversity observed in the replica assemblage was closely matched by the Acheulean artefacts. Indeed, there was near total overlap in their convex hulls and 95% confidence ellipses. Further, both assemblage's principal component standard deviation values are similar, indicating comparable levels of variation. When considered at a broad, technological level (i.e. not at an assemblage level), these results identify substantial variation in the plan-view shape of handaxes produced by Lower Palaeolithic hominins.

It is not surprising that considerable diversity was observed in the artefacts, as previous studies identifying shape variation within and between handaxe assemblages attest to this (Hosfield et al., 2018, Lycett and von Cramon-Taubadel, 2015, McPherron, 1999, Roe, 1981, Vaughan, 2001 and Wynn and Tierson, 1990). Nor is it surprising that plan-view shape variation is controlled independently from a handaxe's third axis (Archer and Braun, 2010 and Gowlett, 2006). Certainly, Acheulean populations appear to place stronger emphasis on controlling the distribution of mass and relative thickness than 2D plan-view shape. What is revealing, however, is that artefact shape diversity matches two replica assemblages that were purposefully produced to be highly variable. This suggests that any universally observed pressures working to limit handaxe plan-view shape in the Acheulean were likely very weak. Given that handaxes were, ultimately, functional objects applied to cutting tasks, these results emphasise the limited impact that plan-view shape variation likely has on their utilitarian capabilities (at least within a generalised functional environment) (Key and Lycett, 2017b, Machin et al., 2007 and Schiffer and Skibo, 1997). Further, these data highlight the ‘area of free play’ available for handaxe's, within which shape can vary without impacting utility (Crompton and Gowlett, 1993, Gowlett, 2009 and Lycett et al., 2016). The artefact shape space outlined in the PC plots provides a framework for understanding this area of free play.

Acheulean handaxe shape variation is contextualised here within a theoretically possible range of tool-forms, as defined by highly diverse modern replica tool sets. It has been demonstrated that handaxe artefacts exhibit substantial diversity in their 2D plan-view shape, displaying near complete overlap with two intentionally diverse replica tool sets, and similar levels of variation. Much stronger restrictions and central tendencies on a handaxe artefact's 3D form were identified; principally driven by an absence of flat and ‘tabular’ tools displaying low thickness to width ratios in the artefact record. It is argued that while there is considerable diversity and variability in the shape of Acheulean handaxes, their form is nonetheless restricted by strong material volume and relative thickness limits (driven, in part, by functional factors). Shape diversity is more restricted at an individual assemblage level, while mean shape differences indicate differential central tendencies and shape preferences between sites. Supporting past works emphasising the impact that cultural evolutionary mechanisms, and other localised environmental factors, can have on handaxe shape.