Experimental impacts were carried out on subadult femoral shafts of Sus scrofa (n = 31). Specimens were obtained defleshed from local abattoirs, kept frozen until the day of testing, then thawed in water and used immediately.

All strikes, both chops (n = 20) and bashes (n = 11), were performed using the Instron 9440 Drop Tower System, which ensured controlled and repeatable delivery of force across all trials (Figure 1A).

Chop impacts were delivered using reproduction axes made of copper, bronze, iron, and modern steel (Figure 1B). These replicas were based on generalized archaeological forms and metallurgical compositions from the southern Levant rather than a single specific artifact (Ashkenazi et al., 2016; Erez et al., 2016; Gottlieb, 2010). The pure copper, bronze (5.5% tin), and hand-forged bloomery iron axes reflect technological developments spanning from the Chalcolithic to the Iron Age, capturing key shifts in material properties and manufacturing techniques that would have influenced chopping performance. Ax dimensions are provided in Table 1 for documentation. Blade width was not treated as an analytical variable, as experimental work demonstrates that fracture mechanics, rather than blade width, govern sharp force trauma expression on long bone shafts under controlled impact conditions (Okaluk et al., 2025). The use of generalized forms provides a representative sample of ancient technological variability while avoiding the idiosyncrasies of individual artifacts. Ancient axes were not standardized and display considerable variation in both form and material (Shalev et al., 2014), so this approach allows for controlled comparison across broad technological categories. A modern steel ax was also included for comparative purposes, offering a reference point for a highly durable cutting edge. Each ax was drilled and fitted with bespoke attachments to allow secure mounting to the impact machine arm.

Bash impacts were delivered using the rounded stainless steel hammer supplied with the Instron 9440 Drop Tower System (Figure 1C). The hammer measured 1.5 cm in diameter (Figure 1D). Using a smooth stainless steel hammer as the percussive impactor ensured that bash and chop trials could be compared under identical and repeatable force and placement conditions. Archaeological hammerstones vary widely in form and surface texture, and their irregularity introduces heterogeneous contact geometries and pressure distributions at the point of impact, which can alter local stress concentrations and fracture initiation even under similar loading regimes. This variability was intentionally controlled by using a standardized blunt impactor in order to minimize within-class variance in the blunt condition and isolate fracture-level responses to blunt vs. edged impacts. The steel hammer thus served as a controlled, reproducible proxy for a rounded blunt impactor, prioritizing comparability and repeatability across trials rather than replicating the full range of archaeological hammerstone morphologies.

Thawed femora were placed unrestricted and unsupported on a flat wooden board beneath the Instron 9440 Drop Tower impact arm, with the bone resting directly on the board and no restraints applied to either end. Alignment of the impact point was visually checked prior to each strike. Impacts were delivered by releasing the drop tower carriage from a fixed height, allowing the impact arm with the mounted tool to fall vertically and strike the bone in a single percussive blow. The machine controlled for speed, mass, direction, and force, and recorded velocity at the moment of impact.

Each bone was subjected to a single impact delivered to the posterior face of the diaphysis at a 90° angle along the transverse plane. Impacts were delivered at 30 J of energy, corresponding to an 8.2 kg mass and an impact velocity of 2.7 m/s. Impact energy was standardized at 30 joules as this was the minimum level determined through pretesting to achieve complete fractures in femora without causing catastrophic fragmentation, ensuring realistic and comparable chopping conditions. Specimens that did not fracture, or that fractured only partially, were classified as incomplete.

Following impact, bones were boiled and treated with detergent as a degreasing agent to remove residual fats. Fragments were then re-fitted with white glue to reconstruct specimens for analysis. Each specimen was examined under raking light, and all fracture characteristics were recorded by the first author according to the variables and definitions outlined below.

Each impact was coded for impact location shape (straight or irregular), gross fracture type (e.g., transverse, oblique, butterfly, comminuted), bone surface modifications (outer conchoidal flake, percussion notch, pecking, none), fracture origin (anterior, posterior, lateral, medial), and the presence or absence of SFT.

These variables were selected to capture both surface-level and internal fracture characteristics relevant to differentiating chopping and bashing. Impact location shape records the geometry of the fracture edge at the point of impact. Gross fracture type documents overall break morphology, which reflects how force propagated through the bone (Figure 2). Bone surface modifications refer specifically to anthropogenic alterations to the cortical surface, including SFT, percussion notches, and conchoidal flakes, following common zooarchaeological usage (Blasco et al., 2014; James and Thompson, 2015; Pickering and Egeland, 2006; White, 1992). Fracture origin records the location where failure initiated on the bone surface, providing insight into the loading mechanics of each impact. Definitions follow established zooarchaeological and biomechanical terminology, with clarifications provided below.

Butterfly fracture and butterfly type

Fracture mechanics (contextual definitions)

Impact area morphology

Bone surface modifications

Sharp force trauma (SFT)

Fractographic features

To locate impact areas and interpret fracture propagation, internal fracture surface features were examined following forensic fractography approaches (Christensen et al., 2021, 2018). Dynamic bending fractures typically exhibit three distinct zones (Figure 7).

Analyses focused on identifying which variables were most predictive of chop vs. bash impacts and evaluating how these variables performed in combination. Three complementary approaches were used.

First, unsupervised divisive hierarchical clustering using Gower distance was performed to explore underlying structure in the dataset and assess whether impacts produced by different ax materials (copper, bronze, iron, modern steel) could be grouped for analysis.

Second, variable-level associations were tested using two-tailed Fisher's Exact Tests. Exact p values were calculated for 2 × 2 tables. Variables were then ranked by p value to guide model building.

Finally, multivariate logistic regression models were constructed by adding variables stepwise in order of their Fisher rankings using a step-wise forward-selection WAIC algorithm. Model performance was compared using the Widely Applicable Information Criterion (WAIC) and Leave-One-Out cross-validation (LOO), both of which balance predictive accuracy with model complexity. Variables were modeled additively without interaction terms.

All analyses were conducted in R using standard statistical and modeling packages, including “brms” for Bayesian logistic regression (Bürkner, 2017), which allowed for model comparison using WAIC and LOO criteria.

A divisive hierarchical clustering using Gower distance was conducted to assess whether impacts created by different tool types could be meaningfully grouped for subsequent analyses. This unsupervised method grouped observations according to their fracture characteristics without using chop or bash labels as a target. Comparing cluster compactness reveals that separation into five clusters is optimal (Figure 8A). At this resolution, the results revealed a clear separation between bashing and chopping impacts (Figure 8B). All steel hammer impacts formed a single, distinct cluster that was completely separated from the clusters containing the four ax materials (copper, bronze, iron, and modern steel axes). Although some internal sub-clustering among ax types was observed, particularly between modern steel and bronze construction, and iron and copper construction, unsupervised clustering provides support for the idea that bash impacts can be clearly distinguished from other types of impact. While inter-ax differences are addressed elsewhere (Okaluk et al., 2025), pooling them in the analyses that follow can further explicate their differences without the loss of statistical power associated with fragmenting the classes, and while reflecting their overall similarity in fracture morphology. The unsupervised clustering thus aligns with the statistical modeling by indicating that chopping and bashing impacts are systematically distinct. For full data see Table 2.

Two-tailed Fisher's Exact Tests were used to identify individual variables that differed significantly between chop and bash impacts. Several variables were strongly associated with impact type, while others showed weaker or non-significant relationships but nonetheless contributed to predictive performance in combination (Table 3). The most significant differences were observed in impact location shape, the presence SFT, and fracture type. Straight impact location shapes were significantly associated with chops rather than bashes (p = 0.0201), and SFT occurred exclusively in chop impacts (p = 0.0331). Fracture type also differed significantly between impact types (p = 0.0416), with comminuted fractures occurring more frequently in bashes, whereas transverse and oblique fracture bases were more common in chops. Several additional variables showed weaker associations: the absence of impact marks approached significance (p = 0.0659), and butterfly type showed a less pronounced but suggestive pattern (p = 0.151). Other variables, such as pecking (p = 0.2409), transverse fracture types considered alone (p = 0.2755), fracture origin categories, and flake scars (p = 0.631, the lowest predictive score), did not show significant individual differences. Nonetheless, all variables were retained for multivariate modeling because even those with weak individual associations may contribute predictive information when combined with others.

Logistic regression models were used to evaluate how combinations of variables predict the probability that an impact was produced by chopping. Variables were added stepwise in order of their Fisher test rankings, and model performance was assessed using WAIC and LOO scores to balance predictive power with model simplicity.

Across all models, three variables consistently emerged as the strongest predictors of chop impacts: straight impact location shape, the presence of SFT, and fracture type. Pecking and butterfly type contributed additional, though weaker, predictive information. Regression coefficients are reported on the logit scale, which represents the log odds of a chop impact when each feature is present.

Straight impact location shape had a coefficient of 0.82, corresponding to approximately 2.3 times higher odds of a chop impact relative to irregular impact locations. The presence of SFT had a coefficient of 0.98, equivalent to 2.7 times higher odds of a chop impact compared to its absence. Comminuted fractures had a negative coefficient of −0.96, corresponding to about 0.38 times the odds of a chop relative to other fracture types. Butterfly type (normal) had a negative coefficient of −0.74, indicating a weaker but consistent trend, while pecking had a positive coefficient of 0.63, reflecting a modest increase in the likelihood of a chop when present.

Importantly, the model could still predict chop impacts with high probability even when SFT was absent. For example, in a case with a straight impact area and pecking, but without SFT, a butterfly fracture, or comminution, the model would still predict the impact to be a chop. This model predicted these variables to be a chop with an 81.7% probability. This demonstrates that SFT is a useful indicator when present, but it is not required for chop identification.

The best-performing model correctly classified the majority of impacts, identifying 18 of 20 chop impacts and 7 of 11 bash impacts correctly, with an overall accuracy of 80.6% and a sensitivity of 90% for chop impacts (Table 4). Several models produced similar WAIC and LOO scores, indicating that the results are stable across alternative model structures.

The results highlight both clear tendencies and meaningful areas of overlap between chopping and bashing impacts. Across all statistical approaches, including Fisher's exact tests, hierarchical clustering, and logistic regression modeling, three variables consistently emerged as the strongest predictors of chop impacts: straight impact location shape, the presence of SFT, and non-comminuted fracture bases. Straight impact location shape showed the strongest positive association with chopping, reflecting the concentrated linear loading applied by an ax blade compared to the more diffuse force of a hammer. The presence of SFT increased the likelihood of a chop classification, although the presence of SFT was not vital for identification. Comminuted fracture bases were negatively associated with chopping, aligning with expectations that blunt impacts are more likely to disperse force broadly and produce shattering.

Logistic regression models that incorporated both direct morphological features and absence features performed slightly better in-sample according to WAIC and LOO scores. This reflects the controlled experimental context, where absence features such as incomplete fractures or unmodified surfaces can meaningfully signal chop impacts. When these features were excluded, the same core set of predictors emerged, and their relative rankings remained stable. This suggests that the relationships between key fracture characteristics and impact type are robust across different observational contexts.

Hierarchical clustering provided further support for this distinction. It revealed a clear separation between the hammer and the axes, while showing little differentiation among the four ax materials. This indicates that the primary differences reflect the mechanics of chopping vs. bashing, rather than metallurgical variation between ax types. Grouping all ax materials together was therefore justified for the main analysis.

Taken together, these patterns provide a coherent framework for distinguishing between chopping and bashing on long bone shafts. Rather than relying on any single diagnostic criterion, the combination of impact area shape, fracture type, and, when present, SFT provides a statistically supported basis for identifying chop impacts. At the same time, the presence of overlapping traits such as bashes occasionally producing straight fractures or chops lacking surface modification emphasizes that these distinctions are probabilistic rather than absolute, and that some individual specimens may remain ambiguous when considered in isolation. Effective interpretation depends on evaluating multiple variables in concert and considering the mechanical and taphonomic context of each assemblage. See Figure 9 for a schematic analytical workflow illustrating how these criteria can be applied in practice.

Fracture morphology provides some of the clearest distinctions between chopping and bashing on long bone shafts. Among the variables tested, impact location shape and gross fracture type were especially informative. Straight impact locations were strongly associated with chopping, reflecting the concentrated, linear loading of an ax blade. Bashes typically produced irregular impact locations, consistent with the diffuse force of a rounded hammer surface. This pattern matches previous percussion experiments, where irregular outlines are typical of hammerstone impacts (Pickering and Egeland, 2006; Galán et al., 2009; Blasco et al., 2014), and experimental chopping studies that show the linear geometry of ax edges produces distinct straight zones (Okaluk and Greenfield, 2022). Gross fracture type also differentiated the two categories. Comminuted fractures were significantly more common in bashes, reflecting broader force dispersal, whereas chops more frequently produced transverse or oblique fractures.

Fractographic features provide a critical tool for locating impact areas and interpreting fracture mechanics. Mirror, mist, and hackle zones are key indicators of how bending fractures propagate. In bending, the fracture initiates on the tensile surface, then travels across the bone toward the compressive side. This means that the point of crack initiation is opposite the actual impact location. By examining the internal fracture surface, analysts can trace the mirror, mist, and hackle zones backward to locate the impact zone on the opposite cortex. This approach is especially valuable when external surface modifications are absent or ambiguous. These fractographic features form during mid-high-energy dynamic loading events, such as blows or strikes, and are not expected in low-energy taphonomic breaks like trampling or sediment pressure (Christensen et al., 2018, 2021). Thus, these features provide the first indication that the fracture may be anthropogenic and should be examined further.

Incorporating fractography strengthens classification by providing a reliable reference point for evaluating impact location shape and fracture morphology. Surface modifications such as percussion notches, flakes, or pecking were infrequent in this study (pecking in 35%, flakes in 45%, a single confirmed notch), echoing previous findings that these features form inconsistently and vary with tool type and number of blows (Blasco et al., 2014; Galán et al., 2009; Pickering and Egeland, 2006). Because surface traces are unreliable on their own, impact identification should center on internal fracture morphology, with surface features treated as supplementary evidence. These patterns were derived from diaphyseal loading, where bending dominates, and should not be generalized to non-diaphyseal contexts without further testing.

Zooarchaeological and forensic studies have traditionally focused on surface modifications such as notches, flake scars, and pitting to identify percussive activity, but they have rarely examined how these features function in distinguishing chopping from bashing in a systematic way. Percussion notches, in particular, have often been treated as hallmark indicators of hammerstone activity, characterized by scalloped cortical margins and the detachment of an inner flake with conchoidal features (Binford, 1981; White, 1992; Fisher, 1995; Pickering and Egeland, 2006; Galán et al., 2009; Domínguez-Rodrigo et al., 2009; Blasco et al., 2014). Flake scars and pitting have similarly been associated with localized impact forces, and their presence has frequently been used to infer intentional tool use (Blasco et al., 2014; White, 1992; Okaluk and Greenfield, 2022). In this study, however, surface modifications were not statistically significant predictors of impact type. This does not reflect a lack of diagnostic potential, but rather their inconsistent occurrence, even under controlled experimental conditions. Pecking was observed in roughly 35% of impacts, flake scars in 45%, and only a single confirmed percussion notch was recorded, produced by a bash. These frequencies mirror patterns observed in experimental percussion studies, which show that notches and related features form unpredictably and vary with tool type, carcass size, and number of blows (Pickering and Egeland, 2006; Galán et al., 2009; Blasco et al., 2014). Because these features occur irregularly, many chop and bash impacts leave no surface modification, even though every impact necessarily produces a fracture and an impact zone. Over-reliance on surface modifications would therefore lead to significant underrepresentation of butchery, since their absence does not imply that no impact occurred.

A deep, incomplete chop mark created by the copper ax (Bone 77) illustrates the interpretive challenges of surface modifications (Figure 10). Unlike most chops in the sample, which produced bending fractures with distinct morphologies, this impact failed in compression and did not propagate through the bone. The blow left a pronounced V-shaped indentation with scalloped cortical margins along a straight impact edge. These scallops resemble percussion notches but were classified as unconfirmed because no interior flake detachment was visible as the inner surface was not exposed. Morphologically, however, they closely mimic hammerstone percussion notches. Had the fracture propagated, these features could have easily been misidentified as hammerstone percussion marks, demonstrating that although uncommon, chop impacts can sometimes resemble bashes when they fail in compression. Crucially, the scallops occurred along a straight impact edge, meaning that the two diagnostic criteria, impact location shape and surface modification, can appear together but be misread if considered separately. In this case, impact location shape provides the clearer indicator of impact type, matching the statistical findings that straight impact edges are strong predictors of chopping. These observations show that surface modifications have diagnostic value when present but should be interpreted alongside fracture morphology and impact zone identification. This combined approach reduces equifinality and improves classification confidence.

Although the analytical framework developed here was designed for zooarchaeological contexts, it also draws on and contributes to methods established in forensic fracture analysis. Both fields confront similar challenges in distinguishing chop impacts from blunt-force trauma, particularly when sharp force traits are absent. Forensic research has long acknowledged that the ax trauma often blurs the line between sharp and blunt force trauma and presents significant challenges for weapon identification (Humphrey and Hutchinson, 2001; de Gruchy and Rogers, 2002; Lynn and Fairgrieve, 2009; McGehee et al., 2023). This overlap in diagnostic challenges reflects the shared mechanical principles that govern how bone responds to dynamic, percussive loading. The results of this study show that consistent, fracture-based criteria can help distinguish between chopping and blunt-force impacts. This alignment underscores the value of shared experimental and analytical approaches in advancing both archaeological and forensic interpretations of bone trauma.