The driving force behind tool-stone selection in the African Middle Stone Age

Significance

Stone tool making requires knowledge and skill. Ancient stone tools allow studying evolutionary processes and settlement patterns, such as complex cognition, migration, or land use. However, the choice of raw materials for making tools has largely relied on interpretation. Here, we introduce a physical model allowing to understand and quantify the quality of different stones for tool making and use. Based on the rocks available to Stone Age foragers living at the South African site Diepkloof Rock Shelter, we demonstrate that the selection of specific rocks for making specific tools was driven by precise criteria. Our findings pinpoint a deep understanding of the physical properties of materials in the African Middle Stone Age.

Abstract

In the Stone Age, the collection of specific rocks was the first step in tool making. Very little is known about the choices made during tool-stone acquisition. Were choices governed by the knowledge of, and need for, specific properties of stones? Or were the collected raw materials a mere by-product of the way people moved through the landscape? We investigate these questions in the Middle Stone Age (MSA) of southern Africa, analyzing the mechanical properties of tool-stones used at the site Diepkloof Rock Shelter. To understand knapping quality, we measure flaking predictability and introduce a physical model that allows calculating the relative force necessary to produce flakes from different rocks. To evaluate their quality as finished tools, we investigate their resistance during repeated use activities (scraping or cutting) and their strength during projectile impacts. Our findings explain tool-stone selection in two emblematic periods of the MSA, the Still Bay and Howiesons Poort, as being the result of a deep understanding of these mechanical properties. In both cases, people chose those rocks, among many others, that allowed the most advantageous trade-off between anticipated properties of finished tools and the ease of acquiring rocks and producing tools. The implications are an understanding of African MSA toolmakers as engineers who carefully weighed their choices taking into account workability and the quality of the tools they made.

The study of raw materials for stone tool making has important implications for our understanding of the economy (1), ecology (2), and mobility (3) of past people. Binford (4), for example, had argued that foraging hunter–gatherers may have “embedded” their raw material acquisition into other activities, such as food-getting, instead of organizing specific raw material procurement trips. Others (e.g., ref. 5) suggest that tool-stone acquisition was subordinate to seasonal rounds of hunter–gatherers. A more moderate understanding might be that people introduced modifications in their food-related mobility pattern to include specific tool-stone acquisition activities. If so, the variability of collected material types (i.e., which rocks were collected and how variable they were) would at least partly be a by-product of the group mobility patterns. This view was perhaps best expressed in Brantingham’s (6) neutral model that predicts archaeological raw material patterns not to be the results of strategic selection at all, but rather to reflect raw material abundances in the area the foraging group happened to be active in. Alternative, non-neutral approaches (e.g., refs. 7 and 8) propose that it was the need for raw materials with certain qualities that led to active selection for these properties. In this case, the tool makers would have had detailed knowledge of the raw materials’ different properties for tool making, being able to link these to, and select these for specific types of tools they intended to make. The implications of such a model of targeted raw material selection are highly relevant in the context of the cultural evolution of early Homo sapiens in Africa. During the past ~200 ka years, the South African subcontinent has witnessed the advent of major innovations (9), technological changes (e.g., refs. 10–12), and population turnovers (13). New traits linked to what some authors have called behavioral modernity (e.g., ref. 14) or behavioral variability (15), include abstract expressions through art and pigment use (16–19), composite tools made by combining different material classes (20, 21) or complex and analogous reasoning (22). The successive appearance of these traits over the last part of the Middle Stone Age (MSA) was accompanied by changes in the stone toolkit. Such changes have been interpreted to reflect socioeconomic and cultural factors (23), changing subsistence patterns (24), or responses to climatic variations (25). What is certain is that changing stone tool production strategies were accompanied by the use of new and more diverse stone raw materials. But what were the reasons for these frequency shifts in the raw material spectrum? Setting Brantingham’s (6) neutral model as null hypothesis, Oestmo et al. (26) tested whether the raw material spectrum at one South African MSA site can be described as neutral, or is better described as a specific selection pattern. They found that the range of tool-stone types cannot be explained by raw material availability alone, suggesting that the foragers of the later part of the South African MSA selected specific raw materials for tool making. The reasons for this selection, however, remain unclear. One hypothesis are novel knapping and tool-stone property requirements. Did newly emerging tool forms require different fracture properties of the stone raw materials? Alternatively, or in addition, is there a relationship between tool-stones and the way these tools were used? In other words, do specific types of use, like cutting, scraping or projectile use, require specific material properties of the finished tools? If so, can these new tool forms and/or usages be meaningfully linked to the specific tool-stones selected at the time? We attempt to answer these questions by understanding the quality of different tool-stones that were used in the MSA of Diepkloof Rock Shelter (henceforth only Diepkloof) on South Africa’s west coast. For this, we establish a quantitative framework for comparing different stones in terms of their advantages and challenges for toolmakers and users.

In terms of stone working, this equals to the concept of knapping quality, or knapping properties, where one of the most important variables appears to be the force needed to detach flakes. This is so because the force applied by a knapper is limited by physical constraints and, additionally, knapping blows are more difficult to control when higher force is to be applied (by greater hammer stone mass or faster blows; this is especially true in free-hand percussion). Such knapping force requirements can be, and have been, tested for in real-world knapping experiments (e.g., ref. 27), but a confounding factor is that the constantly varying core morphologies during knapping induce supplementary constraints, potentially blurring our understanding of the tool-stones’ mechanical properties [see, for example, the criticism of ref. 27 by Luedtke (28)]. Another means of investigating knapping force is based on understanding the energy transitions involved in crack formation and propagation (29, 30). Here, we chose such an approach, rather than actual knapping experiments, because it provides quantitative data that can be compared more objectively. Another variable associated with knapping quality is the predictability of fracture behaviors. While fracture predictability has been investigated by visually evaluating clasts and other heterogeneities, e.g., establishing an “impurity encounter rate” (31–33), or indirectly estimated based on other physical measurements (34), we argue that an experimentally better constraint treatment of fracture predictability is based on the theory of Weibull (35). This is a statistical approach, based on actual mechanical tests, allowing to measure how often materials fracture under similar conditions (see, for example, the use of Weibull statistics in ref. 36). We therefore combine mechanical measurements of fracture force and predictability to compare different stones in terms of their knapping quality.

A few works have also investigated the quality of tool-stones for different use types. Some authors conducted actualistic use-experiments combined with automated cutting tests (37), others fully automated cutting and scraping tests (38) and still others employed standard engineering tests investigating the resistance of different materials to abrasion (32). The latter provides standardized values by which different tool-stones can be compared. We therefore measure abrasion resistance to make statements on tool-stone quality for repeated used types like cutting or scraping. Similar works on the formation of impact fractures found that mechanical properties, similar to those involved in stone knapping, allow to make predictions on the quality of different stones as projectiles [for example, strength (39)]. Thus, investigating knapping force also allows understanding the suitability of different tool-stones to be used as projectiles, although the relationship is expected to be inversely proportional. In this study, we use these measures to compare Diepkloof tool-stones. Our comparison should allow to understand tool production quality and how the stones behave during use as finished tools.

Results

The most important raw materials used at Diepkloof are hydrothermal quartz (this is a white polycrystalline joint infill of ~1/2-cm measuring quartz crystals), quartzite, hornfels, and silcrete (40). We therefore collected eight samples of these rocks (Fig. 1 and Table 1). We chose three quartzite samples to represent the variety of grain sizes observed on Diepkloof quartzite artifacts. One of these quartzite samples (WK-21-07) was matched to a raw material class in the assemblage (18.9% of all lithics), ascertaining the correspondence with artifact raw material (petrographic and mineralogical analyses are detailed in supplementary information, SI Appendix, Figs. S1–S5). Two silcrete samples were chosen to reflect the grain sizes observed on silcrete artifacts. One sample (WK-13-13) was petrographically and mineralogically matched to an artifact raw material class (which represents 6.3% of the assemblage) and the other silcrete (WK-13-08) was found to represent fine-grained silcrete tools (34.5% of all lithics) in terms of mineralogical content, cementing and microfacies. Because silcrete was frequently heat-treated (HT) in the MSA and at Diepkloof in particular (41), we analyzed these two silcretes unheated and HT (for 2 h at 450 °C as in ref. 42). The eight samples include local rocks (quartzite of the rock shelter wall; the closest silcrete outcrop to the site) and rocks sampled more than 70 km away from the site (quartzite from the Olifants River mouth and fine-grained silcrete from south of Diepkloof). Together, the four rock types directly analyzed for correspondence with Diepkloof lithics (cf. SI Appendix) represent 63.1% of all materials knapped at the site. Admitting that there are no differences between samples of hydrothermal quartz that could be linked to specific outcrops, our analysis is representative of 99% of the Diepkloof assemblage (as 35.9% of the assemblage is made of such hydrothermal quartz). The only rocks knapped at Diepkloof for which we could not include geological reference materials are chert and fine-/medium-grained quartzite (our fine-grained quartzite sample WK-21-09 and the medium-grained sample CB-18-06 must be regarded as an approximation of the material used at Diepkloof). Both rock types account for approximately 1% of the assemblage.

Fig. 1.

Different raw material types analyzed mechanically. (A) Quartzite of Diepkloof rock shelter, sample WK-21-07, (B) hornfels from Olifants river mouth, sample WK-21-10, (C) hydrothermal quartz, same location, sample WK-21-08, (D) fine-grained quartzite, same location, sample WK-21-09, (E) silcrete blocks at outcrop, sample WK-13-13, and (F) silcrete blocks at outcrop, sample WK-13-08. Photos (B and C) are shown to the same scale (below C); A is shown to a different scale (above A).

Table 1.

List of samples, descriptions, and predicted knapping force (F_ac) for samples with a volume of 30 mm³ (for an explanation of this volume, see the main text)

Sample N°	Rock	Description	Fac in 30 mm3 [N]	Specific wear rate [mm3/Nm]
WK-21-08	Hydrothermal quartz	Rolled white cobble. Sampled on a terrace of the Olifants River, near river mouth (31°44’58”S 18°13’54”E). Distance to Diepkloof: ~73 km.	229	3.32 × 10−3
WK-21-09	Quartzite	Rolled cobble. Crystal size cement <0.1 mm. Same position as WK-21-08.	284	5.02 × 10−3
WK-21-07	Quartzite	Block from the wall of Diepkloof rock shelter. Crystal size cement ≈ 0.3 mm.	277	4.20 × 10−3
CB-18-06	Quartzite	Rolled cobble. Crystal size cement ≈ 0.15 mm. Sampled in the Cederberg region from a riverbed (near 32°14’35”S 19°00’36”E). Distance to Diepkloof: ~62 km.	450	4.48 × 10−3
WK-21-10	Hornfels	Rolled cobble. No grain visible at 10× magnification. Same position as WK-21-08.	12	7.61 × 10−3
CB-18-04	Hornfels	Rolled cobble. No grain visible at 10× magnification. Same position as CB-18-06.	43	7.35 × 10−3
WK-13-08	Silcrete	Floating texture. Clasts: 32 vol% with average size 0.2 mm. Crystal size cement >5 μm (sampled in secondary position 33°15’05”S 18°34’57”E). Distance to Diepkloof: ~97 km.	100 HT = 15	3.37 × 10−3 HT = 3.32 × 10−3
WK-13-13	Silcrete	Grain-supported structure. Clasts: 66 vol% with average size 0.92 mm. Crystal size cement ≈ 12 μm (sampled at outcrop 32°27’37”S 18°31’01”E). Distance to Diepkloof: ~10 km.	60 HT = 116	4.45 × 10−3 HT = 3.81 × 10−3

For silcrete samples, F_ac and specific wear rates are first reported for unheated silcrete, followed by values of the same silcrete HT at 450 °C.

To understand the quality of these rocks for tool making and projectiles use, we measured the following properties using standard mechanical tests. We experimentally determine fracture toughness K_Ic by Vickers indentation [in this case, K_Ic = indentation fracture resistance (43)], elastic (or Young’s) modulus E with resonance-frequency-damping analysis (RFDA), strength σ by four-point-bending experiments, and variability of fracture (measured as Weibull modulus m) based on four-point-bending data combined with Weibull statistics (35). The resulting values express the strength, with which these eight rock samples resist fracture initiation (σ), the resistance they oppose to crack opening during fracture propagation (K_Ic), and their homogeneity in terms of expected fracture behavior (or their fracture predictability, m). The results of these mechanical tests are summarized in Table 2, all raw data can be found in SI Appendix.

Table 2.

Mechanical properties as experimentally determined on the eight rock samples

Sample N°	Density [g/cm3]	E [Gpa]	Mean c [µm]	KIc [MPa* √m]	σf [MPa]	m	V0 [mm3]	σ0 [MPa]
WK-21-08 (Qrz.)	2.55	84.45	280	1.46	25	4.3	159	25
WK-21-09 (Qrzte.)	2.60	82.36	220	2.02	54	20	27	54
WK-21-07 (Qrzte.)	2.51	68.62	293	1.24	23	6.1	111	23
CB-18-06 (Qrzte.)	2.50	55.88	254	1.43	29	9.7	58	29
WK-21-10 (Hfls.)	2.64	80.52	272	2.53	89	2.9	233	89
CB-18-04 (Hfls.)	2.62	77.53	271	2.51	88	4.2	104	88
WK-13-08 (Silc.)	2.58	67.42	233	1.8	58	5	61	58
WK-13-08 HT: 450 °C	2.58	63.83	303	1.14	63	7	61	63
WK-13-13 (Silc.)	2.65	76.07	254	1.67	38	3.1	208	38
WK-13-13 HT: 450 °C	2.65	59.92	276	1.31	35	9.2	206	35

K_Ic was calculated from the mean crack length c. σ_f is the strength of the bending bars. Weibull modulus m and characteristic strength σ₀ were determined from Weibull plots and V₀ is the volume for which σ₀ is valid. Qrz. = hydrothermal quartz, Qrzte. = quartzite, Hfls. = hornfels, Silc. = silcrete.

However, these values alone do not allow understanding the force required to make knapping flakes or to cause impact fractures in projectiles. It has recently been shown (30) that they must be combined in a mechanical model to make such predictions. The method is based on a graphical approach, reporting the length of cracks actually produced in a material, compared with the minimal length requirement for a crack to become critical (such a plot is shown in SI Appendix, Fig. S6). A crack is said to be critical if it is long enough to run through the entire volume of the material, requiring no more energy for crack propagation [the crack fulfils the Griffith criterion (44), i.e., further opening decreased elastic energy in the material, in our case: flaking]. This model is based on the assumption that cracks produced in a material by indentation with a pointed pyramid behave similarly to cracks produced during knapping. This is probably not true in terms of the absolute magnitude of the predicted force. The stress field under a hammer stone impact in knapping is related to Hertzian fracture mechanics (ball contact), which is different from the point-like contact of a pyramid (i.e., the force value obtained by modeling indentation cracks must be considered an approximation of the real knapping force). However, the relative magnitude of the force predicted for different tool-stones can be expected to be similar (as compared to Hertzian cone cracks), allowing a valid comparison between samples. We therefore take Nickel and Schmidt’s (30) method as a basis for our force prediction, introducing a compact formulation that facilitates its applicability to archaeological problems. The derivation of this formulation is given in supplementary information (SI Appendix, Eqs. S1–S9). The force needed to obtain a critical crack F_ac in a tool-stone can be estimated by

$$F_{\mathit{ac}}=5.804\ast {10}^6 \frac{{K_{\mathit{Ic}}}^4{\left(\frac{\mathit{HV}}{E}\right)}^{2/5}}{{{\sigma }_1}^3}.$$ [1]

Fracture toughness K_Ic, Vickers hardness HV and Young’s modulus E are material properties that are independent of a sample’s volume. Strength σ₁ depends on sample volume. Thus, σ₁ in Eq. 1 is the strength of a specific volume and F_ac is the force needed to create a crack in this volume. For the treatment of different volumes, σ₁ of a volume V₁ can be obtained by

$${\sigma }_1={\sigma }_0\sqrt[m]{\frac{V_0}{V_1}},$$ [2]

where σ₀ is the experimentally measured strength (by four-point bending) of volume V₀ and m the sample’s Weibull modulus (SI Appendix, Eqs. S10–S12). Thus, Eqs. 1 and 2 predict the force necessary to make knapping flakes, or to induce impact fracturing, in rock samples with different volumes. For better comparability, we report F_ac for a standard volume of 30 mm³ (following ref. 30) in Table 1 but show its volume dependence in Fig. 2. It is important to note that V₁ is not the volume of the entire core (knapping) or the whole projectile (impact fracture formation). V₁ is the volume within which the stress field, which ultimately leads to a critical crack, is building up and is connected to the probability of distribution of the sizes of existing faults as measured by the Weibull modulus. In knapping, this is immediately under the impact point. The exact size of the stress field is constrained by work-piece morphology, flaking platform depth, and striking angle, i.e., the geometry of the core and flaking impact, but it cannot be directly understood through the volume of the finished knapping flake or blade. In impact fracture formation, it is driven by point morphology, such as tip cross-sectional area, but also by impact angle.

Fig. 2.

Volume dependence of the force F_ac predicted to reach critical crack length a_c. (A) Comparison of hydrothermal quartz and different types of quartzite; (B) Two hornfels samples; (C and D) different silcrete types, unheated and heat-treated. Note that ordinate scales are different in the four plots for better visibility of the curvature of the functions.

Comparing F_ac of the eight South African tool-stones allows understanding some parameters related to the ease of knapping and, on the other hand, highlights their potential weakness to failure if they are used as projectiles (raw data in SI Appendix, Tables S2 and S3). In 30 mm³, hornfels and the HT fine-grained silcrete WK-13-08 fracture at forces ~20× lower than quartzite and 15× lower than hydrothermal quartz. Both unheated silcretes still require less than half the force of hydrothermal quartz and about three times less force than quartzite. The exact volume dependence of F_ac (Fig. 2) shows a more complicated trend. At low volumes, the relation between some samples is inversed compared to larger volumes. This is so because calculating σ₁ with Eq. 2 contains the Weibull modulus m as root degree. The volume dependence of F_ac is therefore not linear, with m ≲ 3 yielding steep increases of F_ac in larger volumes and m ≳ 3 causing F_ac to plateau out with increasing volume.

Fracture predictability of the eight South African tool-stones, as expressed by their Weibull modulus m, is listed in Table 2 (SI Appendix, Fig. S7). Quartzite has a significantly more homogeneous fracture behavior than all other rock types (as expressed by its m, which is at least 30% higher than that of all other rocks, with the exception of HT silcrete). The fine-grained quartzite WK-21-09 even has a value of m more than double that of all other rocks, pinpointing excellent predictability of fracture during knapping. For quartz-bearing rocks, the variation of m might be in relation with the samples’ grain size. Hornfels lies in the lower range of values obtained in this study, suggesting that its fracture is less predictable during knapping. Heat treatment of silcrete allows improving its fracture predictability by at least 30%, adding to its improvement already achieved by lowering required knapping force.

To obtain data on the use-related properties of the eight South African tool-stones, we measured their resistance to abrasion using pin-on-disc tribometry (SI Appendix, Fig. S8). These measurements provide a complementary use-related dataset allowing to predict the durability of tools during repeated use activities like cutting and scraping. These abrasion resistance values are summarized in Table 1 and graphically compared in Fig. 3 (SI Appendix, Table S4 and Fig. S9). Hornfels is least resistant to abrasion, with a specific wear rate almost double that of all other analyzed rocks. All other tool-stones contain quartz as their main mineral phase and have specific wear rates more similar to each other. Hydrothermal quartz and fine-grained silcrete (WK-13-08) are the most resistant tool-stones to abrasion. Coarse-grained silcrete has a specific wear rate similar to quartzite samples and heat treatment allows improving its resistance against abrasion slightly. In quartzites, a grain-size dependence of the specific wear rate is observed, the finest-grained samples (WK-21-09) being least resistant and the coarsest samples (WK-21-07) most resistant.

Fig. 3.

Mean specific wear rate of the eight tool-stones, as measured by the pin-on-disc abrasion test.

Discussion

The Quality of Our Data and Comparability.

The perhaps greatest source of uncertainty in the raw data we produced on Diepkloof raw materials is the use of indentation fracture resistance instead of real fracture toughness. We do this because the direct experimental determination of fracture toughness, for example, using single-edge notch beams (SENB), is relatively elaborate and cost intensive (e.g., ref. 45). Our approach provides a more practicable way of producing raw data for our physical model. In indentation fracture resistance measurements, the exact nature of obtained cracks is disputed (see, for example, discussion in ref. 46). However, Lawn’s (47) analysis of the indentation crack length as a function of the resistance against crack elongation is valid. Thus, our measurement of toughness through indentation is not identical to toughness values measured by more traditional methods (using SENB), but it is internally consistent. This means that values produced using the same methodology (e.g., different values produced in this study) are comparable. Further, studies using the same methodology but different sets of samples from different geographic spaces (e.g., refs. 30, 36, and 48) are also comparable and their data can be used to make predictions of knapping quality using our model.

Raw Materials at Diepkloof.

Raw material spectra in different stratigraphic units at Diepkloof (as taken from ref. 40) are shown in Fig. 4. Hornfels accounts for the least represented tool-stone in almost all units throughout this sequence, despite its very low knapping force requirement (hornfels requires some of the lowest knapping forces of all raw materials). The apparent preference for tools-stones other than hornfels may be explained by the need for tools that are resistant to abrasion, in other words, for raw materials that make durable tools [this has previously been suggested (49)]. This is true for repeated use types (scraping, cutting), where higher wear rates are expected to lead to rapid dulling, and for projectile use, where the low force at which fracture is initiated would cause projectiles to break more easily. The selection might also have been influenced by fracture predictability. Hornfels has low predictability of fracture and therefore does not allow predicting the outcome in a knapping situation. Our mechanical analysis also allows making statements that help understanding specific technocomplexes in the Diepkloof sequence. However, to understand raw material selection in technocomplexes, like in the Still Bay and Howiesons Poort, it is necessary to first work out what the implications of the volume dependence of F_ac really are (Fig. 2).

Fig. 4.

Raw material distributions in Diepkloof Rock Shelter (squares M6-N6). Percentages of blades/bladelets and bifacial products were calculated to the base of all pieces in the lithic assemblage. Bifacial products include the bifacials themselves and their shaping flakes. Data recalculated from (40), total lithic assemblage = 14,333 (only artifacts > 20 mm).

Implications of the Volume Dependence of F_ac for Stone Knapping.

Overall, rocks that fracture at relatively low forces during knapping can be expected to also fracture at lower impact forces during projectile use (points), imposing an inverse relationship between perceived good knapping quality and durability in projectile use. However, the exact volume dependence of this flaking force makes some of the rocks slightly better in specific situations (knapping or impact resistant) than expected/calculated. The implications of this volume dependence are best illustrated by the two curves of silcrete WK-13-13 in Fig. 2C. In this sample, edge chipping in a small volume (e.g., for retouching stone tools) would require more force if this silcrete is HT than in its unheated state. For this specific application (retouch, shaping), it is not advantageous to heat-treat the tool-stone. In a similar fashion, when that same silcrete is subjected to (near) frontal impact in a point, the stress field dissipates in a larger volume (>80 mm³) and, consequently, heat treatment would provide no benefit because it reduces the force needed to induce fracture (in other words, the point becomes less resistant to damage). However, such a volume dependence of the fracturing force likely also has consequences for stone knapping because different knapping techniques may cause the stress field, which ultimately leads to flake detachment, to be dissipated in different volumes (compare Fig. 4 A–C). It might therefore be useful to heat-treat the coarse-grained silcrete WK-13-13 to produce larger blanks. A similar phenomenon can be observed for different types of quartzite. Quartzite WK-21-07, coming from the rock shelter wall of Diepkloof, is likely the best choice for retouch or final shaping, where effective volumes are small due to (freehand, or pressure) flaking near the edge of tools. The important factor limiting this volume is striking angle (compare Fig. 5 A and B). The fine-grained quartzite WK-21-09, from 70 km further north of Diepkloof, allows low fracturing forces in large volumes (in other words, points are less resistant to damage) but would impose significantly larger forces than the coarser-grained WK-21-07 in near-edge flaking conditions (in other words, more force is needed for retouching and shaping of WK-21-09). Thus, the selection of specific tool-stones, or the choice to heat-treat silcrete or not, may be guided by the knapping strategy or the intended end product.

Fig. 5.

Hypothetical volumes of the stress field created during impact of a projectile (A and B) and knapping (C). Black arrows are the directions of the applied force, stress fields are shown in red, and broken lines indicate crack formation. (A) Near-axial application of stress to a Still Bay point in a head-on impact during projectile use. (B) Stress field in a near-edge chipping situation applied during bifacial thinning. Note that the volumes of these stress fields in both samples of the same morphology are different, the near-axial application of force allowing stress to be dissipated in a larger volume. (C) Stress field created during knapping of blades (schematic, not to scale). Note that its extension is limited by core morphology toward the edge of the core and weakens out away from the point of impact.

Tool-Stone Selection in the Still Bay.

The Still Bay is a South African technocomplex directly preceding the Howiesons Poort. It is typified by the production of several-centimeter-long and relatively thin (≤10 mm) bifacial points said to be characterized by a lenticular cross-section, a foliate to lanceolate morphology, and a semicircular to wide-angled base (50). The so-called Still Bay points, yet recorded in a handful number of archaeological sites in South Africa, document the application of pressure-flaking in their last stage of shaping (51). Current studies support their use as projectiles and knifes (52).

In stratigraphic units Keeno to Leo, which yielded the Still Bay occupation at Diepkloof (Fig. 4), coarse-grained quartzite from the shelter wall (WK-21-07) represents 65% of the used raw materials (all raw material percentages here are recalculated from the data in ref. 40). In bifacial products (Still Bay points and their shaping flakes), this quartzite accounts for 71%. Coarse-grained local silcrete (WK-13-13, called YB-silcrete in ref. 40) is slightly overrepresented in lithics associated with bifacial shaping (11% of the bifacial shaping products are made from this silcrete vs. 8% of all lithics). Hydrothermal quartz and fine-grained silcrete were less extensively used for bifacials than in the overall assemblage (Hydrothermal quartz: 2% of bifacial products vs. 7% in the assemblage; fine-grained silcrete: 14% of bifacial products vs. 16% in the assemblage). Thus, the Still Bay occupants of Diepkloof preferred to make points from coarse-grained quartzite coming from the site’s shelter wall. Coarse-grained silcrete was also readily used. At first glance, this is surprising because quartzite requires the highest knapping force of all Diepkloof raw materials, and therefore allows the least control of knapping blows. Coarse-grained silcrete also requires higher knapping force than finer varieties and, additionally, has the lowest fracture predictability of all analyzed quartz-based rocks. Great knapping forces, and therefore poor control over the knapping blows, appear to be in contradiction with the interpretation that Still Bay points required considerable skill and investment to be crafted [(53), also requiring more precise knapping strategies such as pressure flaking (51)]. High-investment knapping techniques, such as bifacial shaping with pressure, are often interpreted as requiring raw materials with good knapping quality (e.g., ref. 54). The apparent contradiction is resolved by considering the volume dependence of knapping force. Among all quartzites available at Diepkloof, the coarse-grained rock shelter quartzite requires the lowest knapping force to make knapping flakes in small volumes. During bifacial shaping in the last stage of the production of Still Bay points, knapping force is typically applied near edge, where stress builds up in a small cube, measuring a few millimeters length (Fig. 5B). For example, in a volume of 10 mm³, the rockshelter quartzite requires only half the force that would be necessary to produce a flake in the next finer-grained quartzite. However, if knapping quality were the only selection criterion, silcrete would be a far superior choice for making Still Bay points. Quartzite requires more than double the force to produce shaping flakes than silcrete. Additionally, the initial stages of making a Still Bay point imply the detachment of large flakes (including for the blank) where effective volumes, in which the stress field builds up during a knapping blow, can also be expected to be larger. The apparent preference for quartzite over silcrete might be explained by the potential use type of the finished Still Bay point [projectile use has been suggested for other Still Bay assemblages (10)]. During projectile use, a head-on impact at low angles with respect to the point’s long axis builds up tension in a relatively larger volume behind the tip (Fig. 5A). A rock shelter quartzite volume of 100 mm³ withstands more than 2.5× the impact force tolerated by silcrete. If silcrete is used to make projectiles (a fourth of all bifacial products are made of silcrete), it appears to be preferable to either use coarse-grained silcrete or finer-grained silcrete in its unheated state (both withstand similar forces at 100 mm³, coming second to quartzite). And indeed, it has been suggested that the prevalence of silcrete HT decreased in the Still Bay compared to before and after this technocomplex (55). Thus, the relatively good fracture predictability of quartzite, its robustness (when used as projectiles), and good resistance against abrasion (which is required for scraping or cutting) make Still Bay points ideal multipurpose tools. The material choice additionally had the advantage of relatively good knapping quality for the specific requirements of bifacial shaping (i.e., fractures develop in small volumes). Thus, choosing these materials represented an advantage for making Still Bay bifacials, even if they were to be used for purposes other than projectile. Although quartzite and coarse-grained silcrete are locally available, their selection to manufacture a tool type for which they apparently present the best of all properties is pointing toward a good understanding of the mechanics of different tool-stones and a carefully considered raw material acquisition scheme in the Still Bay.

Tool-Stone Selection in the Howiesons Poort.

The Howiesons Poort is a more recent technocomplex, immediately following the Still Bay at Diepkloof. It is characterized by the abundant production of small blades and bladelets and by the manufacture of a broad range of backed pieces. Those backed tools are associated with lateral or oblique hafting, cutting (56), and projectile use (57, 58). The variation of backed pieces, in terms of frequency and morphologies, correlates with broader technological and typological changes that document the existence of different stages during the Howiesons Poort. At Diepkloof, stratigraphic units Jeff to Fred yielded the intermediate Howiesons Poort, a technocomplex closely resembling the Howiesons Poort found at other South African sites (compare, for example, with ref. 59). Raw materials used in the intermediate Howiesons Poort were dominated by fine-grained silcrete (accounting for 59% of the assemblage and 69% of all blades/bladelets). Rockshelter wall quartzite (12%), hydrothermal quartz (16%) and coarse-grained local silcrete (10%) are also represented. More than 93% of all silcrete was HT in this phase (41). It therefore seems that Howiesons Poort toolmakers preferred good knapping quality and tools that are highly durable in repeated use situations: HT fine silcrete yielded the lowest of all knapping forces in volumes >45 mm³, a relatively good fracture predictability similar to that of quartzite, and (together with hydrothermal quartz) the best resistance against abrasion. Low knapping forces provide ideal condition to detach blades and bladelets. The exact volumes in which the stress field builds up during such a knapping blow (Fig. 5C) cannot easily be measured but they might be larger than those effective in edge shaping (Fig. 5B). And indeed, the abundance of blades and bladelets is associated with a preference for silcrete throughout the Diepkloof sequence (note the disappearance of blades/bladelets in layer Jack, correlating with a raw material shift away from silcrete, Fig. 4) and in the Howiesons Poort in particular. Silcrete’s good resistance to abrasion makes these blades suitable tools for cutting and scraping but segments made from these blades are potentially fragile upon impact when they are used as projectiles. In lateral hafting situations, as suggested by archaeological data from other Howiesons Poort sites (60), the fragility of HT silcrete might be less important. Thus, silcrete’s fragility is not necessarily a drawback in the context of Howiesons Poort composite hunting weapons (57, 61). The likelihood that small lateral or axial implements break loose from their haft, potentially worsening injuries in the hunted animals, is higher in such composite tools than for the larger and more solidly hafted Still Bay points (see, for example, ref. 10). The focus in the Howiesons Poort appears to lie on cutting with durable and sharp blades and bladelets, at least the raw material choice can be understood to indicate this. Another notable event is the increasing relative prevalence of hydrothermal quartz at the end of the Howiesons Poort. This raw material shift coincides with an increase of convex backed pieces, lager blanks and decreasing small bladelets (40). Generally, the late Howiesons Poort lithic assemblage points toward lower standardization than what is known from before. The use of higher quantities of quartz is consistent with these technological transitions. Hydrothermal quartz requires relatively high knapping forces while, at the same time, presenting relatively low predictability of fracture. In other words, hydrothermal quartz is of generally low knapping quality. Thus, its increase in the end of the Howiesons Poort reflects the flexibility of Diepkloof stone workers in terms of the supports they needed, while still providing good resistance to abrasion. Thus, raw material selection in the Howiesons Poort also documents a deep understanding of the constraints and advantages provided by the variety of mechanical properties of available tools-stones.

Can Raw Material Procurements Be Explained by Other Factors?

Some of the tool-stones used at Diepkloof were transported over larger distance. The closest point where fine-grained silcrete can be found is the Olifants River ~40 km inland from the site or in the region near Hopefield ~70 km further south (40). Hornfels and fine-grained quartzite can, to our knowledge, be found in the Olifants River and its nearby catchment only (closer terrains that now are submerged might yield these rocks, however). Changing territories or accessibilities to raw material outcrops might impose limits on the availability of these tool-stones (6). In this case, raw material changes would reflect socioeconomic changes instead of the purposeful selection of tool-stones. For Diepkloof, this explanation appears unlikely. Raw materials that were transported over longer distance are present in all stratigraphic units (Fig. 4), even in those where other tool-stones were preferred. Consequently, they were at least occasionally accessible at all times. More importantly, many of the raw materials used at Diepkloof were available in a radius <10 km around the site, and therefore accessible at all times. The transition from the preference for locally available tool-stones, like quartzite from the rockshelter wall, to other materials that must be transported over longer distances, such as fine-grained silcrete, occurs at the beginning of the early Howiesons Poort (Fig. 4). After that moment, there is only one short period (centered around stratigraphic unit Jack) where finer silcrete regresses in the raw material spectrum. Except for the MSA Jack, fine-grained silcrete remains present at the site, suggesting that it was always accessible. Still, a result of the relatively long transport distances of fine-grained silcrete, with respect to other raw materials, might be that it was a rare resource, which was subject to specific acquisition limitations (trade, excursions organized in less accessible territory). This might be a likely scenario if silcrete tools show traces of curated technology (as in ref. 4). The Howiesons Poort lithic system is based on the production and use of raw blades and bladelets, some being modified into backed and truncated tools (40). No evidence of broken backed pieces that would have been reshaped can be observed in the Diepkloof material so that it appears that these elements were not meant to be curated. In the late Howiesons Poort, strangulated notched tools (i.e., blades and elongated blanks modified by a series of notches) might be considered as curated elements but their exact function is yet to be characterized. Recycling can be observed on some of the Diepkloof artifacts, for example, in the form of blanks being modified as scaled pieces/bipolar cores, but it remains unclear whether we deal with recycling or variations in the production steps. Thus, it is not straightforward to argue for silcrete being a rare commodity in the Howiesons Poort, based on curation. As it stands, the selection of the different tool-stones used at Diepkloof is best explained by the technical and functional needs they satisfied. These findings pinpoint the sophisticated knowledge and understanding of the properties of stones in the South African MSA.

Methods

Four-point-bending experiments were performed with an Instron 4502 10kN universal testing machine. Standard bending bars with chamfered edges on their tensional side were loaded with a speed of 1 mm/min until failure (after preloading of 10 N). Upper and lower bearing distances were 15 and 30 mm for silcrete samples but 20 and 40 mm for all other samples. Silcrete samples were bent in a bending module (MB2) made by Kammrath & Weiss GmbH (Dortmund), all other samples were bent using bearings directly mounted in the testing machine. Bending bar dimensions are reported in SI Appendix, Table S1. All single measurements reported therein result from bending bars that broke within the limits of the two upper bearings (roughly in the middle of the bars). Values obtained from bars that broke outside of the bearings were discarded tor further analysis.

Indentation fracture resistance was measured using the same Instron and Vickers diamonds. The applied load was 100 N, hold time was 30 s. Each sample was indented multiple times in a diamond-polished surface to obtain indentation cracks to be statistically representative. K_Ic was calculated from the mean length of these cracks using SI Appendix, Eq. S1. Crack length measurements are summarized in SI Appendix, Table S2.

Young’s modulus was measured in flexural vibration mode using an IMCE (Belgium) RFDA professional unit. The frequency range was 0 to 100 kHz. The calculation of E requires the material’s Poisson’s ratio, which we admitted to be equal to the one of silica glass: 0.17 (62).

Abrasion resistance was measured using a pin-on-disc tribometer. 80-grit SiC sandpaper was used as rotating phase (a new sheet was used for each measurement). The rotational speed of the tribometer was such to obtain a linear speed of 1 m/s under the sample pins, resulting in ~300 m of sandpaper passing under each sample in 5-min measuring time. Load was set to be 10 N. Three individual pins were measured for each sample and their values averaged. Raw data of these tests are summarized in SI Appendix, Table S4.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All raw data are provided in the main text and SI Appendix.