A taphonomic analysis of PTK (Bed I, Olduvai Gorge) and its bearing on the interpretation of the dietary and eco-spatial behaviors of early humans

will increase signi ﬁ cantly our understanding of early human adaptive patterns. Here, we present PTK as another assemblage where faunal resources were acquired by hominins prior to any carnivore, and where stone-tool assisted bulk de ﬂ eshing was carried out. The abundance of juvenile individuals extends our understanding, as in Kanjera (Kenya), about the hunting skills of early Homo sensu lato . The increasing number of sites, where bulk de ﬂ eshing of small and medium-sized carcasses took place is underscoring the importance of meat in the diets of some of the early hominins, and their patterned use of the space for food processing and consumption. The patterning emerging has a profound importance


a b s t r a c t
Here, we present a thorough taphonomic analysis of the 1.84 million-year-old site of Phillip Tobias Korongo (PTK), Bed I, Olduvai Gorge.PTK is one of the new archaeological sites documented on the FLK Zinj paleolandscape, in which FLK 22 level was deposited and covered by Tuff IC.Therefore, PTK is penecontemporary with these sites: FLK Zinj, DS, AMK and AGS.The occurrence of these sites within a thin clay unit of ~20 cm, occupying not only the same vertically discrete stratigraphic unit, but also the same paleosurface, with an exceptional preservation of the archaeological record in its primary depositional locus, constitutes a unique opportunity to explore early hominin behavioral diversity at the most limited geochronological scale possible.The Olduvai Bed I sites have been the core of behavioral modelling for the past half a century, and the newly discovered sites, excavated with 21st century technology, will increase significantly our understanding of early human adaptive patterns.Here, we present PTK as another assemblage where faunal resources were acquired by hominins prior to any carnivore, and where stone-tool assisted bulk defleshing was carried out.The abundance of juvenile individuals extends our understanding, as in Kanjera (Kenya), about the hunting skills of early Homo sensu lato.The increasing number of sites, where bulk defleshing of small and medium-sized carcasses took place is underscoring the importance of meat in the diets of some of the early hominins, and their patterned use of the space for food processing and consumption.The patterning emerging has a profound importance

Introduction
Hominin primary or secondary access to carcasses has important implications for the reconstruction of the behavior of early Homo sensu lato.Early access to carcasses would imply regular meat-eating, and it might have promoted food-sharing and other cooperative behaviors (Isaac, 1978(Isaac, , 1983)).In contrast, secondary access to largely defleshed carcasses would have provided limited surplus resources (Speth, 1989;Blumenschine and Madrigal, 1993;Domínguez-Rodrigo, 1999;Domínguez-Rodrigo et al., 2014a;Gidna et al., 2014), which, in turn, would have prevented food-sharing and, probably, the development of further complex cooperative behaviors.For decades, FLK Zinj (Bed I, Olduvai Gorge) has been used as the main testing background to infer features of early Homo behavior.Ethnographic analogues and experimental studies applied to the analysis of this site have led to different interpretations about primary versus secondary access of hominins to carcasses (e.g., Binford, 1985;Bunn, 1986;Bunn and Kroll, 1986;Blumenschine, 1986;Domínguez-Rodrigo, 1997a;Capaldo, 1998a;Pante et al., 2012: Parkinson, 2018).
Through the use of skeletal part profiles, some researchers have argued that hominins had early access to carcasses, and that they selectively were transporting high-yielding carcass parts (Isaac, 1978(Isaac, , 1983;;Bunn, 1981Bunn, , 1982Bunn, , 1983Bunn, , 1991;;Bunn andKroll, 1986, 1988;Potts, 1988;Bunn and Ezzo, 1993;Oliver, 1994;Rose and Marshall, 1996;Domínguez-Rodrigo, 1997a;Domínguez-Rodrigo and Pickering, 2003;Bunn and Pickering, 2010).Using the same skeletal profiles, other researchers have argued that hominins had secondary access to partially or fully defleshed carcasses (Capaldo, 1995(Capaldo, , 1997)), that hominins scavenged carcasses initially accumulated by other carnivores at kill-sites or dens (Binford, 1981), or that hominins scavenged the brain and marrow-bearing bones from defleshed felid kills and transported them to refuges (Blumenschine, 1986(Blumenschine, , 1991(Blumenschine, , 1995;;Blumenschine et al., 1994;Selvaggio, 1994).In order to overcome the equifinality produced by skeletal part profiles, some researchers focused on the analysis of bone surface modifications and bone breakage patterns.Blumenschine (1988Blumenschine ( , 1995) ) experimentally argued that the frequencies of tooth-marked specimens by limb bone portion could be an indicator of the order of carnivore access to carcass resources.However, more recent research has shown that tooth mark frequencies can be similar in carnivore-to-hominin and hominin-tocarnivore experimental models when felids rather than hyenas are the primary agent in carcass modification (Domínguez-Rodrigo et al., 2007a;Organista et al., 2016).Additionally, cut-marked bones have been interpreted as the result of hominins exploiting whole fleshed carcasses (Bunn and Kroll, 1986), or the outcome of removing marginal scraps of flesh that survive after carnivore consumption (Blumenschine, 1991(Blumenschine, , 1995)).This purported equifinality is methodological and can be overcome by quantifying the frequency and anatomical distribution of cut marks per bone section and element combined (Domínguez-Rodrigo, 1997b;Domínguez-Rodrigo et al., 2007b).Recent Machine Learning (ML) techniques have also been effectively used to solve some of these taphonomic challenges (e.g., Cifuentes-Alcobendas and Domínguez-Rodrigo, 2019;Domínguez-Rodrigo et al., 2020).Recently, we carried out a 3D and deep learning analysis of cut marks on limb bones to the upper level (equivalent to FLK Zinj's level 22A) of FLK Zinj, PTK and level 22B of David's Site (DS), and the results indicate that hominins might have had primary access to carcasses (Domínguez-Rodrigo et al., 2021).However, in this previous work different important aspects, like carcass acquisition strategies or what type of carnivores intervened in the PTK site were not tested.That study was also limited to the analysis of cut marks and excluded all the other taphonomic variables that inform us about site formation processes.
David's Site (DS) is a new site located on the same paleolandscape as FLK Zinj.Recent taphonomic analyses of the DS faunal assemblage indicate that hominins had early access to carcasses (Cobo-S anchez, 2020;Domínguez-Rodrigo et al., 2021).They transported complete or partial carcasses to the site to be processed and potentially consumed collectively (Cobo-S anchez, 2020).PTK, also contemporaneous with DS, shows the same spatial properties (i.e., mono-cluster) as FLK Zinj (Domínguez-Rodrigo and Cobo-S anchez, 2017a).The site presents an exceptional preservation of the archaeological materials and bone cortical surfaces, which offers the possibility of testing previous interpretations about carcass acquisition strategies, and shedding new light on early Homo behavior.It can also contribute to widening the range of behaviors documented at these early sites.The main aim of the present work is to provide new evidence to understand carcass acquisition strategies by early humans and the potential impact of meat consumption in their diets.Additional questions that will be addressed in the present study are: What function did early Oldowan sites have in the socio-economic structure of early humans?Is there any patterning in how early humans used the space in these loci and in the activities that they carried out in them or do they display wide variability?How can we use these early sites to understand better the lifestyle and behaviors of early Homo sensu lato?
2. The PTK site and its archaeofaunal assemblage PTK is located on the north side of the secondary gorge, just upstream from the junction with the main gorge of Olduvai Gorge, Tanzania (Fig. 1).The site is situated 370 m south of the FLK Zinj and 330 m westward of DS (Domínguez-Rodrigo et al., 2017).Its chronology is well constricted by two volcanic tuffs: tuff IB and tuff IC, which have been dated by argon-argon (40Ar/39Ar) to 1.848 ± 0.003 Ma and 1.848 ± 0.008 Ma respectively (Deino, 2012).
In the present study, we focus on the archaeological material that is within the same clay stratum and same paleosurface as FLK Zinj (upper Zinj or level 22A).The Zinj clay consists of a minimum of two clay depositional events, each containing separate archaeological levels formed at different times with mainly similar environmental conditions.Level 22A shows a more hydrologically closed system than level 22B, but both levels were deposited under a very low-energy conditions (Martín-Perea et al., 2019).Each archaeological level is distributed over a maximum vertical depth of 10 cm (most commonly less).The apparent vertical distribution of materials shown in Fig. 1 is due to the tilting of the modern terrain caused by tectonics occurring after site formation.Paleogeographically, PTK had a slightly elevated profile, compared to the adjacent lake margin, and the area was probably well-drained during the wet season, showing runoff shapes, like rills and small depressions (Uribelarrea et al., 2014) that resulted in alkaline soil conditions which optimised bone-surface preservation (Pizarro-Monzo et al., 2021).
From 2012 to 2019, an open excavation consisting of twenty-six adjacent trenches was carried out at PTK unearthing an area of 185 m 2 (Fig. 2).Small hand tools were utilized for the excavation of the fossiliferous levels and all the sediments were systematically sieved using 5 mm and 3 mm screen-meshes.Stone tools and fossil bones !2 cm and identifiable remains <2 cm, were plotted with a laser total station.The humidity of the sediments had affected some of the fossil bones, which presented cracks and recent breakage planes that ultimately resulted in the fragmentation of some specimens as they were removed from the soil.However, when possible, bone specimens risking fracture during excavation were consolidated with Paraloid B72® (5e10% dissolved in acetone).Additionally, to preserve the morphological structure of some fossil remains, it was necessary to apply gauze before their removal.Some specimens show chemical weathering because they were diagenetically exposed to a constant variation of humidity within the soil matrix.These were mainly found in the northeast area of the archaeological site, where Tuff 1C had been affected by erosion.
In this work, we present the results of the analyses on the faunal materials recovered in level 22A (PTK's upper level).This level is composed of a dense accumulation of fossil bones (n ¼ 1514) and stone artefacts.Flakes and flake fragments represent the bulk of the lithic assemblage (n ¼ 497) (Fig. 3).

Age profiles
To establish dental age of faunal remains, two methods were used considering the different taxa and following the approach detailed by Spinage (1967Spinage ( , 1976) ) for waterbuck and gazelle; Talbot and Talbot (1963) and Attwell (1980) for wildebeest; Grobler (1980) for Hipotragini remains; and Loch and Bradley (1998) and Marín et al. (2017) for equids.First, the degree of the eruption of deciduous teeth and the eruption and development of permanent teeth was explored.Second, the degree of tooth wear was estimated.Both processes were explored under the assumption that tooth eruption and wear are similar in the mandible and the maxilla, though it is known that they are slightly different (Spinage, 1967(Spinage, , 1976).Thus, it is possible that the individuals catalogued using mandibles are slightly older than reported here.However, this bias has not affected the determination of age groups in the present study.Table S1 shows the potential ecological lifespan in years for the different species considered in this analysis and Table S2 shows the material that was used to determine the minimum number of individuals (MNI).We examined 35 isolated dental pieces and 33 teeth anchored in their mandibular or maxillary alveoli to determine the mortality profiles of the bovids (n ¼ 64) and one equid (n ¼ 4) recovered from level 22A.
The age profiles were analyzed by means of several methods, including ternary diagrams (Stiner, 1990), Principal Component Analysis (PCA), and Canonical Variate Analysis (CVA), which were used to compare PTK mortality profiles with age profiles derived from studies with carnivores -lion, hyena, leopard, wild dog- (Bunn and Pickering, 2010), modern foraging ethnographic samples -Hadza and Kua- (Bunn and Gurtov, 2014), and archaeological assemblages -FLK Zinj, DS and Kanjera-(Cobo-S anchez, 2020; Oliver et al., 2019).Individuals were also assigned to five age groups based on their potential ecological longevity (PEL) (Bunn and Pickering, 2010), but given that the MNI obtained from teeth differs from postcranial MNI, we sorted the age groups into three: juvenile, prime adult and old adult.The different corners of the triangular diagram represent 100% of each age category.Between these corners, two zones indicate the attritional (left side) and catastrophic (right side) profiles (although see Discamps and Costamagno, 2015, because their mathematical simulations identified potential problems in the zoning of ternary diagrams).The attritional profile is represented by very young and old individuals, while prime adults are relatively rare.This pattern has been associated with cursorial predators such as cheetahs, wild dogs and hyaenas (Stiner, 1990).The catastrophic or living profile is represented mainly by prime-adult individuals with progressively fewer individuals belonging to other age classes, emulating the structure of a living population (Klein, 1982).This pattern is associated by Stiner (1990) with ambush predators.The software uses maximum likelihood statistics to produce a density contour that is sensitive to sample size and approximated to the 95% confidence interval (Steele and Weaver, 2002).
The second method is a combination of Principal Component Analyses (PCA) and Canonical Variate Analyses (CVA), which reduce the dimensionality of the original data set, generating new variables, namely principal components and canonical variates, which provide a simplified version of the sample variance by the generation of fewer but highly explanatory scores at the expense of minimal accuracy loss.Main differences between PCA and CVA techniques stem from the a priori knowledge required by CVA on group membership, in order to be able to minimize intragroup variation while maximizing inter-group variation, which results in a higher degree of separation among groups when compared to PCA analyses.Canonical variables are defined by square distances between the means of the groups obtained by Mahalanobis's D 2 .PCA and CVA analyses were displayed with a 95% confidence interval ellipse.To homogenize the available data and to make statistically valid assumptions samples were bootstrapped 100 times.Multivariate analyses and bootstrapping were performed in R (R Core Team, 2020) using the R library "BiplotGUI" (La Grange et al., 2009).

Skeletal profiles
In the present study, the standard zooarchaeological units to determine skeletal part abundance have been used: the Number of Identifiable Specimens (NISP), the Minimum Number of Elements (MNE), the Minimum Number of Individuals (MNI), the Minimum Animal Units (MAU), and its relative frequency (%MAU) (see Lyman and Lyman, 1994).
Due to the high bone fragmentation observed in PTK and the virtual absence of epiphyseal specimens, which allow determining taxa, skeletal part profiles have been analyzed by carcass size according to Bunn's (1982) classification.Small refers to sizes 1 and 2 (up to 120 kg), medium-sized refers to size 3 (ranging from 120 to 450 kg), and large refers to sizes 4 to 6 (exceeding 450 kg).However, to compare with other archaeological sites, such as FLK Zinj or DS, size 3 and 4 carcasses have been merged in a single group.This is justified because only few bone specimens studied here belong to size 4 (n ¼ 28).Carcasses have been divided into four anatomical regions: skull (horn, cranium, mandible, and teeth), axial (vertebrae, ribs, pelvis, and scapula), appendicular (limb bones) and compact (carpals, tarsals, patella, phalanges, and sesamoids) bones.
The MNI has been estimated through two methods; first, using maxillary and mandible specimens and isolated teeth, and second, using long limb bones.The first method, as can be seen, offers information about the taxa present at PTK, whereas the second offers a more approximate estimate of carcass MNI.The survivor index (SI) has been applied to evaluate the existent deviation between MNE and MNI values using long limb bones.The survival rate has been obtained as the result of multiplying our MNE value for each skeletal element by the MNE expected given the MNI (obtained from our sample) and dividing it by 100 (Brain, 1969).
To determine the MNE, ribs have been divided following Rodríguez-Hidalgo et al., (2017) system: portion 1 belongs to the epiphysis which includes the head, neck and costal tubercle; portion 2 belongs to the coastal angle; portion 3 belongs to the proximal body; portion 4 is composed by the distal body and portion 5 belongs to the sternal zone.The high frequency of fragmentation documented in the PTK's ribs and the lack of landmarks made us determine the rib's MNE through portions 1 and 2. Long limb bones were also divided into upper (humerus and femur), intermediate (radius-ulna and tibia), and lower (metacarpal and metatarsal) limb bones (Domínguez-Rodrigo, 1997a, b).These, in turn, have been divided into proximal and distal epiphysis, proximal shaft, midshaft and distal shaft.The MNE estimates were based on the systematic inclusion of shaft specimens using a manual overlap approach (Yravedra and Domínguez-Rodrigo, 2009), and taking into account the size, side, landmarks, and ontogenetic age of each identifiable specimen (Barba and Domínguez-Rodrigo, 2005).Given that destructive processes like differential preservation, carnivore ravaging, and water flows tend to eliminate or underrepresent less dense bones (e.g., Voorhies, 1969;Brain, 1967Brain, , 1969;;Lyman, 1984Lyman, , 1992;;Marean and Spencer, 1991;Marean et al., 1992;Capaldo, 1998b;Pickering et al., 2003).%MAU was compared with bone density values.Density values were acquired from Lyman (1982Lyman ( , 1984) ) for small carcasses and Lam et al. (1999) for medium-sized carcasses.In order to approach the exploitation and transport strategies that took place at PTK, skeletal profiles have been evaluated through the statistical correlation between %MAU and the utility index (Binford, 1978).The modified general utility index (MGUI) for small carcasses has been obtained from Binford (1978), and the standardized food utility index (SFUI) for medium-sized carcasses stems from Metcalfe and Jones (1988).Spearman's correlations were applied first on all skeletal elements and second on the high-survival elements (skull, mandible, and long limb bones).For this latter analysis, the samples were bootstrapped 1000 times to avoid type II error, as described by Faith et al. (2009), since at PTK the samples are small (n ¼ 37 for small size and n ¼ 77 for mid-size carcasses).Finally, high-survival elements were used to identify the most probable transport strategy through a uniformity index.Shannon evenness index measures the uniformity of the distribution of skeletal elements concerning a complete carcass.Evenness values between 0.961 and 0.996 suggest a complete carcass transport or no transport while values lower than 0.96 suggest a selective transport if sample sizes are equal or higher than 100 elements (Faith and Gordon, 2007;Egeland et al., 2008).If MNE samples are equal to 50 then the threshold lowers to 0.92 (Faith and Gordon, 2007).

Testing site integrity
Due to the clear bias documented in the skeletal profiles, we applied a series of statistical analyses targeting specimen size distribution and the shape and composition of the faunal remains, in order to detect the degree of site integrity.Experimental studies have shown that post-depositional processes, such as water flows (e.g., Badgley, 1986a,b;Coard and Dennell, 1995;Domínguez-Rodrigo et al., 2014b;Isaac, 1967;Kreutzer, 1988;Schick, 1984;Toots, 1965;Voorhies, 1969), trampling, or even gravity (Kerbis-Peterhans, 1990;Domínguez-Rodrigo and Martínez-Navarro, 2012;Krajcarz and Krajcarz, 2014) can produce a preferential distribution of major axes of material (anisotropy).All the archaeological specimens with a longitudinal axis (A-axis) at least twice as long as their width (B-axis) were measured with compasses and clinometers to obtain their horizontal (azimuth) and vertical (plunge) orientations.(Voorhies, 1969;Fiorillo, 1991;Alcal a, 1994;Howard, 2007).Measurements were taken along an A-axis that divided the specimen symmetrically along its longitudinal axis since elongated objects tend to orient according to this axis (Toots, 1965;Voorhies, 1969;Domínguez-Rodrigo and García-P erez, 2013).The uniformity of the archaeological assemblage was statistically evaluated by combining three tests.The Rayleigh test (R) was used to evaluate if the material had a preferent unimodal orientation (Fisher, 1995), and the omnibus Kuiper (V) and Watson (W) tests were further used to determine if the orientation of the archaeological assemblage is bimodal or multimodal (Fisher, 1995) in the likely event of an anisotropic distribution.The three tests were applied in this study using the R library "CircStats" (Lund and Agostinelli, 2018).
It has been shown that the shape and composition of bones influence their transport (e.g., Schick, 1984;Visher, 1969;Reineck and Singh, 1980;Stein, 1987;Domínguez-Rodrigo et al., 2014b).Bone fragments were measured to determine their shape (flat, cube, and tube) and composition (dense or trabecular).Bones were defined as "flat" when the thickness was less than one-third of the width; "cube" when the thickness was greater than one-third of the width, and "tube" when the specimen was from a long bone that preserved the complete shaft section.The bone composition was divided into "dense" if dense tissue represented >2/3 of the specimen, and "trabecular" if cancellous bone structure constituted more than one-third of the composition of the bone (following Domínguez-Rodrigo et al., 2014b system).The data obtained from level 22A were compared with two experimental assemblages, which represent "transported" and non-transported "lag" samples, and with an ethnoarchaeological assemblage without hydraulic disturbance ("Maasai Camp") (Domínguez-Rodrigo et al., 2014b;Domínguez-Rodrigo and Lezana, 1996).Data were treated statistically using Bootstrapped Confidence Regions for Correspondence Analysis (Ringrose, 2012).This was done with the R "caboot" library (Ringrose, 2019), which performs correspondence analysis on a two-way contingency table and produces bootstrap-based elliptical confidence regions around the projected coordinates for the categories.

Bone surface modifications
As mentioned earlier, bone surface modification (BSM) analysis (especially cut marks) is the best method to discern the type of carcass access.Cortical surface preservation was evaluated by taking into consideration the stages of subaerial exposure (Behrensmeyer, 1978) and water-induced modification (C aceres, 2002).Cortical surfaces were examined with 10e20x hand lenses under a strong oblique light source (Blumenschine, 1995).Several types of marks could be identified: cut marks, percussion marks, tooth marks, trampling, abrasion/polish, and biochemical marks.The analysis applies a "configurational approach" where mark morphology, the anatomical placement of marks, and the sedimentary context of the specimen are taken into account (Fisher, 1995;Pickering and Wallis, 1997;Pickering et al., 2005;2004;Domínguez-Rodrigo et al., 2005).The identification of tooth and percussion marks was made following the diagnostic criteria outlined by Blumenschine (1988Blumenschine ( , 1995) ) and Blumenschine andSelvaggio (1988, 1991).Cut marks were identified and tallied following Bunn (1981) and Domínguez-Rodrigo et al. (2007b, 2009, 2010).Since comparisons of archaeological data with modern experimental assemblages involve processes that do not apply to the latter (diagenetic bone breakage or differential bone surface preservation), long bone mark frequencies were corrected following Pickering et al. (2008) method.This method attempts to correct for both the inflation of specimen frequency through dry breakage and artificial reduction of mark frequencies on poorly preserved cortical surfaces.
A multiple discriminant analysis (MDA) was conducted on the well preserved and dry breakage corrected sample to evaluate BSM information.MDA differentiates among distinct experimental groups (which represent primary or secondary access) and classified the PTK data accordingly (Domínguez-Rodrigo et al., 2014c).Prior to statistical analysis, samples were bootstrapped (1000 replications, with replacement), to normalize sample distribution, a prerequisite to applying standard parametric tests such as MDA.The statistical analysis was performed in the R software (R Core Team, 2020) using the R library "BiplotGUI" (La Grange et al., 2009).
The analysis of the breakage planes at PTK is based on the identification of green and dry breakage planes following Villa and Mahieu's (1991) criteria and the measurement of the plane angle (longitudinal or oblique relative to the long axis) with a goniometer.Measurements were taken at the point of maximum angle because they represent the extreme values that most of the breakage planes show (see Mocl an et al., 2019).
In a previous experimental approximation, Mocl an et al. ( 2019) proposed a method to analyse bone breakage of medium-sized animals produced by carnivores (hyenas and wolves) and direct anthropic percussion based on machine learning techniques (for more information about the experimental samples see Mocl an et al., 2019).This approximation combines variables related to the type and the angle (in degrees) of green fracture planes, the length of the specimens, and the presence or absence of epiphyses and notches (types A, C and D).The analyses proposed by Mocl an et al. ( 2020) are based on the idea that longitudinal and oblique fracture planes (and planes < and >90 ) could be analyzed separately and jointly creating five different types of analyses (i.e., all fracture planes, longitudinal <90 and >90 and oblique <90 and >90 ).
Recently this method has been reviewed by other authors (Yezzi-Woodley et al., 2022) which have pointed out the existence of some problems related to the data curation (especially in the case of the fracture planes broken by wolves) and the structure of the dataset (there are two variables [length of the specimen and length intervals] which can be considered as redundant) of the original work of Mocl an et al. ( 2019).Currently we are trying to improve the data curation of the sample broken by wolves (Mocl an and Domínguez-Rodrigo, work in progress).Yezzi-Woodley et al. ( 2022) also proposed that the inclusion of some fractures planes of the same type documented on the same single bone specimen (e.g., longitudinal <90 ) could create methodological biases.
Thus, in this work we have implemented the use of these methods excluding the sample produced by wolves and deleting the 'interval length' variable (see Mocl an et al., 2019).We have also discarded the use of some fracture planes of the same type observed on the same bone specimen.If there is more than one per specimen, we have only used the most extreme values (i.e., lowest values in the case of <90 and highest in the case of >90 ) to perform the analyses.In addition, analyses were performed without considering the size of the fracture plane (>4 cm or < 4 cm) variable that was included in Moclan's et al., works (2019Moclan's et al., works ( , 2020)), because this variable was not collected during the taphonomic analysis of PTK.
Here we have used five mathematically different algorithms in order to classify the experimental samples of fracture planes broken by hyenas and anthropic percussion: neural networks (NNET), radial kernel support vector machines (SVM), k-nearest neighbours (kNN), random forest (RF) and naïve Bayes (NB).Then, we used the best models (which showed the most accurate results following kappa and accuracy indeces) to classify the fracture planes of PTK.All these algorithms were trained using 70% of the samples as training, and 30% as testing.Cross-validation (10 folds, 10 repeats) was also used.In addition, 10 different configurations of the hyperparameters of the algorithms were used with the function 'tuneLength' of the 'caret' library of R (Kuhn et al., 2020) in order to define the most accurate model per analysis.As the samples were clearly unbalanced, we used the function 'upSample' to overcome the possible bias produced by the data structure.In the original works of Mocl an et al. (2019,2020) bootstrapping was used, here we have only use the upsampling method to create balanced samples.We have also noted which cases have been classified with a probability higher than 0.9 because of the reliability in how they can be considered as true positives.
As can be seen in Table S3 some algorithms have provided kappa values which can be considered as valid (i.e., >0.4) to classify the archaeological samples.These algorithms are NNET and RF for all the fracture planes, NNET, kNN and NB for longitudinal fracture planes <90 , NNET for longitudinal fracture planes >90 and NNET for oblique fracture planes >90 .In this sense, we have discarded the use of the oblique fracture planes <90 to classify the archaeological samples (instead of they are included in the analysis of 'all fracture planes').
Fracture notches were identified according to the typological classification proposed by Pickering and Egeland (2006) (modified from Capaldo and Blumenschine, 1994): 1) complete notches (Type A) have two inflection points on the cortical surfaces and a nonoverlapping negative flake scar; 2) incomplete notches (Type B) are missing one of the inflection points; 3) double overlapping notches (Type C) have negative flake scars that overlap with an adjacent notch; 4) double opposing complete notches (type D) are two notches that appear on opposite sides of a fragment and result from two opposing loading points; and 5), micro notches (Type E, <1 cm).
Bootstrapped Correspondence Analysis (Mocl an and Domínguez-Rodrigo, 2018) was used to identify the more probable agent that produced the notch assemblage in PTK.This approximation allows for comparison of the notches present on PTK small and medium-sized animals with those generated by anthropic percussion (Blasco et al., 2014;de Juana and Domínguez-Rodrigo, 2011; Mocl an and Domínguez-Rodrigo, 2018), anthropic battering (Blasco et al., 2014) the activity of hyenas (Domínguez-Rodrigo et al., 2007b), and lions (Arr aiz et al., 2017).The analysis was performed using the 'cabootcrs' R library (Ringrose, 2019).
Shaft circumference type frequency provides information about the agent of breakage (Bunn, 1982(Bunn, , 1983) ) and the integrity of a site (Marean et al., 2004;Pickering and Egeland, 2006).Based on data from experimental hammerstone-broken assemblages and hyena dens, three types of shaft circumference were established following Bunn's (1982) classification: Type 1 fragments retain <50% of the original circumference, Type 2 retain >50% and Type 3 preserve the complete shaft circumference.
In order to evaluate the degree of ravaging intensity at PTK, the indices proposed by Domínguez-Rodrigo and Organista (2007) were used: 1) the ratio of axial to appendicular remains, and 2) the ratio of proximal humerus and distal radius to distal humerus and proximal radius.Since the assemblage was broken mostly by hominins, the ratio of femur to tibia was excluded from the analysis.The values of these ratios classify the assemblage into one of the four different ravaging stages being stage 0 indicative of nonravaging and stage 4 complete ravaging.Indeterminate axial and appendicular specimens were not taken into account in the calculations.

Age profiles
The MNI determined by teeth, including isolated teeth, from mandibles and maxillae is 20.Of these individuals, 15 are bovids, and the other five belong to Equus, Hippopotamus, suidae, Theropithecus, and Homo habilis.We focused on the age at death of the bovids and the equid.The small carcasses are represented by three individuals: two prime adults and one old individual (Table 1).Medium-sized carcasses are represented by 13 individuals: two young juveniles, three subadult juveniles, seven prime adults and one old individual (Table 1).
Considering the postcranial skeleton, the MNI is slightly higher.For small carcasses, it has been estimated at 10 by the humerus.We documented nine left caudal distal shaft specimens; one belongs to an adult individual (size 1) and eight to subadult individuals of carcass sizes 1 (n ¼ 4) and 2 (n ¼ 4).Right humeri have been quantified by two distal shafts from adult individuals, but one of them could be the antimere of an adult left humerus, and the other belongs to a size 2 carcass.Regarding medium-sized carcasses, the MNI of 11 has been determined by both femoral and tibial elements.In the first case, the femora are composed of 10 right caudal shafts (three of them belong to juvenile animals) and one left caudal distal shaft with a very thick cortical section.In the case of the tibia, eight left caudal-lateral proximal shafts with the presence of foramina and one shaft belonging to Equus have been documented.The left tibiae are represented by two cranial proximal shafts which were still in the growing process.The long bone cortex growth patterns show the typical fibrolamellar bone tissue (highly vascularized) during subadult age growing into dense periosteal and endosteal lamellar bone in adult age (Manilay et al., 2013).According to Carter et al. (1976), bovine long bones are fully grown in about 2 years.If we quantify the elements with woven fibrous bone and less dense as juveniles (young and/or subadult juveniles) their MNI is 11: eight belong to small carcasses and three to medium-sized carcasses.Four of these small carcasses probably belong to juvenile Kobus which as adult is a size 3 animal, and were therefore included as medium-sized carcasses in the mortality profiles, since aging is different from taphonomic analysis in carcass size information.
At PTK, DS and Kanjera, the small carcass mortality profiles include more juvenile individuals than FLK Zinj, which is dominated by older individuals.PTK (like DS and Kanjera) falls in the attritional area (using Stiner's model), with a clear trend toward young individuals (Fig. 4A).The mortality profile for medium-sized carcasses falls in the middle of the attritional and catastrophic area (Fig. 4B).Its confidence interval includes all the samples except the lions and hyenas suggesting that the profile is closer to ethnographic and archaeological samples with ambush behaviors.However, due to the archaeological samples being too small, the large confidence ellipses in the ternary diagram cover all carnivore samples being statistically indistinguishable.Taking into consideration the ternary model produced by Discamps and Costamagno (2015) both samples fall into the JPO zone (juvenile, prime, old) which indicates that both samples are dominated by juvenile individuals and could represent both types of profile: attritional or catastrophic.To avoid this equifinality we resort to PCA and CVA analyses using a multiple-age variable partition.
Fig. 5 shows the biplots of the PCA and the CVA analyses which produce three different groups for small-sized carcasses.The first group is formed by FLK Zinj, which appears separate and close to prime-adult dominated individuals.The second group is composed of the lion and hyena samples and DS and PTK archaeological samples which are slightly overlapping with them, suggesting an ambush hunting strategy.The third group includes the cursorial samples from wild dogs and leopards, the ethnographic sample of Hadza and the archaeological site of Kanjera.The PCA and CVA biplots also show three different groups for medium-sized carcasses (Fig. 6).The first group is formed by FLK Zinj and DS; both archaeological sites appear close to prime-adult dominated individuals.Their location and their separation from other samples, as Cobo-S anchez (2020) indicates, is probably because they reflect a hominin-specific carcass-acquisition (hunting) model (Bunn and Pickering, 2010;Domínguez-Rodrigo and Pickering, 2017).The second group comprises the lion and hyena samples.The third group includes the ethnographic samples and the archaeological sites of Kanjera and PTK, which show a bias toward juvenile individuals suggesting a cursorial hunting strategy if the agency was a nonhominin carnivoran.The wild dog prey mortality pattern resembles that of ethnographic samples and it slightly overlaps with them.
Given the closed-vegetation environment surrounding the PTK-DS-FLK Zinj sites (Arr aiz et al., 2017), we argue that hominins may have employed an ambush strategy to acquire small and mediumsized carcasses (Bunn and Gurtov, 2014).However, this may have differed from mammal carnivore ambush strategies in that the latter produce a more eclectic age profiles, since it is nondiscriminatory and it samples age distribution in the prey species cohort demographics (Bunn and Gurtov, 2014), whereas hominin ambush strategy at PTK may have been more selective, targeting juvenile and prime adults over the rest of potential age profiles.

Skeletal profiles
As explained above, the MNI differs depending on whether only teeth or teeth plus long bones are used (Table 1).The faunal assemblage is composed of 1265 specimens out of which 344 (27%) belong to small carcasses, 786 (62%) belong to medium-sized carcasses and 143 (11%) could not be assigned to carcass size.Tables 2  and 3 show the NISP and MNE per element for each carcass size.Correlations between NISP and MNE for small and medium-sized carcasses result in significant and high correlation coefficients (Spearman's rho ¼ 0.8) in both samples.
Small animals are mainly represented by long bones, mandibles and pelvis (Fig. 7).Long limb bones constitute 40% (n ¼ 139) of the sample, with the humerus being represented almost twice as much as the rest of the long limb bones.Axial elements, such as ribs, vertebrae, and scapulae are underrepresented, as measured by MNE and MNI.Compact bones (carpals, tarsals, and phalanges), and long bone epiphyses are almost absent (n ¼ 13) (Fig. 8).Proximal epiphyses from the radius-ulna are the best represented followed by both epiphyses from humeri, while femoral and tibial epiphyses are missing.Only two distal epiphyses from tibiae were recovered, one of them digested.Medium-sized carcasses show a similar pattern of skeletal abundance (Fig. 7).Axial and compact bones are underrepresented.Long bones are most frequently represented by shaft portions (n ¼ 407; 52%) while the epiphyses are underrepresented (n ¼ 24), and they mainly belong to metapodials (n ¼ 14; 58%) (Fig. 9).Table S4 shows skeletal element abundances in Minimum Number of Elements (MNE) and Percent Minimal Animal Units (%MAU).
As a consequence of the low level of cancellous bone representation, the attrition effects were estimated through Spearman's rank-order correlations between %MAU and bone mineral density (Table 4).Taking into account that most of the small carcasses belong to juvenile animals, the correlations between %MAU and bone density are only approximate.The coefficients show a significant correlation for medium-sized carcasses, but not so for small carcasses.Considering the high-survival element subsample, the correlation coefficients are negative, showing an inverted relationship for small and medium carcasses.
When analysing the skeletal parts according to food utility indexes, the results are non-significant in both carcass size groups (Table 4).Considering the high-survival bootstrapped subsample the results are similar for small carcasses, whereas medium-sized carcasses show a significant but weak correlation.According to these results and taking into account that the MNE samples are lower than 50 for small and lower than 100 for medium carcasses, the evenness index (Table 4), suggests that small and mediumsized carcasses may have been transported complete to the site (or unselected), since their confidence interval values are within the 95% confidence limits of even element representation (0.92e0.99 see Faith and Gordon, 2007).

Testing site integrity
The documented underrepresentation of cancellous bones suggests that abiotic and/or biotic agents acted in the PTK faunal assemblage.Hydraulic flows easily transport the small and lowdensity elements.This would explain in part why most cancellous elements and portions are highly underrepresented.However, the analysis on specimen length distribution shows that the fragments between 21 and 40 mm are the most abundant ones, and that, in general, frequencies decrease until reaching specimens >100 mm.This pattern is echoed in the sub-sample of green-broken long bone specimens (Fig. 10A).The high frequency of very small specimens and the low frequency of polished and rolled specimens (n ¼ 49), the majority of them in 1st degree (n ¼ 37) and the rest in 2nd degree, according to C aceres ( 2002) system, indicate a limited postdepositional disturbance by abiotic factors (Fig. 10B).In addition, the analysis on bone shape and composition locates PTK next to the experimentally undisturbed assemblage (Maasai Camp), indicating that PTK is an assemblage without significant hydraulic disturbance (Fig. 10C and Table S5).This statement is corroborated by circular statistical tests, which reflect an autochthonous uniform distribution of the stone tools and fossil bones (R ¼ 0.888, p ¼ 0.412; V ¼ 1.946, p¼ <0.02; U2 ¼ 0.135, p¼ >0.1).Additionally, the clustering of long limb shaft fragments according to element (i.e., fragments from the same type of bone occur concentrated in the same loci), indicates that the site represents the original depositional locus, and that post-depositional disturbance only scattered few shaft fragments in the vicinity of their concentration (Fig. 11).

Bone surface modifications
Once the hydraulic flows were discarded as an important factor explaining bone deletion from the PTK archaeofaunal assemblage, focus was placed on BSM and the attrition produced by biotic agents.In general, all bone material is well preserved: 68% of the sample shows good cortical preservation, with the majority of the specimens showing a stage 0 and 1 (58% and 28%, respectively) of subaerial exposition, suggesting that the assemblage was quickly Fig. 5. Multiple discriminant analysis using A) Principal Component Analysis (PCA) and B) Canonical Variate Analysis on the bootstrapped samples of the mortality profiles generated by carnivores and modern hunter-gatherers, as well as the age profiles documented at PTK (current study), DS (Cobo-S anchez, 2020), FLK Zinj (Bunn and Pickering, 2010) and Kanjera (Oliver et al., 2019), regarding small bovids and using three age classes.
buried after deposition and that the site formed during a short time span.The preservation of original cortical surfaces allowed us to examine if alteration of the outer cortical layer was caused by subaerial weathering or chemical weathering during diagenesis.At PTK, most of the cortical modification is of chemical origin.
The low representation of axial elements and epiphyses suggests intense carnivore ravaging or other density mediated destruction processes that do not include water flows.However, a low frequency of tooth-marked specimens (n ¼ 9 for small carcasses and n ¼ 28 for medium carcasses) has been documented.Carnivore damage is mainly documented on long limb bones (77% and 78% of all documented tooth marks respectively according to carcass size) (Tables 2 and 3).Cut mark frequencies on the NISP are moderate, with 6% of small-sized identified specimens and 9% of medium-sized identified specimens presenting cut marks.The majority of the cut marks occur on midshaft sections of long limb bones: 55% and 52% of the total cut marks observed on small and medium-sized identified carcasses appear on long bone midshaft.ULB and ILB display the highest frequencies of cut marks (Table 5).
The BSM frequencies documented on the corrected long limb bone sub-sample are shown in Table 6 and show that 12.9% of small-sized long bone midshafts bear cut-marks, and the percentage is 17% among medium-sized carcasses (Fig. 12).The low percentage of tooth marks and the higher percentage of cut-marked long bone specimens occur mainly on hot zones (which are areas on limb bones where flesh scraps rarely survive after lion consumption) (Figs.S1 and S2).This confirms that the assemblage was produced and exploited first by hominins, who consumed the flesh and extracted the bone marrow (see below).Cut marks were also identified on cranial, mandibular, and axial bones, including scapulae and pelves (Tables 2 and 3), which also confirms that hominins had primary access to bulk flesh from whole carcasses and exploited them intensively.This is supported by the predominance of juvenile individuals, which would be utterly defleshed had hominins acquired them after felids initiated consumption (Cavallo and Blumenschine, 1989;Stiner, 1990Stiner, , 1991;;Domínguez-Rodrigo, 2001).Cut marks have been also detected on two phalanges belonging to small-sized animals, being one of them on a juvenile individual, suggests that hominins could also be skinning the animals (Fig. 12).Finally, the MDA analysis locates PTK within the experimental model of primary access more than secondary access (Fig. 13 and Table S6).
Fig. 6.Multiple discriminant analysis using A) Principal Component Analysis (PCA) and B) Canonical Variate Analysis on the bootstrapped samples of the mortality profiles generated by carnivores and modern hunter-gatherers, as well as the age profiles documented at PTK (current study), DS (Cobo-S anchez, 2020), FLK Zinj (Bunn and Pickering, 2010) and Kanjera (Oliver et al., 2019), regarding medium bovids and using three age classes.
The percentages of percussion marks on long bones are 7% for small carcasses and 16.1% for medium-sized carcasses (Fig. 14).The lower percentage of percussion marks on small carcasses may be explained by the fact that most of them belong to juvenile individuals and they do not have a fatty marrow because the marrow cavity is composed of red blood cells -red marrow-which is low in fat content (Blumenschine and Madrigal, 1993;Munro and Bar-Oz, 2005;Currey, 2006).Nevertheless, both values fall into the 95confidence interval of hammerstone to carnivore experimental samples (Blumenschine, 1988(Blumenschine, , 1995;;Capaldo, 1997).Faced with this situation, other intervening carnivores were limited to scavenging the remains abandoned by hominins.Considering that tooth and percussion marks frequencies are low, notches and breakage planes were analyzed to confirm the main agent(s) of bone breakage at PTK.

Bone breakage patterns
A total of 868 bone specimens could be identified as bearing green and/or dry breakage.Of these, 696 (80.2%) specimens show green breakage only, 100 (11.5%) have dry breakage only, and 72 (8.2%) specimens have both types of breakage.The majority of the green breakage is present on appendicular bones (n ¼ 483, 69.4%), with the rest documented on axial (n ¼ 131, 18.8%) and cranial while Type 2 and Type 3 show lower frequencies: 5.1% (n ¼ 27) and 6.2% (n ¼ 33), respectively.This suggests two things: first, it confirms the integrity of the site, and second, that biotic agents were breaking the bones (Fig. 15).A total of 64 notches have been identified in the assemblage (Fig. 15).The frequency of the different type notches has shown a greater presence of types B (n ¼ 25) and A (n ¼ 24), followed by types C (n ¼ 11) and D (n ¼ 4).Notches are distributed across all bone types including two hemimandibles and one phalanx, but the majority of the notches are present on long bones.Humeri and femora present the highest number of notches (n ¼ 19 and n ¼ 13, respectively) followed by tibiae (n ¼ 8), radii-ulnae (n ¼ 6), metacarpals (n ¼ 7) and metatarsals (n ¼ 2).Notch type A is the most frequent (43.7%) in ULB followed by notch type B (31.2%).
The analysis of the notches (Fig. 16) shows that there is a clear overlap between the 95% confidence interval of the sample of small carcasses of PTK and the anthropic experimental samples.However, the small sample size provides a very large ellipse that is also overlapping with battering activities and direct percussion.Nevertheless, the mean value of the sample seems to be related to direct percussion activities.In contrast, medium-sized carcasses show a different pattern.This sample clearly overlaps with the notch samples produced by direct percussion, as well as slightly with the one generated by hyenas.Thus, it seems difficult to infer agency from the notch sample.However, a combined model including primary human access and secondary hyena scavenging might be able to explain the patterns observed at PTK, as highlighted by the Correspondence Analysis (Fig. 16).
A total of 218 fracture planes have been measured.Among longitudinal fracture planes 52 fracture planes are <90 and 46 fracture planes are >90 .In the case of the oblique fracture planes, the sample is formed by 57 fracture planes are <90 and 63 fracture planes are >90 .
The new trained models (Table S3) have identified most fracture planes as anthropogenic (Table 7).First, when the analysis is performed using all the fracture planes at the same time (i.e., longitudinal þ oblique > and < of 90 ) NNET and RF classify 80% and 92.38% of the PTK sample as anthropogenic.These percentages are increased is only those cases classified with probabilities higher than 0.9 are considered, reaching up to 98.33 and 100%.
The results obtained on the longitudinal fracture planes <90 suggest the anthropogenic nature of the assemblage at PTK (between 59,62% and 94,23% of the fracture planes), but in general (with the clear exception of kNN which is the most accurate model      S5. in this case) the algorithms show a higher number of hyena-broken fracture planes than in the other samples (between 21.15% and the 40,38%).In this case, none fracture plane have been classified with a probability higher than 90%.
In the case of longitudinal fracture planes >90 , NNET has classified the most part of the sample as anthropogenic (80.43%) while 5 fracture planes have been identified as anthropogenic with a probability higher than 0.9.
Finally, in the case of oblique fracture planes >90 most of the fracture planes are also classified as anthropogenic by NNET (92.06%).In this case there are 48 cases classified as anthropogenic with a probability higher than 0.9.

Discussion
A previous taphonomic analysis on the cut mark distribution of the zooarchaeological assemblage of PTK concluded that the main accumulation of the sample was produced by anthropogenic activity (Domínguez-Rodrigo et al., 2021).However, that analysis did not address other questions about the processes that created the faunal assemblage final configuration.The taphonomic analysis of PTK that has been detailed in this paper suggests that hominins were processing different animal taxa (e.g., Antidorcas, Kobus, Parmularius, Hippotragus) from different ages, with an emphasis on juvenile and prime adult individuals.Bulk flesh and marrow exploitation is supported by the frequency and distribution of cut marks on the high meat-yielding upper and intermediate limb bones and by the frequency of percussion marks on all long limb elements (Figs. 12 and 14).The analysis of breakage planes also confirms that at least between 59,62% and 94,23% of the long bone sample was broken by hominins.Carcasses seem to have been transported in a non-selective way to the site, as the evenness index suggests.Nevertheless, the correlation between %MAU and the density in the small size sample is too small to be significant.The correlation only explains 15% of the sample.From this perspective, complete small carcasses could have been originally transported to the site.This would make more sense vis-a-vis other data, since most of the carcasses are juvenile and easier to transport as complete units (Bunn et al., 1988).In essence, hominins were repeatedly transporting fleshed animal resources into the site, and all taphonomic evidence indicates that these were acquired through primary access (see below).
It must be noted that abiotic processes have produced a low rate of modification of the original bioestratinomic assemblage.However, it is clear that other non-hominin biotic agents modified the original anthropogenic accumulation of carcasses at the site.The use of skeletal profiles to interpret hominin behavior is complex because it is difficult to associate certain skeletal patterns to human activity or specific taphonomic agents.Overall, the skeletal part profiles provide an approximation to interpret hominin carcass transport and exploitation strategies when the anthropogenic agency can be otherwise determined, as well as to evaluate the degree of integrity of the assemblage impacted by taphonomic agents.The PTK's low frequency of axial bones and the high frequency of limb bones without epiphyseal sections is indicative of post-depositional carnivore ravaging at the site (Capaldo, 1995(Capaldo, , 1998;;Domínguez-Rodrigo et al., 2015a) rather than the result of a selective transport by hominins.Only 13 long bone specimens display evidence of carnivore gnawing.The hyenid activity on the assemblage is confirmed by the machine learning analysis applied to the breakage planes.Depending on the model and fracture plane, the activity of the hyenas on the assemblage can range from 5,77 to 40.38%.As reported here, the scarcity of epiphyseal sections suggests that bone-crunching hyenas may have deleted these portions after butchery and marrow extraction by hominins.The presence of digested bones (n ¼ 5) belonging to these parts from size 3 juveniles confirms this, and suggests a consecutive use of the site by both hominins and other carnivores.The paucity of tooth marks can also indicate that small carnivores (i.e., jackals) may have intervened by transporting bones away.
When evaluating the degree of ravaging at PTK, the ratio of axial (vertebrae and rib) to long limb bones, and the ratio between proximal humerus plus distal radius to proximal radius plus distal humerus (Fig. 17), the small carcasses value falls inside of stage 2, which represents moderate ravaging, where at least half of the axial and long bones have been reduced by the action of carnivores.While the medium-sized carcasses value is indicative of an intensive carnivore ravaging represented by stage 3, where bones frequencies have been reduced by one-third or less (Domínguez-Rodrigo and Organista, 2007).These stages imply a moderate to high competition habitat.However, the PTK environment was characterized by the presence of trees (Arr aiz et al., 2017) that could have served as a refuge by temporarily decreasing the trophic competition (i.e., the timing of access to resources could have been longer than in an open setting, thus lowering the overall competition).Nevertheless, the competition degree also depends on carnivores' nutritional stress.The low representation of migratory animals at PTK, as in FLK Zinj and DS, suggests a dry season occupation (Domínguez-Rodrigo et al., 2019;Linares-Mat as and Clark, 2021).The hominin occupation of PTK potentially at the end of the dry season, which is the moment of higher nutritional stress for carnivores, would also explain the potentially higher degree of ravaging or bone deletion) documented in the assemblage.
At PTK, 50% (n ¼ 172) of the small carcass' specimens belong to juvenile animals and 7.5% (n ¼ 13) are cut-marked.The percentage of juvenile specimens belonging to medium-sized carcasses is 21% (n ¼ 167) and 14% of these (n ¼ 24) are cut marked.The predominance of juvenile individuals that preserve a high frequency of cut marks, similar to primary access models (Domínguez-Rodrigo, 1997a), is suggestive of hunting and not of scavenging, since carnivores rapidly destroy small carcasses leaving no remains available for secondary consumers (Schaller, 1968;Blumenschine, 1987;Cavallo and Blumenschine, 1989;Domínguez-Rodrigo, 2001).A similar age pattern has been documented at Kanjera South, Kenya (2.0 Ma) where 50% of the sample belongs to juvenile carcasses, which were also interpreted as having been acquired through hunting (Ferraro et al., 2013;Parkinson, 2013Parkinson, , 2022;;Oliver et al., 2019), but with the difference that in Kanjera South the midshaft specimens from size 1 carcasses are not cut-marked while at PTK, the midshaft sample (n ¼ 36) from small size carcasses shows 16% of cut marks (n ¼ 6).As in Kanjera (Parkinson et al., 2022), a substantial amount of cut marks not only occur on meat/bearing long bones (namely, upper and intermediate elements), but are also located on hot zones where carcasses are usually devoid of scraps if accessed after felids.The juvenile-dominated Kanjera assemblage is probably one of the earliest pieces of evidence of primary access by hominins, and the pattern of abundance of juveniles and anatomical distribution of cut marks (in higher frequencies) is replicated in the PTK archaeofaunal assemblage.Oliver et al. (2019) explained the high frequency of juvenile individuals in Kanjera South due to the open paleolandscape of the site, where hominins had to adopt a cursorial strategy to get the most vulnerable prey.In the case of PTK, the paleolandscape is reconstructed on a high lacustrine terrace (Uribelarrea et al., 2014), where phytoliths show the presence of forest/palm plants and the presence of ferns suggesting a shaded and humid habitat.PTK would be characterized by mixed paleovegetation, dominated by forest and woodland with patches of open spaces occupied by grasses (Arr aiz et al., 2017).In this environment, an ambush strategy seems to be the most appropriate.
At PTK, the carcasses belong mainly to juvenile and adult Kobus sigmoidalis.Considering the ethology of the waterbuck, one can speculate about the hunting behavior of hominins.The waterbuck is gregarious and nonmigratory (Foley, 1993).Their territoriality and preference for habitats close to the water with cover and grass (Kiley-Worthington, 1965) could have made them easy to locate.Given that the paleolandscape of PTK responds to these characteristics, it is possible that even hominins and waterbucks shared the same territory.Therefore, it is plausible that prey hunting decisions by early Homo were conditioned by the presence of juvenile prey.This could indicate a potential seasonality in the selection of subadults.Ethnographic and ethological studies demonstrate that the seasonal changes and the selection of certain habitats would have profound implications in food-procurement strategies (e.g., Speth, 1987;Abraham et al., 2022;Marlowe and Berbesque, 2009;Oelze et al., 2014).Then the selection of juvenile animals at PTK could be in connection with the season (dry or wet), the habitat (resource availability), and prey vulnerability, just as San, !Xo and G/wi hunters of the Kalahari have been observed to do (Schapera, 1965;Liebenberg, 2006).The predominance of prime adult individuals has been seen as the result of the selective ambush strategy (Bunn and Pickering, 2010); however, at the end of the dry season and the beginning of the wet season, the ungulates, and more specifically, the prime adults show higher nutritional stress, which is reflected in a decrease of their fat deposits (Speth, 1987Tomlinson, 1979, 1981;Murray, 1982;Van Rooyen, 1993).The power struggle to maintain territories results in a substantial decline in adult male body condition (Murray, 1982;Van Rooyen, 1993).Therefore, hominin hunting strategies could have avoided exclusively targeting the prime adult males and focused on other age ranges with more fat-rich resources, such as pregnant females (Linares-Mat as and Clark, 2021).This strategy has been documented at FLK Zinj where hominins also targeted fat resources through the acquisition of pregnant females (Bunn and Pickering, 2010).The presence of juveniles, given that the peak of births in waterbucks, as in other antelopes, is documented during the wet season, would be indicative of a dry season occupation.This could also explain the absence of non-local potentially migratory fauna.This situation is coherent with those data provided by the carnivore competition analysis described above.
The comparison of PTK with other pene-contemporaneous sites from Bed I, such as FLK Zinj and DS, shows some differences, but in general, all of these sites represent specific locations selected by hominins on the landscape, with the presence of trees and freshwater resources nearby (Ashley et al., 2010;Uribelarrea et al., 2014;Arr aiz et al., 2017).Hominins transported fleshed carcasses to these places to be processed and consumed.The three assemblages are represented by similar bovid taxa and by similar proportions of each species.Despite the fact that the three archaeological sites are located in closed-vegetation habitats , and were probably occupied during the dry season (Domínguez-Rodrigo et al., 2019), which should be relatively low-competition habitats (Domínguez-Rodrigo and Organista, 2007), at PTK the carnivore activity seems to be slightly more intensive than at FLK Zinj and DS, where a relatively higher presence of epiphyses has been documented (n ¼ 166 from 721 limb bone specimens, 23%; n ¼ 133 from 549 limb bone specimens, 24% respectively) as well as more axial elements (Domínguez-Rodrigo and Barba, 2007;Cobo-S anchez, 2020).The frequency of juvenile individuals at PTK (52%) is higher than at DS (35%) and at FLK Zinj (18.5%).At all the three anthropogenic sites, the juvenile taxa are similar: Kobus and Parmularius.The selection of young individuals by Oldowan's hominins could be because they may have been more vulnerable and abundant at the end of the dry season.Based on the optimal foraging theory, this strategy could provide them with a higher energy benefit for the lowest cost, maximizing the net energy obtained (Winterhalder, 1981).
It must be noted that PTK also presents a similar size and internal spatial distribution of archaeological materials as documented at FLK Zinj, with most of the assemblage documented in a single spatially-discrete cluster area (Domínguez-Rodrigo and Cobo-S anchez, 2017a).Inside this discrete area, three peaks of intensity (or sub-clusters) close to one another can be observed (Fig. 3).The similarity to FLK Zinj in the taxonomic composition, amount of materials, skeletal part distribution, taphonomic properties, lithic technology, and spatial distribution indicates a similar hominin behavior at both sites.In this sense, it is probable that the functionality of the site was similar to the one attributed to FLK Zinj  S6. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)(Domínguez-Rodrigo and Cobo-S anchez, 2017a).Thus, we propose that PTK was used as a central-provisioning place by hominins to where they transported some animal carcasses to be consumed and, probably, shared with other members of the group.PTK also shows the great potential of these nicely preserved archaeological records to unravel cues about the social organization and use of the space by those early humans, with most carcass consumption repeatedly taking place on the same discrete area, without the typical distribution according to household space that is typical among modern human foragers (Domínguez-Rodrigo and Cobo-S anchez, 2017b).Future dental microwear analyses at PTK will test the hypothesis that we introduce here about the seasonal

Table 7
Classification of the different bone fracture plane samples with the selected Machine Learning algorithms.In the table, the accuracy and kappa values of the models can be seen.
The number of classified PTK fossil specimens per agent (and its percentage) and the number of classified PTK specimens over 90% of confidence (p > 0.9) is also shown.nature of the assemblage.Our challenge now is to find the proper models and analogues that make this patterning understandable.The repeated use of the same loci for food transport, and (presumably) collective consumption, in the absence of fire, would indicate that site selection was determined by low-competition settings, where hominins could engage in food processing, consumption and other activities.This is indicated by the low carnivore input in the immediate paleolandscape surrounding these Bed I sites (Domínguez-Rodrigo et al., 2010).This low-competition condition must have existed at least on a seasonal basis, alternating with higher mammalian carnivore input at other times, where bone remains at these sites were still attractive for durophagous bonecrunching scavengers.
Reconstructions like those provided by the strong anthropogenic signal of sites like PTK, FLK Zinj or DS are not usual and have been referred to as "the Olduvai effect", implying their anomalous nature (Barr et al., 2022).Some zooarchaeologists seeking longitudinal trends mix chronologically-similar assemblages disregarding if they can be effectively compared (Barr et al., 2022).For such a comparison to be heuristically solid, the basic requirement is to use sites that contain similar qualities in agency and preservation.Most early Pleistocene sites in Africa are affected variously by different degrees of impact by biotic and abiotic processes during their biostratinomic and diagenetic phases.Comparing sites whose material preservation is deficient, whose integrity is low, and whose study involve different types of taphonomic analyses and methods to sites where hominin agency is prominent and preservation is exceptionally good, biases temporal and behavioral information, and epistemologically flaws interpretations.We argue, contra this approach, that effective longitudinal comparisons can only be made using exceptionally-preserved archaeofaunal records, where hominin behavioral information has been preserved and is minimally biased.This is why sites like PTK contribute to increasing our limited sample of such exceptional records, and documents hominin activities in a degree that other sites impacted by a higher range of taphonomic processes do not.

Conclusions
The taphonomic analysis of the PTK faunal assemblage has shown that the site is autochthonous and that it contains a high degree of integrity.The assemblage was formed as a result of the selective transport of fleshed carcasses to the site by hominins who consumed their meat and marrow.After that, hyenid activity was intensive probably deleting a substantial part of the original bone assemblage.The study concludes that the site may have been used by hominins as a central place, where the closed vegetation could have sheltered them from carnivores long enough to allow them to carry out several activities at the site.Its location, near a fluvial input shared with the nearby site of DS, would have enabled hominins to have easy access to water and herbivores using the area as a drinking location.The abundance of young individuals suggests that the hominins in PTK employed a selective hunting strategy slightly different from that documented at FLK Zinj and DS.This analysis reinforces the patterning in prey selection and use documented at other pene-contemporaneous sites, as well as the patterning in the use of the space in these loci by hominins.This clearly has important repercussions in how this information is used to interpret socio-reproductive strategies of early Homo.PTK also contributes to increasing the behavioral range previously documented in other sites, in a time in which the earliest evidence of Homo aff.Erectus (using morphological and dimensional criteria) is documented at the gorge (Domínguez-Rodrigo et al., 2015b).The chronologically sympatric co-existence of three hominin taxa (Paranthropus boisei, Homo habilis and Homo aff.erectus) on the same landscape, one of them with fully terrestrial adaptation, also requires revising the role of Homo habilis in the agency of the Oldowan archaeological record at Olduvai Gorge.

Fig. 1 .
Fig. 1.A) Location of PTK in the Olduvai Gorge.B) Detailed stratigraphic section of PTK with the archaeological levels 22A and 22B.C) Topographic model of PTK with the distribution of bone (red points) and lithic (blue point) remains at level 22A.(For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Fig. 2 .
Fig. 2. PTK site with a detailed of its stratigraphy and the archaeological materials discovered in trench 10 level 22A.

Fig. 3 .
Fig. 3. Spatial distribution of the archaeological materials in level 22A (red ¼ bones; blue ¼ lithics).(For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Fig. 4 .
Fig. 4. Triangle graphs showing the mortality patterns for small and medium-sized bovids killed by lions, hyenas, cheetahs, leopards, wild dogs, and Hadza and Kua hunter-gatherers, as well as the mortality profiles documented at PTK, Kanjera, DS, and FLK Zinj.

Fig. 7 .
Fig. 7. Skeletal part profiles and survivor index by carcass size.

Fig. 10 .
Fig. 10.A) Specimen size distribution quantified by %NISP.B) Distribution of rolled and polished bones at PTK level 22A.C) Biplots of the bootstrapped CA of bones showing the relationship between referential models (Maasai camp, lag and transported modern assemblages) with respect to the shape and composition of bones at PTK level 22A.Data from Domínguez-Rodrigo et al., 2014b.Ellipses with 95% confidence intervals for referential models (left) and shape and composition of bones (right) are also displayed.Length of the axes shows the importance of the contribution of each variable to the inertia.Loading scores for the variables are displayed in TableS5.

Fig. 11 .
Fig. 11.Spatial distribution of long bones according to element at PTK level 22A.

Fig. 13 .
Fig. 13.Multiple discriminant analysis (using a canonical variate approach) on a bootstrapped sample of experimental sets: primary access to completely fleshed carcasses butchered with stone tools reproducing the HeC model (red alpha bag), secondary access to variously defleshed carcasses in the FeHeH model (yellow alpha bag), and secondary access to defleshed carcasses from the FeH model (blue alpha bag).Data from Domínguez-Rodrigo et al. (2014c).Loading scores for the variables are displayed in Table S6.(For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Fig. 15 .
Fig. 15.Several examples of breakage planes on long bone shafts, some of them with notches at PTK level 22A.

Table 1 A
) Minimum number of individuals represented at PTK level 22A by teeth elements.B) Minimum number of bovids and equids represented by teeth elements, postcranial elements and teeth plus postcranial elements.
a Include four individuals' size 2 that probably belong to Kobus sigmoidalis.b Include the four Kobus sigmoidalis catalogued as size 2 in the postcranial MNI, one equus and three bovids.E. Organista, A. Mocl an, J. Aramendi et al.Quaternary Science Reviews 300 (2023) 107913

Table 2
Skeletal part profiles of small size (size 1e2) carcasses at PTK level 22A.NISP, number of identified specimens; MNE, Minimum number of elements; TM, tooth marked specimens; PM, hammerstone-percussed specimens; CM, cut marked specimens.

Table 4
Spearman's statistics for the relationship between %MAU e density and %MAU e food utility per carcass size.Skeletal element evenness and 95% confidence limits per carcass size.

Table 5
Cut mark frequencies per long limb bone and bone section.ULB, upper limb bone; ILB, intermediate limb bone; LLB, lower limb bone; EPI, epiphysis.Numbers in the numerator are total numbers of cut-marked specimens.Numbers in the denominator are the total numbers of specimens.Numbers in parentheses are percentages.* Not included the indeterminate long bone fragments.

Table 6
Cut mark, percussion mark and tooth mark frequencies on long bones (LB) and midshaft (MSH) sections per carcass size on corrected sample.Numbers in the numerator are total numbers of marked specimens.Numbers in the denominator are the total numbers of specimens.Numbers in parentheses are percentages.