Engineering

Abstract: Radio-frequency (RF) reflection measurements provide a non-ionizing approach for estimating biological tissue thickness; however, accurately resolving deeper tissues is difficult because their reflections are attenuated and superimposed with those from overlying layers. This study presents a physics-guided stacked-regression framework for estimating subcutaneous adipose tissue (SAT), muscle tissue (MT), and visceral adipose tissue (VAT) thicknesses from simulated scattering parameters. A planar multilayer abdominal model containing 120 combinations of SAT, MT, and VAT thicknesses was evaluated. The input features consisted of the magnitude of S₁₁ at ten frequencies from 300 MHz to 2 GHz, including four intermediate frequencies at 600, 700, 800, and 900 MHz that were absent from the previous model. Target-specific frequency subsets were selected and evaluated using polynomial regression and five-fold cross-validation. Multilayer reflection theory was used to explain the nonlinear relationship between tissue thickness and the measured RF response. The extended-frequency models achieved errors of 1.96%, 7.60%, and 19.86% for SAT, MT, and VAT, respectively, compared with 2.56%, 11.42%, and 39.10% in the previous model. A selective stacked architecture was subsequently evaluated by using an estimated overlying tissue thickness as an additional input for a deeper-tissue estimator. Adding predicted SAT thickness degraded MT estimation, whereas adding predicted MT thickness improved VAT estimation. The final stacked VAT model achieved an RMSE of 0.14±0.11 mm, an R² of 0.97±0.02, and an error of 12.49%, corresponding to a 68.06% error reduction relative to the previous VAT model. These findings demonstrate that denser frequency sampling and selectively applied anatomical guidance can improve RF-based estimation of deeper tissue layers.

Abstract: Wearable and handheld devices record a Lead I-like electrocardiogram (ECG), but it is unclear whether a single lead retains the information that deep learning extracts from the 12-lead ECG for coronary disease. We developed two one-dimensional residual network models, oneusingthe12-leadECGandonerestrictedtoLeadI,withidenticalpreprocessing, patient-level partitions and five-fold cross-validation, to detect coded coronary artery disease/coronary heart disease (CAD/CHD). The cohort comprised the most recent ECG of 161,306 adults in MIMIC-IV-ECG linked to MIMIC-IV, a hospital-based population spanning emergency, inpatient (including intensive care) and outpatient settings, in which controls included patients with other cardiovascular disease. In a shared holdout cohort of 31,497 patients (943 cases), the area under the receiver operating characteristic curve (AUC) was 0.764 (95% CI 0.751-0.777) for the 12-lead model and 0.744 (0.730-0.759) for Lead I (p < 0.001). Predictions were strongly correlated (Pearson r = 0.78), binary classifications agreed in 81.8% of patients, and sensitivity did not differ between models (75.0% vs. 74.0%, p = 0.55). Both models also discriminated heart failure with reduced ejection fraction, including after excluding CAD/CHD. Grad-CAM showed that both models shifted attribution towards the ST segment in ECGs classified as positive, although their attribution profiles were only partly shared. Adding the Lead I prediction to age, sex and race/ethnicity increased the AUC from 0.734 to 0.773. A single Lead I ECG retained most, but not all, of the CAD/CHD-related information available to the 12-lead model.

Abstract: Cardiovascular diseases remain the leading cause of death worldwide, and most of this burden falls on low- and middle-income countries where access to electrocardiography (ECG) is limited. In the last decade, integrated biopotential front ends, open-source microcontroller platforms and ubiquitous smartphones have made it possible to build ECG monitors at a fraction of the cost of conventional equipment, while consumer wearables have shown that single-lead ECG can be recorded at population scale. This narrative review examines how low-cost, open-source and wearable ECG systems are designed and evaluated for remote cardiac monitoring. We describe the end-to-end architecture (electrodes, analog front ends, microcontrollers, wireless links, gateways and cloud back ends), the signal processing chain (denoising, signal quality assessment, QRS detection, delineation and heart rate variability), and the emerging role of artificial intelligence running on the cloud or directly on the device. We then discuss validation methodology, including reference databases, standardized metrics and agreement analysis, and summarize the regulatory, safety and data protection requirements that separate a research prototype from a medical device. An Arduino-based remote ECG system previously developed by the author is used as a reference case to illustrate common design choices and pitfalls. We conclude that technical feasibility is no longer the bottleneck; rigorous validation in target populations, signal quality management and regulatory-grade engineering now determine clinical impact.

Abstract: Antimicrobial resistance (AMR) and biofilm-associated infections continue to compromise the effectiveness of conventional antibiotics, highlighting the need for complementary therapeutic approaches. Nanoparticle-based systems offer two distinct but potentially convergent strategies: direct antibacterial activity and improved delivery of antimicrobial agents. This narrative review critically examines inorganic, organic, and hybrid nanoplatforms, including metal and metal-oxide nanoparticles, carbon-derived materials, mesoporous silica, polymeric carriers, lipid-based systems, and stimuli-responsive formulations. Their mechanisms of action, physicochemical determinants of efficacy, and potential applications against multidrug-resistant bacteria and biofilm-associated infections are discussed. Particular attention is given to the distinction between bacterial resistance, tolerance, and persistence, as well as the challenges associated with nanoparticle penetration, antimicrobial synergy, and reproducible antibacterial activity. Although preclinical findings demonstrate considerable therapeutic potential, translation into clinical practice remains limited. Inhaled liposomal amikacin illustrates the clinical relevance of indication-specific nanodelivery, while emphasizing that therapeutic benefits cannot be generalized across nanoparticle formulations. Major challenges include cytotoxicity, bacterial adaptation, biodistribution, stability in biological environments, manufacturing scalability, regulatory requirements, and environmental safety. Future research should prioritize standardized characterization, clinically relevant infection models, robust comparisons with conventional treatments, and integrated assessments of safety and resistance evolution. Within a One Health framework, nanoparticle-based technologies may contribute to more effective antimicrobial therapies, provided that their development is guided by demonstrable clinical benefits, responsible antimicrobial use, and consideration of their broader environmental impact.

Abstract: Ag/AgCl gel electrodes are the clinical standard for electrocardiogram (ECG) recording, but at the skin–electrode biointerface, gel drying and skin irritation limit accurate recording over several days. To explore gel-free alternatives, we sputtered gold over a titanium adhesion layer onto serpentine textile dry electrodes and examined how Au:Ti thickness affects electrical, electromechanical and simulated ECG performance. Four variants (50:10, 80:15, 100:15 and 100:20 nm) were characterised by scanning electron microscopy, collinear four-probe measurement and mandrel bending, and the textile geometry by tensile testing. The two best coatings were then compared with commercial Ag/AgCl electrodes on a two-layer skin phantom using impedance measurements and simulated ECG acquisition. All coatings lowered sheet resistance relative to the uncoated textile, with the 80:15 nm variant giving the lowest value. The 100:20 nm variant was the most stable under bending and showed the least post-bend surface damage. The serpentine track did not significantly reduce bend-induced resistance change. On a porous phantom, both dry electrodes recorded identifiable ECG waveforms; mean signal-to-noise ratio was 24.4 ± 0.2 dB for Ag/AgCl, 15.9 ± 3.9 dB for 80:15 nm and 18.4 ± 0.5 dB for 100:20 nm (p = 0.061). Au:Ti-coated textiles are a feasible, gel-free biointerface for conformable ECG electrodes, but motion, durability and on-skin testing are required.

Abstract: Calibration remains a practical barrier for steady-state visual evoked potential (SSVEP) brain–computer interfaces. Task-related component analysis (TRCA) is accurate when several labelled trials per class are available, but its performance degrades when calibration is reduced to one trial and the test block follows a change in recording conditions. We asked whether a fixed modification of ensemble TRCA (eTRCA) could reduce this loss without any per-subject hyperparameter tuning. The proposed SST-eTRCA appends a three-sample delayed copy of each channel, z-scores observations before covariance accumulation, and pools 80 class filters into one projection. On the public 35-subject Tsinghua benchmark (40 classes, six posterior channels), it reached 95.6% with one calibration trial per class versus 93.4% for eTRCA (1400 decisions; McNemar p = 1.5E-5), and 95.0% versus 91.8% without extra calibration. The advantage was largest at 1 s (68.9% versus 63.9%). Under leave-one-subject-out, SST-eTRCA and eTRCA were indistinguishable (83.6% each), and self-training changed accuracy by only 0.1 points (p = 0.75). The gain is therefore specific to within-participant block shift, not to cross-subject transfer. These results define a narrow but useful operating range: a fixed, reproducible decoder for reducing recalibration across blocks, with the negative self-training result serving as a reliability check.

Abstract: Poly(vinyl alcohol) (PVA) hydrogels are promising for meniscal replacement, but reproducing native meniscal anisotropy remains unsolved. We developed a fabrication strategy for imparting controlled anisotropy to PVA hydrogels. We hypothesized that modulus and alignment would increase with applied strain, and that circumferential strain would produce direction-dependence in a meniscal-like geometry. Dogbone specimens at three PVA concentrations were strained 0–200% during freeze–thaw processing. Meniscal-like specimens were fabricated at 27.5% PVA under 200% circumferential strain. Constructs were characterized by tensile testing, stress-relaxation compression, and polarized light microscopy. In dogbone specimens, tensile modulus rose exponentially and coherency linearly with applied strain (adjusted R² > 0.92; R² ≥ 0.96). In meniscal-like specimens, stretching increased circumferential peak stress (587 ± 173 vs 439 ± 96 kPa), and direction-dependence appeared only after stretching. Stretched constructs showed higher optical coherency (p = 0.002, r = 0.77), concentrated in the outer third of the cross-section. Compressive moduli trended lower in stretched constructs, reaching significance anteriorly. Strain applied during freeze–thaw processing programs direction-dependent stiffness and alignment into a homogeneous hydrogel, though absolute stiffness remains below native tissue. More strain during processing is the most direct route to a stiffer construct, and further development is required before functional replacement.

Abstract: Atrial fibrillation (AF) is a heart rhythm disturbance affecting approximately 1% of the global population, being the most common cardiac anomaly. Visual inspection of extensive electrocardiograms (ECGs), patient variabilities, and long-term dependencies make manual AF diagnosis difficult and heavily reliant on a cardiologist’s expertise. Aiming to aid medical diagnosis, this study develops: an innovative R-peak detector to correct database annotation inaccuracies, and machine learning classifiers to enhance AF detection from long-term raw ECGs. The R-peak algorithm combines self-convolved windows filtering, signal derivatives, Shannon energy, and the Hilbert transform. Using the MIT-BIH Atrial Fibrillation database, extracted ECG features included RR intervals (RRi) durations, Jitter and Shimmer disturbances, Shannon and logarithmic energy entropies, instantaneous frequencies, spectral entropy and Wavelet Scattering Transform (WST) coefficients. These were evaluated using Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM) networks. To our knowledge, this is one of the first times that acoustically inspired parameters, Jitter and Shimmer, have described ECG signals. The proposed R-peak algorithm achieved a high accuracy of 98.95%. For AF classification, the best accuracy results were 91.96% for MLP and 98.17% for LSTM, which matches typical state-of-the-art values (≈ 98.00%). Findings indicate that RRi remained the most discriminatory feature, while incorporating P- and T-waves entropies and Jitter significantly enhances performance. Besides LSTM’s superior performance, MLP approaches validate that precise feature engineering allows simpler models to approach state-of-the-art results, highlighting their potential on times of deep learning overuse and of searching for explainable AIs in clinical practice.

Abstract: Long-term fluorescence imaging of live cells enables visualization of dynamic cellular processes but often requires expensive, specialized instrumentation for operation within cell-culture incubators. We developed a low-cost, modular fluorescence microscope based on the open-source UC2 framework for automated time-lapse imaging under incubator conditions. The system uses standardized UC2 modules for motorized objective positioning and camera mounting, alongside custom 3D-printed modules for laser illumination, beam expansion, neutral-density filtering, dichroic beam splitting, tube-lens imaging, and sample mounting. An infinity-optics configuration allows optical components to be readily added or rearranged to accommodate different imaging modalities and incubator constraints. A Raspberry Pi controls objective positioning and laser illumination through custom Python software, enabling automated time-lapse acquisition. This work demonstrates a reconfigurable, low-cost fluorescence microscope potentially capable of automated long-term imaging of cell behavior responses and provides an adaptable platform for live-cell imaging applications in constrained laboratory environments.

Abstract: This review systematically examines the mechanisms underlying the efficacy of Cold atmospheric plasma (CAP), with particular emphasis on the interaction pathways between reactive oxygen and nitrogen species (RONS) and biological or chemical contaminants. It evaluates the impact of key processing parameters and food matrices on treatment outcomes and synthesizes recent advances in CAP applications for mitigating diverse food hazards. Limitations and current research gaps are discussed, especially regarding the complex interactions between plasma-generated species and food components. Future research directions are outlined, focusing on molecular-level mechanistic studies, the development of tailored treatment protocols for specific food categories, and the integration of real-time monitoring systems to enable precise, optimized plasma processing.

Abstract: Background: Cognitive resources are allocated based on task difficulty, duration and immersion level. The development of virtual reality (VR) technology, based on head-mounted displays (HMDs), has led to increased immersive user experiences, affecting mental load and effort, as measured by Electroencephalography (EEG). However, brain dynamics under different difficulty and immersive levels have not been fully described. Methods: EEG recordings and microstate (MS) analysis were used in classical desktop (DSK) and HMD-based stimulation environments to explore the effects of cognitive demand and immersion level on brain dynamics and their relationship with attention domains. Twenty-seven healthy participants performed a two-level N-back task in both environments, while EEG signals were acquired. Brain dynamics were evaluated through MS-based dynamic parameters and power-based indices were computed to estimate attention and engagement. Results: The increased task difficulty modulates EEG microstate dynamics, regardless of the immersion level. Specifically, the highest mental demand was associated with an increased presence of MS D, related to the task active network, confirming its involvement in cognitive tasks. Conclusions: Although immersion level did not significantly influence MS dynamics, it affected the relationship with attention level, with a stronger correlation observed in the DSK condition and a weaker one in the HMD scenarios.

Abstract: Background: Telerehabilitation can improve access to rehabilitation services for older adults, but limited clinical supervision may increase the risk of undetected compensatory movements. This study evaluated whether inertial measurement unit (IMU) based machine learning model could automatically detect compensations during rehabilitation exercises. Methods: Five exercise-specific classifiers were developed using a controlled dataset of compensatory movements performed by fifteen healthy kinesiologists. For each exercise, 450 correct and 450 compensatory repetitions were collected while participants wore both the prototype and commercial IMUs. Biomechanical and statistical features were extracted, and LASSO regression was applied for feature selection. Five different classifiers were optimized through grid search and evaluated using subject-wise leave-one-out cross-validation. The selected models were subsequently validated in eleven community-dwelling older adults aged 69.7 ± 7.1 years. Results: In controlled conditions, the prototype achieved performance comparable to the commercial IMU (accuracy and precision ≥ 95%). In older adults, accuracy ranged from 67% to 96%, whereas precision ranged from 32% to 100%, indicating exercise-dependent differences. Conclusions: A low-cost IMU combined with lightweight machine learning can detect compensatory movements with performance comparable to a research-grade sensor under controlled conditions. Larger clinical samples, different exercises and multi-level compensation labels are required to improve generalizability and support real-time telerehabilitation feedback.

Abstract: This paper proposes a tri-modal speech reconstruction method that synergistically fuses laser speckle vibrometry, depth-camera articulatory capture, and ultra-wideband (UWB) radar. By exploiting the RVTALL multimodal dataset, the method extracts highly complementary physical information: laser signals capture physiological vibrations homologous to voice sources, depth cameras record articulatory geometry, and UWB radar detects radio-frequency micro-motions. To maximize the utility of these heterogeneous data streams, the framework adopts a two-stage training paradigm, where modality-specific encoders are pretrained with an auxiliary classification objective and then frozen for label-free log-mel spectrogram regression. A hybrid loss combining global spectral consistency and local dynamic constraints is designed to optimize fine-grained formant transitions and energy onsets. The proposed method demonstrates significant superiority over single-modality approaches without requiring any textual or phonetic supervision. Experimental results under a strict leave-two-subjects-out protocol show that the full tri-modal system achieves an 81.7% word-level Whisper accuracy, a 16.3% absolute improvement over the single laser modality (65.4%). Furthermore, the wav2vec 2.0 sentence-level cosine similarity reaches 0.932 (compared to 0.799 for laser only), and the word and sentence mean opinion scores (MOS) reach 3.08 and 2.67, respectively. These results confirm that the rich physical complementarity within the tri-modal dataset enables highly intelligible, natural, and noise-resilient speech recovery.

Abstract: Biomechanical analysis of human movement supports rehabilitation decisions requiring objective and reproducible measurements. However, motion capture technologies are commonly evaluated as separate technological families. This critical review compares marker-based optoelectronic, markerless, inertial, and hybrid systems in terms of accuracy, clinical validity, reliability, and implementation feasibility, and proposes methodological criteria for future validation studies. Scopus, Web of Science, PubMed, and IEEE Xplore were searched, supplemented by forward and backward citation searching from eight seed studies. The selection process was reported using a PRISMA 2020 flow diagram. No protocol was registered, and no formal risk-of-bias assessment was conducted. Eligible studies involved human participants in clinical, rehabilitation, or functional assessment settings and reported quantitative performance metrics. Evidence was synthesized by technological family and application scenario. Ninety-six documents were included. Marker-based systems provided the highest accuracy for joint kinematics, whereas markerless and inertial systems showed comparable validity for spatiotemporal variables and global movement characterization. However, errors in specific joint angles remained above clinically desirable thresholds. Calibration procedures and assessment duration substantially influenced performance. No technology was universally superior. Therefore, selection should be made by variable, task and population, using explicit validation criteria.

Abstract: Fluid-Structure Interaction (FSI) has been the standard when simulating cerebral aneurysms. Nonetheless, obtaining a realistic aneurysm wall geometry without ex-vivo methods has proven difficult. In this article, a nodal assignment method based on empirical distributions is presented, and the effect of wall thickness is studied using one-way FSI simulations for normal and hypertensive pressure cases. Results show that while generally there is an inverse relation between structural parameters (Von-Mises stress and de-formation) and wall thickness, the final distribution of these parameters is defined by a balance between the fluid dynamics (Wall Shear Stress and hemodynamic pressure) and the thickness distribution. Comparisons between the thickness distribution and fluid dynamics were made to illustrate how these two factors affected the final structural parameter contours in the variable assignments, where hemodynamic pressure and aneurysm size were found to influence the leading factor in this duality.

Abstract: Hydrogels are being investigated as delivery platforms for stem cells and extracellular vesicles (EVs) in regenerative medicine, but their relative advantages remain unclear. The objective of this critical narrative literature review is to compare stem‐cell‐loaded and EV‐loaded hydrogels in terms of biological activity, cell viability where applicable, retention and controlled release, stability, storage, safety, and clinical development. Published scientific studies were critically examined, considering experimental conditions, reported outcomes, methodological limitations, and levels of evidence. Stem‐cell‐loaded hydrogels can support living cells and their dynamic interactions with surrounding tissues but require maintenance of cell viability and control of cellular behavior. EV‐loaded hydrogels offer a cell‐free approach that avoids the administration of living therapeutic cells, although vesicle characterization, stability, dosing, and effective delivery remain challenges. Preclinical studies suggest improved local retention and sustained EV release, along with regenerative and immunomodulatory effects in specific experimental models. However, differences in hydrogel formulations, experimental conditions, and evaluation methods limit direct comparisons, while clinical effectiveness remains insufficiently established. Stimuli‐responsive hydrogels may provide additional control over EV delivery and regulate the microenvironment surrounding encapsulated stem cells, although their therapeutic superiority over conventional systems has not been demonstrated. Overall, neither strategy can currently be considered universally superior. Further research involving standardized characterization, direct comparative studies, long‐term safety evaluations, and clinical validation is necessary to determine their relative therapeutic potential.

Abstract: UpperMyo is an open six-muscle surface electromyography (sEMG) dataset for upper-limb action detection, temporal prediction, and cross-dataset transfer, evaluated through internal validation and paired external experiments. Analyses use 264 recording blocks from 24 healthy male participants, sampled at 1 kHz with 12-bit resolution; the nine-class dataset contains 2,640 four-second samples spanning eight actions and background. Internal evaluations cover recognition, channel ablation, and elbow-flexion/rest temporal classification. A convolutional neural network (CNN) achieved a macro-F1 of 0.9860 under nested leave-one-subject-out validation. Temporal classification yielded CNN accuracies of 89.5% and 75.3% at 0 and 2,000 ms backward offsets from an sEMG onset reference. Four external experiments compare transfer with target-only learning at identical labeling budgets for matched actions, new actions, activities without class mapping, and continuous grip-force estimation. WristMotion balanced accuracy increased by 14.21 percentage points with four labeled trials; mean flexor carpi ulnaris (FCU) grip-force R2 increased from 0.6431 to 0.6899 at the full budget. Transfer yielded gains across multiple tasks and budgets, with benefits varying by task and labeling budget. UpperMyo provides traceable data and evaluation evidence for sEMG action analysis and cross-dataset learning.

Abstract: Improving the stability of protein–mineral interfaces while preserving biological activity is a major challenge in biosensor and functional biomaterials design. Ancient tooth enamel offers a unique opportunity to identify protein sequences that have persisted in mineralized environments for millions of years. Here we investigate whether archaeological protein evidence can guide the selection of new peptide candidates for biomaterials engineering. We analyzed amelogenin (AMELX) peptide coverage from 12 ancient enamel specimens dating from approximately 0.4 to 2 million years ago. The data reveal, beyond the established C-terminal hydroxyapatite-binding region, two new peptides regions, archaeologically persistent regions (APR1 and APR2), with substantial archaeological coverage (82.7% and 100%, respectively) and different chemical and physicochemical properties from the C-tail, including differences in electrical charge, isoelectric point, and hydropathy. These findings identify additional AMELX sequences for investigating protein–mineral interactions and suggest that archaeological proteomics can guide the selection of new peptide motifs and candidate protein-mineral interfaces whose potential to improve biosensor durability can be evaluated experimentally.

Abstract: Wearable sensing technologies have enabled continuous, non-invasive monitoring of physiological signals, accelerating advances in neuromuscular assessment, rehabilitation, and human–machine interaction. Among these technologies, surface electromyography (sEMG) has become a cornerstone for decoding muscle activity. However, the reliability of sEMG remains highly dependent on user–device interaction factors—including skin–electrode impedance, contact pressure, and skin temperature—that introduce signal variability, degrade long-term stability, and limit the generalizability of machine learning models. Despite their importance, these interaction variables are rarely acquired synchronously with sEMG in existing wearable systems, limiting systematic investigation of their influence on signal quality. Here, we present MyoProBand, a compact multimodal forearm wearable that records multichannel sEMG along with skin–electrode impedance, contact pressure, skin temperature, and inertial measurements. By combining neuromuscular, physiological, and mechanical sensing within a synchronized acquisition framework, the platform enables the characterization of the factors governing sEMG reliability during prolonged use. The MyoProBand incorporates custom analog front-end electronics, multimodal sensor conditioning, and embedded signal synchronization while maintaining a lightweight form factor suitable for experimental and translational research. We evaluated the system in healthy participants to assess hardware performance, multimodal signal synchronization, and operational stability during forearm muscle activity. The results show that the sensing modalities can be recorded within the same experimental session while preserving the primary sEMG acquisition and provide initial observations of how interface variables change during use. MyoProBand is intended as a research platform for collecting multimodal datasets that can support later studies of sEMG variability and robust myoelectric control.

Abstract: Background: Stem cell bioprocessing requires timely assessment of cell state and control over variable expansion and differentiation. Automated culture can execute predefined steps, whereas adaptive manufacturing uses observations to guide decisions during an ongoing batch.Objective: This review examines how artificial intelligence and machine learning connect process monitoring to prediction and feedback control, and assesses which links have been demonstrated in stem cell systems.Key Content: This topic covers image-based phenotyping, physical and multimodal sensing, growth and differentiation forecasts, process optimization, hybrid models, and proposed digital twins. A descriptive Level 0–6 continuum distinguishes endpoint assessment, monitoring, state estimation, prediction, operator decision support, within-run feedback, and coordinated workflow autonomy. Reported studies demonstrate image-guided adjustments to differentiation, robotic optimization across experiments, and rule-based feedback within a culture. However, these capabilities do not demonstrate autonomous production of a clinically releasable cell product. Translation is further limited by reduced model performance when entire experiments are withheld from training, fragmented datasets, variation across donors and processes, uncertain links between surrogate measures and potency, and the need for model governance.Conclusion: Supervised adaptation and supervised autonomy are credible near-term targets, characterized by bounded, traceable interventions with human oversight and a defined response to uncertainty. Prospective studies should compare adaptive and fixed processes across independent batches and assess cell identity, relevant function, safety-related attributes, and resource use. Claims of full autonomy require verified performance across the manufacturing workflow.