Despite substantial excitement around the use of AI in law, little information exists on the performance and associated risks of the domain's widely marketed tools. Recent work, for instance, has demonstrated the significant potential for "hallucinations"-wherein models make up facts, law, and precedent-leading Chief Justice Roberts to spotlight this risk in his annual report on the judiciary. We argue that there is a need for public AI benchmarking in law. First, relative to other AI application domains, the legal AI ecosystem lacks legibility-there is little information about the design and performance of many commercial legal AI systems. Legal AI has not benefited from the types of benchmarking that have catalyzed, measured, and informed AI innovation and responsible use in other domains. Second, we articulate the challenges of the institutional design of benchmarking. We illustrate how benchmarks can be captured, watered down, and abused. Careful institutional design around the why, who, what, and how of benchmarking will be critical to navigate difficult tradeoffs of transparency, objectivity, expertise, and resources. Third, addressing legal AI's illegibility requires matching institutional models to available resources and constraints. Rather than advocating for a single "best" approach to benchmarking, we show how benchmarking strategies depend on available resources.

There is obscurity about the role of the knee joint during quiet standing (QS) postural control. Examining the knee joint's contribution during QS in individuals with damaged knee proprioception can help resolve the ambiguity. We investigated whether anterior cruciate ligament reconstruction (ACLR) changes the relative contribution of lower-limb joint motions and postural control during QS. This study included 12 individuals with ACLR and 12 able-bodied individuals. A motion capture system and two force plates were used. The acceleration of the joints, the acceleration of the body center of mass (COM), and the COP parameters were calculated during QS under eyes open (EO) and eyes closed (EC) conditions. In the control group under the EO condition, ankle, knee, and hip accelerations, and under the EC condition, ankle and hip accelerations were primary predictors of COM acceleration. In the ACLR group, under the EO condition, the knee and hip accelerations, and in the EC condition, the ankle, knee, and hip accelerations were prominent predictors. The ACLR group exhibited higher COP parameters than the control group in the closed-eyes condition. The knee plays a vital role in QS postural control. Improving knee proprioception modalities and lower-limb joint coordination is important in ACLR rehabilitation.

Despite therapeutic efforts, the majority of individuals post-stroke will experience chronic motor deficits, particularly in the hand. Assistive devices such as hand exoskeletons have the potential to improve hand function, but these instruments must provide sufficient flexibility to support a variety of tasks. This study tests the robustness and utility of a novel device, the Bidirectionally Actuated Cable (BAC)-Glove, which provides independent actuation of each digit. The BAC-Glove supports a variety of control modes, including direct user control through electromyography (EMG). Device capabilities in terms of digit movement and fingertip force production were assessed in 5 neurotypical participants. Kinematics were captured with motion tracking and kinetics were captured with a load cell. Another 10 neurotypical participants completed four training sessions with the BAC-Glove to examine the viability of EMG control. Evaluations were performed at the end of the first and last sessions to examine user performance with the device. The device produced coordinated flexion and extension motions of all three finger joints and peak fingertip forces of 15.1 N in flexion and 9.6 N in extension. Participants improved EMG control of the BAC-Glove over the training sessions, as evidenced by significant improvement on measures of object manipulation with the device, such as the Box and Block test (p < 0.05). BAC-Glove is a robust and potentially useful device for assisting manipulation of objects to perform activities of daily living. The proposed EMG control scheme showed sufficient viability to warrant future examination in clinical populations.

A 31-year-old man presented with an acute, ligamentous perilunate dislocation and paresthesia in the median nerve distribution after a dirt bike accident and underwent closed reduction. The following morning, he was treated with carpal tunnel release and open reduction and internal fixation using a 4-anchor suture tape internal bracing treatment technique through a dorsal carpal approach. The construct consisted of a series of internal braces dunked into four carpal anchors to reinforce/recreate the scapholunate interosseous ligament, lunotriquetral interosseous ligament, and dorsal intercarpal ligament without Kirschner wire retention, with an additional limb used to resist excess scaphoid flexion. The patient was immobilized in a short arm cast for six weeks before initiating full range-of-motion and hand therapy. Full weight bearing and strengthening were allowed at 12 weeks. At six months, the patient reported minimal to no pain, was able to form a full composite fist, and could tolerate a dart thrower's motion. He demonstrated a wrist flexion-extension arc of 90o (25o flexion/65o extension) and reported complete resolution of paresthesia since surgery. No radiographic evidence of carpal instability, fracture, or peri-implant failure was observed, and the Quick Disability of the Arm, Shoulder, and Hand (Quick DASH) score obtained through telephone was 14 out of 100 at 32 months postoperatively.

Pancreaticoduodenectomy remains among the most technically demanding abdominal operations because of complex vascular dissection, multiple reconstructions, and persistent risk of procedure-specific morbidity. Robotic pancreaticoduodenectomy (RPD) has been developed to extend minimally invasive surgery to pancreatic head and periampullary diseases, offering stable three-dimensional visualization, articulating instruments, and improved suturing ergonomics. However, technical feasibility alone does not establish broad clinical value. This structured narrative review maps current evidence on RPD across perioperative outcomes, oncologic safety, clinically relevant postoperative pancreatic fistula (CR-POPF), learning curve, reconstruction, complex cases, cost, and implementation readiness. Evidence was interpreted according to study design, center experience, outcome consistency, and clinical generalizability. Available data support the feasibility of RPD in selected patients treated by experienced teams and suggest possible reductions in blood loss, transfusion, and length of stay compared with open surgery. In contrast, evidence remains less mature for CR-POPF reduction, major morbidity, long-term oncologic outcomes, cost-effectiveness, and routine use in complex vascular or post-neoadjuvant cases. The clinical role of RPD should therefore be determined by alignment among patient complexity, pancreatic remnant risk, surgeon-team maturity, institutional volume, and rescue capacity. We propose an evidence-informed framework for staged implementation, patient selection, program readiness, and predefined conversion triggers. RPD should be viewed as a selective, team-dependent strategy rather than a universal replacement for open pancreaticoduodenectomy.

Achieving the critical view of safety during cholecystectomy is paramount for preventing bile duct injury but contingent on effective gallbladder retraction. Although robotic platforms offer enhanced dexterity and visualization, they present an opportunity to improve the static retraction methods common in laparoscopy. Herein, we describe our advanced dynamic multidimensional retraction (DMR) technique for three-port robotic cholecystectomy. This method uses the external joints of the robotic arms to provide continuous, multi vector exposure of the triangle of Calot, aiming to enhance surgical safety and efficiency. We conducted a retrospective review of the first 131 consecutive patients who underwent a three-port robotic cholecystectomy using our standardized DMR technique between January 2023 and July 2025. All procedures were performed on the DaVinci Xi or X platform by four surgeons during their initial robotic case series. Key demographic, intraoperative, and postoperative data were analyzed. Primary outcome measures included console times, length of hospital stay, intraoperative complications, and conversion rate to other surgical techniques. We performed 131 standardized robotic cholecystectomies via three-arm robot deployment by four teams. The mean patient age was 53 years; 50 were men and 81 were women. All teams started their robotic surgeries sequentially in a 4-months period in 2023. Most cases were elective, seven patients had acute cholecystitis, and 35 patients had a cholecystectomy integrated with other procedures like intraoperative cholangiogram, liver biopsy, repair of postoperative ventral hernia and one case of internal hernia repair after bariatric surgery. The mean console time was 57 min. There was no conversion to other surgical techniques. DMR is a safe and reproducible technique for three-port robotic cholecystectomy. By leveraging the robotic platform capabilities, our method provides non-inferior exposure of the critical anatomy. This advanced robotic approach represents a valuable tool for optimizing the surgical field and may improve safety and efficiency during robotic cholecystectomy. Further investigation in larger, comparative studies is warranted.

Robotic total knee arthroplasty (TKA) enables repeated intraoperative quantification of preoperative alignment, flexion deformity, and medial and lateral compartment gaps in extension and at 90° flexion before component positioning, but the preoperative patterns guiding functional-alignment planning remain incompletely characterized. This study aimed to derive a pragmatic driver-based classification from preoperative robotic data and evaluate its internal coherence across the surgical workflow. We analyzed 68 consecutive primary functionally aligned robotic TKAs performed by a single surgeon using a CT-based robotic-arm platform. Preoperative variables included hip-knee-ankle angle (HKA), flexion deformity, and independent medial and lateral gaps in extension and at 90° flexion. A 10-case formative series informed concept generation; the finalized hierarchy was then applied to all cases using preoperative variables only. Four phenotypes were derived: bone-driven, ligament-driven, flexion-dominant, and mixed/complex (defined as the co-occurrence of two or more severe deformity drivers). Internal coherence was assessed through release escalation, corrected-state behavior, plan-to-final HKA fidelity, and mechanical and functional-alignment balance. Inter-observer reliability was tested by having five independent surgeons classify all cases blinded, with agreement quantified by Fleiss' and Cohen's kappa. All knees were classified: bone-driven, 40/68 (58.8%); ligament-driven, 6/68 (8.8%); flexion-dominant, 10/68 (14.7%); and mixed/complex, 12/68 (17.6%). Release escalation occurred in 1/68 (1.5%), with no posterior capsulotomy. Overall, 58/63 (92.1%) finished within ± 2° of planned HKA; strict mechanical balance was achieved in 58/62 (93.5%) and functional-alignment balance in 61/62 (98.4%). Flexion-dominant knees showed lower coronal fidelity but complete final balance, suggesting a coronal-sagittal trade-off during functional planning. Inter-observer agreement among the five surgeons was almost perfect (Fleiss' kappa 0.88, 95% CI 0.81-0.93; 92.4% overall agreement). Preoperative HKA, sagittal deformity, and medial-lateral gap asymmetry revealed recurring patterns organized into a four-phenotype driver framework. This classification is best interpreted as an internally coherent derivation framework that showed almost perfect inter-observer reliability (Fleiss' kappa 0.88) but still requires external, multicenter validation and threshold-sensitivity analysis before its clinical utility can be established.

Rhodococcus equi is an important bacterium that affects foals up to six months of age, especially those that are immunocompromised. The main manifestation is pyogranulomatous pneumonia; nevertheless, extrapulmonary lesions have not only been reported, but are not uncommon, including arthritis and osteomyelitis. Despite the widespread occurrence of rhodococcosis, the macro- and microscopic lesions in the musculoskeletal system are still poorly understood, and a detailed study is needed to better understand the pathogenic potential of the organism. The aim of this report is to present the anatomopathological aspects of bone and joint lesions associated with Rhodococcus equi in two foals. The first case was a two-month-old male Quarter Horse foal with suspected cervical trauma. This horse was admitted in lateral recumbency, semi-comatose, and with atlantoaxial region enlarged. Necropsy revealed atlantooccipital arthritis with pyogranulomatous osteomyelitis, resulting in pathologic fractures of the atlas. There was also pyogranulomatous pachymeningitis leading to compressive myelopathy. In the second case, a two-month-old Brazilian Sport Horse foal presented with lameness of the left pelvic limb and enlargement of the left femorotibial and femoropatellar and right scapulohumeral joints. Necropsy revealed femorotibial and femoropatellar arthritis and purulent osteomyelitis in the femur, as well as extensive necrosis and bone rarefaction with focal detachment of the epiphyseal plate. Both animals had pyogranulomatous pneumonia and lymphadenitis with extracellular and intracytoplasmic gram-positive coccobacilli in neutrophils and macrophages. In the first case, R. equi was isolated and confirmed by MALDI-TOF. R. equi infection should be considered in the differential diagnosis of osteoarticular lesions in foals.

AI models are increasingly explored for radiographic assessment of knee osteoarthritis, but their reliability as an experimental model for KL grading remains uncertain. This study evaluated ChatGPT-5 agreement with expert consensus for KL grading and compared its performance with a secondary AI model. This controlled cross-sectional observer agreement analysis used 630 hand-compiled synthetic posteroanterior fixed flexion knee radiographs selected for technical adequacy, anatomical suitability, and interpretability. Three experienced clinicians, one orthopedic surgeon and two rheumatologists, independently graded all images, and a three-rater majority-consensus expert reference standard was established. ChatGPT-5 assessed the same images using a standardized prompt and image-level protocol. Gemini 2.5 Flash assessed a 112-image subset. Agreement was analyzed using weighted Cohen's κ, Gwet's AC2, and mean absolute error (MAE). Binary classification behavior was evaluated at the KL ≥ 2 threshold. Inter-rater agreement before consensus formation was high: Fleiss' κ was 0.884, Gwet's AC2 was 0.921, and pairwise weighted κ ranged from 0.902 to 0.929. Complete three-rater agreement was observed in 520 radiographs (82.5%), and only 14 cases (2.2%) required adjudication. Compared with the expert reference standard, ChatGPT-5 showed moderate-to-substantial agreement, with weighted κ = 0.680 (95% CI 0.616-0.737), MAE = 0.589, and 60.9% exact five-grade agreement. Exact agreement was highest for KL 0 (81.5%) and KL 4 (66.9%) and lower for intermediate grades. For binary KL ≥ 2 classification, ChatGPT-5 correctly classified 513/630 radiographs, with 81.4% accuracy (95% CI 78.2-84.3), 82.1% sensitivity, 80.3% specificity, 87.6% PPV, and 72.6% NPV. A small but significant shift toward higher grading was observed (p = 0.013), with up-grading more frequent in KL 0-1 and down-grading more frequent in KL 3-4. In the 112-image exploratory subset, Gemini 2.5 Flash showed limited agreement with the expert reference standard (weighted κ = 0.23, MAE = 0.66, exact agreement = 50.0%) and poor agreement with ChatGPT-5 (weighted κ = 0.06). For KL ≥ 2 classification, Gemini achieved 87.5% accuracy, 100% sensitivity, 0% specificity, and 87.5% PPV; NPV was not estimable because no KL < 2 classifications were generated. In this controlled synthetic radiograph study, ChatGPT-5 showed moderate-to-substantial agreement with expert consensus under standardized experimental conditions. However, the curated synthetic dataset, limited technical reproducibility of web-based inference, systematic grading tendencies, exploratory nature of the secondary-model analysis, and limited ecological validity preclude any inference of clinical readiness or real-world diagnostic performance without validation on real-world, multicenter, prevalence-based knee radiograph datasets.