Large language models (LLMs) are increasingly deployed in politically sensitive contexts, raising concerns about their susceptibility to ideological biases. In this work, we examine how synthetic persona conditioning shapes ideological expression across seven open-weight instruction-tuned models (7B-72B parameters) using the Political Compass Test (62 statements) as a standardized behavioral probe. Across three studies involving 200,000 synthetic personas and more than 260 million model responses, we analyze implicit and explicit malleability, as well as theme-associated variation. We find that: (i) larger models exhibit broader implicit ideological coverage, increasing from 14%-35% for 7-8B models to up to 49% for 70B+ models; (ii) explicit ideological priming induces large and statistically significant shifts, with right-authoritarian cues moving all models in the intended direction and generally producing larger shifts than left-libertarian cues across model-axis comparisons; (iii) left-libertarian priming produces more heterogeneous responses, with counter-directional economic shifts in three of four 7-8B models and intended-direction shifts across all 70B+ models; and (iv) theme-associated semantic content in persona descriptions is linked to systematic and interpretable directional shifts in ideological space. While our results identify an upstream mechanism through which persona conditioning can alter model responses under a standardized ideological probe, we do not test whether such shifts affect users' beliefs, decisions, or political behavior. Our findings are best understood as evidence of ideological malleability at the generation layer, highlighting the need to account for interactional factors when evaluating political neutrality, fairness, and safety in English-prompted, persona-conditioned language models.
Purpose To evaluate agreement of proposed CT-based variants with the Vizzielli laparoscopic risk-adjusted score in suspected advanced ovarian cancer and assess inter-reader agreement of CT-derived components. Methods This retrospective single-center study included consecutive adults evaluated between January 2021 and November 2025 for suspected advanced ovarian cancer who underwent contrast-enhanced abdominopelvic CT and staging laparoscopy. Proposed CT-based modifications were operationalized as six score variants combining CT-derived peritoneal tumor burden, CT-estimated ascites, cancer antigen 125 (CA-125), and either Eastern Cooperative Oncology Group performance status (ECOG-PS) or CT-derived body composition parameters. Agreement with the laparoscopic risk-adjusted score was assessed using the intraclass correlation coefficient (ICC), Spearman correlation coefficient, and weighted Cohen’s kappa. Results The final cohort included 80 patients. All CT-based variants showed agreement with the laparoscopic risk-adjusted score. For the variant combining CT-derived PCI, CT-estimated ascites, ECOG-PS, and CA-125, the consistency ICC was 0.76, Spearman’s ρ was 0.806, and weighted Cohen’s kappa was 0.76. This variant showed 70.0% agreement across the three risk categories and 83.8% agreement for high-risk versus non-high-risk classification, classifying 20 of 22 laparoscopic high-risk patients as high risk. The PCI-based skeletal muscle index (SMI) variant maintained substantial agreement: ICC, 0.73; Spearman correlation coefficient, 0.742; weighted Cohen’s kappa, 0.74. Inter-reader agreement was high for CT-derived PCI, with an ICC of 0.91. Conclusion Proposed CT-based variants showed agreement with the Vizzielli laparoscopic risk-adjusted score. The PCI-based ECOG-PS variant showed the highest agreement metrics, suggesting that structured CT assessment may help approximate laparoscopic risk categories before invasive evaluation.
Background Multiple sclerosis (MS) is an immune-mediated central nervous system disorder; relapsing-remitting MS (RRMS) is the most common form. Self-perceived health-related quality of life (HRQoL) has been linked to psycho-behavioral and immunological factors, but their interplay remains unclear. We examined psycho-behavioral, clinical, and immunological correlates of perceived physical and mental health. Methods In this cross-sectional study, 85 patients with RRMS (78% female; mean age 40.6 ± 10.3 years; predominantly mild disability) were assessed for fatigue, sleep quality, daytime sleepiness, anxiety/depression symptoms, and perceived cognitive dysfunction. Blood samples were analyzed for complete blood counts and, in a subgroup, lymphocyte subpopulations; the systemic immune-inflammation index (SII) was calculated. HRQoL was measured using the SF-36 Physical and Mental Component Summaries (PCS and MCS). Hierarchical linear regression models included sociodemographic, clinical, and treatment-class covariates, with sensitivity analyses. Results Disability, fatigue, and depressive symptoms were associated with lower PCS, and sleep disturbances and depressive symptoms with lower MCS. Higher lymphocyte count was associated with higher MCS (β = +0.268, p = 0.006; ΔR2 = 0.030), consistently across all four sensitivity analyses. SII was not significantly associated with either outcome. In exploratory flow-cytometry analyses (n = 33–34), NK cells were associated with PCS (β = +0.345, p = 0.003), but not significantly with MCS. Conclusions HRQoL in RRMS showed psychological and clinical associations, with an additional lymphocyte-MCS association. The clinical relevance of the immune findings, including the exploratory NK-PCS association, requires further investigation.
“Flavescence dorée” (FD) is one of the most impactful diseases for European viticulture, characterized by a complex epidemiology. While Scaphoideus titanus is the primary vector and Vitis vinifera (grapevine) the main host, several wild plant species, such as Alnus glutinosa, Ailanthus altissima, Clematis vitalba, Corylus avellana, and Salix spp., can serve as reservoirs for certain FD phytoplasma strains. This study aims to determine whether wild plants represent a source of inoculum for vineyards in North-Eastern Italy. Phytoplasma strains from grapevine and wild plants were collected across three provinces, identified via the 16S rRNA gene, and further characterized based on map and vmpA genes. The results demonstrated that the phytoplasma genotypes (M54 and M3) detected in grapevine were typical of the surveyed area. Furthermore, wild plants harbored genotypes previously detected in FD outbreaks and transmissible by S. titanus in other European regions, with the exception of the M3 genotype, reported here for the first time in wild plants in North-Eastern Italy. For instance, regardless of the sampling area, C. vitalba harbored strains ascribable to the M51 genotype, which is prevalent in Serbia. Similarly, A. glutinosa hosted the M50 genotype, which is associated with FD outbreaks in several countries. Analysis of the vmpA gene confirmed that phytoplasma strains belonging to VmpA-II cluster colonize only grapevine in the studied areas, while a few grapevines and all wild plants contained strains of VmpA-III cluster. Overall, these results suggest that grapevine and wild plants represent two rather distinct compartments in FD epidemiology within this region.
Control-flow analysis in the style of Flow Logic has long been a central tool in the Pisa school of language-based security: it provides an over-approximating static semantics whose soundness ensures that statically established security properties are preserved at run time. In this paper we revisit a representative instance of this programme, the static analysis of tuple trajectories in a language of the KLAIM family, and certify it in the Rocq proof assistant. We work with the located-network fragment on which the original trajectory analysis is defined: first-order tuple-space coordination over a finite, statically known set of localities, with no code mobility. For this fragment we mechanise the syntax, operational semantics and Flow Logic specification of the analysis using a locally nameless representation with multi-binders of statically varying arity. We then machine-check the main metatheoretic properties of the analysis: closure of acceptable estimates under meets, yielding a least acceptable estimate when one exists; subject reduction; and a trajectory-confinement theorem stating that sensitive data never reach forbidden sites whenever the static estimate excludes them. The mechanisation also exposes a methodological point that is easy to leave implicit on paper. The 0-CFA distinguishes binders by their canonical labels, and is therefore equivariant, rather than invariant, with respect to alpha-renaming: renaming binders preserves acceptability only when the abstract environment is renamed accordingly. Making this naming discipline explicit reconciles locally nameless binding with finitely supported abstract environments. Finally, acceptability is proved decidable, and the decision procedure is extracted to OCaml, yielding a certified checker for the analysed fragment.
