Jered Mcclain, Erydir Ceisiwr, SEXA Institute of Technology and Interdomain Galactic Advisors, United States of America
Autonomousaero space and mech anical systems require supervisory architecture scapable of determining whether candidate operating states remain mutually consistent before execution. This paper evaluates a hardware-verifiable recursive admissibility architecture derived from the SEXA computationalframework using a controlled mechanical digital twin. Ten state variables are evaluated through six sequential Γadmissibility stages against nominal trajectories and ten controlled fault classes. Parameters were calibrated on separate nominal trajectories and frozen before held-out testing. Thefinal benchmark comprised 1,230,000 mechanical-state evaluations and achieved 97.57% classification accuracy, 95.70% fault sensitivity, 99.59% nominal-state acceptance, and a 4.30% false-admission rate. An independent scalar-threshold baseline produced an 85.55% false-admission rate. The results support recursive admissibility as a reproducible computational engineering constraint architecture under the synthetic conditions tested while preserving explicit boundaries between computational validation and physical hardware or flight validation.
Recursive Admissibility, Mechanical Digital Twin, Fault Detection, Autonomous Aerospace Systems, Hardware Verification.
Hansell Solís Ramírez and Gustavo López, University of Costa Rica, Costa Rica
In programming, large language models that deliver complete solutions endanger students’ reasoning instead of supporting it. This paper reports the human-centered design and integrated evaluation of a Socratic conversational agent for introductory programming at the University of Costa Rica. Twenty students participated by solving a timed Python task and five evaluations were triangulated: perceived usability (SUS), cognitive load (NASA-TLX), interaction logs(411 prompts), an audit of fidelity to the system prompt (374 responses), and a focus group. The agent proved learnable and accepted (SUS = 55.75; moderate load).
Socratic tutoring, Pedagogical scaffolding, Large language models, Usability, Cognitive load.
Olajide Jolugbo, United Kingdom
Teacher professional capability is increasingly exercised under conditions that change faster than conventional professional development cycles. Existing digital and AI competence frameworks specify contemporary expectations, but provide less guidance on whether evidence of previously demonstrated capability remains sufficient when technologies, regulation, assessment and professional contexts change. This conceptual paper develops capability currency, defined as the contemporary sufficiency of a specified professional capability claim in relation to current requirements, available evidence, context and conditions of enactment. Through an integrative conceptual synthesis of teacher competence, adaptive expertise, professional learning, professional digital competence, continuing competence and skill obsolescence, the paper develops the Teacher Capability Operating System (T-COS). T-COS links currency-trigger sensing, materiality triage, review, diagnosis, targeted renewal, demonstration where warranted, proportionate revalidation and evidence updating. Six propositions identify relationships for empirical testing. The model distinguishes changed professional conditions from demonstrated capability gaps and places professional agency, proportionality, privacy and contestability across the lifecycle.
capability currency, teacher professional capability, artificial intelligence, adaptive expertise, professional learning.
Kelechi P. Okpara, University of the Cumberlands, KY
Predictive artificial intelligence (AI) systems are increasingly embedded in tasks once reserved for expert judgment. This raises a question for information systems research: does algorithmic assistance sharpen professional judgment, or erode it? Answering this question with longitudinal registry data is a substantial undertaking. The statistical properties of the required identification strategy are rarely established beforehand. This paper takes that preparatory step. We construct a synthetic panel of 50 experts evaluating 40 cases across four periods, yielding 2,000 observations per replication. The panel embeds a known relationship among algorithmic reliance, decision variation, and diagnostic error. Across 500 Monte Carlo replications, the study tests whether a two-way fixed-effects framework, a placebo design, a U-shape test, and a sensitivity check can recover this relationship. The framework recovers the decision-homogenization effect with negligible bias and nearnominal confidence-interval coverage. The placebo arm returns a null effect in most replications, supporting its use as a falsification check. The framework recovers the U-shaped reliance-error relationship, correctly rejecting monotonicity in favor of a U-shape in 87.6% of replications via the Lind-Mehlum test, and estimates the turning point with a mean of 42.1% but a substantially wider sampling spread than a single analytic confidence interval would suggest. A simplified sensitivity-bounds check appears numerically unstable in many replications, underscoring that this diagnostic should not be interpreted from a single point estimate. We conclude that the proposed pipeline is a statistically defensible candidate for a future study built on real, prospectively collected registry data pairing expert decisions with AI recommendations and adjudicated outcomes, while outlining the data infrastructure such a study would require. All results reported here describe the statistical behavior of an estimator under a known synthetic data-generating process; they are not empirical claims about the behavior of real radiologists or any other professional population.
Algorithmic Augmentation; Cognitive Over-Reliance; Two-Way Fixed Effects; Monte Carlo Simulation; HumanAI Collaboration; Automation Bias; Decision Homogenization; Purposeful Cognitive Friction; Econometric Methods
